The following drivers are critical and complex challenges that underlie artificial intelligence (AI). Drivers impact and alter multiple systems at once. This list was compiled with an emphasis on American society.

Think of these drivers as an inventory of the current system. It does not reflect what the emerging possibilities of the ecosystem might be in a few years. The promises of AI have yet to materialize; the perils are emerging rapidly. We may add drivers as the current conversation evolves. 

Each driver is situated next to/clustered with associated drivers, but they are presented in no particular order. Some drivers directly impact AI while others contribute to the context AI lands in. Not all drivers are equal, e.g. AI safety outweighs the loss of eroticism though both are factors. 

Download as a PDF
American Mythologies
Economic
Work & Jobs
AI Companies & Entrepreneurs
Hardware
Software
Information Ecosystems
Surveillance, Cybersecurity, & Privacy
Art & Media
Social
Tech Marginalization
Young Americans
Critical Systems
The Self
Environmental
Mental Health
Geopolitical
Ontological Challenges
National Politics
Epistemic Challenges

American Mythologies

Definition: The stories and myths that underpin AI in America.

Why it Matters: Narratives are powerful and can shape our society. AI narratives are dominating the public discourse at the moment.

  • America is “an invented country” that is “built around a constellation of ideals—namely, individualism, liberty, equality, hard-work, and the rule of law—that comprise the American Creed.” It is rooted in Puritan-Protestant heritage that valorizes individual merit and “implicitly links work to divine salvation.” Though tensions and disagreements exist within this identity, it is deeply entrenched. American AI embodies machine puritanism. It places an emphasis on the goodness of productivity and output. It has no desires and seeks no pleasure, reflecting the “trainability of self-control.” It displays and helps usher in conformity through homogenizing algorithms and surveillance. AI is the ultimate puritan. See also, Work & Jobs: Productivity.

  • American exceptionalism is the belief that America is “either “distinctive” (meaning merely different), or “unique” (meaning anomalous), or “exemplary” (meaning a model for other nations to follow), or “exempt” from the laws of historical progress (meaning that it is an “exception” to the laws and rules governing the development of other nations).” It sometimes includes features such as the absence of class conflicts or the presence of a predominant middle class. Regardless, America no longer lives up to this narrative of itself. It ranks 31st in education and has the highest income inequality among G7 nations. China’s research and development spending exceeds America’s, with more than a third of top journal articles now written by Chinese researchers. See also, Geopolitics.

  • Prosperity gospel is the belief that God rewards the good with financial gain and health. This belief permeates American society, insulates the rich, and implies that oligarch wealth and power is a divine gift. This deeply held belief may be empowering Silicon Valley and its leaders to pursue unfettered wealth. It also played into the election of Donald Trump, with voters flocking to his promises of prosperity. See also, Economic.

  • This belief asserts that America has a divine right to expand across North America. It is a colonial belief that allows Americans to justify the antagonization of its allies and has been recently propagated by Trump and tech companies through various outlets e.g. the attempt to rename Lake Ontario, ‘Lake America’. See also, Al Companies & Entrepreneurs: Corporate Colonization; Social: English Imperialism.

  • Item desAmerica has a long and fraught history with racism, rooted in slavery and segregation. Racial tensions persist in the US, permeating systems and technology including AI, though awareness has grown in previous years. The unexamined and persistent racial views might be playing a role in the AI race. This is, perhaps, why Americans were so surprised about China’s advancements in AI, particularly the announcement of DeepSeek. See also, Tech Marginalization.

  • This white supremacist myth states that immigrants are conspiring to displace the white population of America. There is overlap between the people touting this myth against immigrants and those who claim AI will displace the human population. In either case, the myth is problematic. The US Census reports that 72% of the population identifies as “white alone”, and that claims of AI labor displacement are questionable. See also National Politics: Immigration; Work & Jobs.

  • More people are longing for a past: a 1950’s America that never existed for most people outside of advertising or to a Gilded Age that ran rampant with wealth inequality. An estimated “62% of Americans reported feeling at least ‘somewhat’ nostalgic for the past.” However, “by politicizing nostalgia, politicians sow seeds of fantasy among their citizens—visions that may never reach the hoped-for fruition.” This promise of simpler times is undercutting our socio-economic and political reality. Americans can’t go back to a past where a single income could easily support a family, and AI will not allow it to do so in any way, shape, or form.

  • AI is the new Wild West, with its frontier models and its ‘gold rush’ metaphors. There are few laws and regulations. It also signifies the new American Dream: a state in which the laws don’t apply to you if you’re rich enough. See also AI Companies & Entrepreneurs: Responsibility Gap.

  • Rugged individualism is the “practice or advocacy of individualism in social and economic relations emphasizing personal liberty and independence, self-reliance, resourcefulness, self-direction of the individual, and free competition in enterprise.” America ranks as the most individualistic society in the world. AI seeks to maximize engagement and intimacy at the expense of real relationships, further weakening our social bonds. Consequently, higher individualism predicts lower intensity of experienced love.

  • Survival of the fittest is often misinterpreted, especially given America’s propensity for individualism and competition. The ‘fittest’ does not mean being the biggest, strongest, or fastest; the theory states that the species that best adapts to its environment is the one that survives. Sometimes, that requires being smaller, softer, and slower. The misreading of this theory underpins tech’s attitude towards AI and its race with China. See also, Geopolitics: Arms Race With China.

  • American society has no shortage of AI depictions in the media. They range from Terminator (destructive) to Spiderman: Brand New Day (helpful assistant) to Wall-E (emotionally intelligent and good). Even without direct mention of it, pop culture is full of references that allude to AI like Obsession (prompt and control, sycophancy) or Backrooms (an analog nightmare) or Pluribus (an always positive hivemind). These ideas have permeated society and influenced our views of AI, but have also been used as training data by AI companies. This is particularly important given that science fiction depictions of AI have influenced military practices. It’s worth noting that concepts of AI or non-human entities differ in other cultures. Unlike their Western counterparts, fictional Japanese “machines are partners to humans,” and in Nigeria, they serve a community rather than serve as a mechanism for personal advancement. See also, Art & Media.

  • DescriSilicon Valley is intent on downplaying human intelligence, often reducing the brain to a computer. Cognitive scientist Romain Brette notes that “the appeal of the brain-computer metaphor is that it promises to bridge physiological and mental domains. But it is misleading because the basis of this promise is that computer terms are themselves imported from the mental domain (calculation, memory, information). In other words, the brain-computer metaphor offers a reductionist view of cognition (all cognition is calculation) rather than a naturalistic theory of cognition, hidden behind a metaphoric blanket.” See also, Epistemic Challenges.

  • Silicon Valley has enjoyed an almost mythical status in the world. This admiration was, in part, justified because Silicon Valley attracted the best talent from around the world for a number of years. This is no longer the case. Far from “making the world a better place,” Silicon Valley has spent the past 20 years making “assisted living apps for Millennials.” Now, we are more aware of the harms that Silicon Valley has unleashed on the world, and that knowledge of harms (particularly, social media harms) has made us wary of AI. A slew of new films and documentaries are seeking to expose Silicon Valley leaders. See also, Al Companies & Entrepreneurs. 

  • The idea that tech will save us is a “utopian myth [that] provides hope and social dreaming of a radically different future society, while simultaneously naturalising contemporary social and political order.” The ultimate challenge with this myth is that it helps “to justify any means necessary to realise the utopian dream.” AI leaders are dangling the idea that AI will save the world. Evidence does not support that.

  • Tech leaders have made an abundance of promises about AI and its ability to solve everything from cancer to climate change. These promises have yet to materialize in any meaningful way. AI can and likely will serve as an effective and versatile tool that scientists and researchers can use to progress knowledge. However, it is an aid, not the be all and end all solution.

  • AI doom has become a hot topic in recent weeks. A lot has been said about the revelations from industry-insiders and whether or not AI poses an existential risk. The doomer narratives are questionable claims that are distracting from real world harms caused by AI, all while distorting the public conversation about its capabilities. Genuine and alarming risks of AI are going underreported. For instance, CNN reported that the US almost went to war with China because an AI hallucinated reports that a Chinese ship carried nuclear components to Iran.

Critical Systems

Definition: The foundational systems of the planet, in human society, and how those interact with AI. This includes AI safety as a systemic concern. This section is deeply linked to other drivers including Environmental, National Politics, and Economic which required further analysis and deconstruction.

Why it Matters: AI is increasingly being embedded into the complex systems that humans depend on.. How these systems impact AI, and how AI impacts those systems, will play major roles in shaping the promise and peril of AI.

  • The recent emphasis on deregulation, tariff policies, and war footing have left US systems fragile and more susceptible to AI risks. Critical systems are often made more resilient through redundancy, the ability to continue functioning when one component fails. Systems optimized for efficiency remove these so-called redundancies that ensure buffers for disruptions in service of reducing time and cost. AI’s potential to improve efficiency makes it appealing to business owners and managers seeking to reduce costs. But resilience will also mean building in non-AI workflows that can serve as fallbacks in the event of errors or unexpected behaviors, the impacts of which can compound and cascade if left unattended. Implementing such processes may become more difficult if human expertise atrophies via AI over-reliance and cognitive offloading/surrender.

  • Critical systems are complex. Interventions, no matter how well-intended or designed, sometimes produce unintended consequences. AI is simultaneously capable of generating more complexity and of simplifying complex challenges by failing to account for important context and interdependencies.

  • AI safety cuts across every system and all uses of AI. Currently, “the dominant safety framework for [AI] assumes that humans, properly positioned in the decision chain, will catch what AI gets wrong.” AI safety advocates argue that this is insufficient to ensure pro-human futures. AI alignment refers to making sure “an AI system’s goals and behavior match what people actually want … getting the AI to do the ‘right thing’ even in new situations, not just follow instructions literally in ways that cause harm.” There are ongoing debates about the best approach to AI and the risks that it presents, with different factions and motivations emerging. AI safety is a global issue that warrants global cooperation.

  • AI is dependent on critical systems, including power, water, telecommunications, global supply chains, land management, human labor, and more. Society is increasingly betting on AI by embedding it in critical systems while AI simultaneously competes with those systems for scarce resources and infrastructure. We’re already witnessing how this becomes a governance issue; Texas has halted new state-issued data-center permits pending an audit of grid, water, and cost impacts and California has adopted new requirements around data-center energy and water use.

  • AI can introduce new chokepoints and points of failure when individuals and entities rely on the same software and hardware providers, AI models, datasets, and cybersecurity infrastructure. In this context, a single cyberattack, outage, or flawed update can impact organizations across a range of industries, such as what occurred during the global CrowdStrike and Microsoft IT outage in 2024. These risks are amplified by the interdependence of critical systems. For example, disruptions in power, communications, or cloud infrastructure can severely disrupt finance, medicine, and emergency services. As AI becomes more pervasive and embedded, it deepens these interdependencies, which means individual problems can cascade more rapidly, widely, and severely. 

  • AI can increase velocity, especially when agents are granted the ability to act with any degree of autonomy. This can enable both accelerating coordination and failure cascades. The notion of human-in-the-loop systems seeks to respond to these potential risks by ensuring that AI speed is checked by human oversight. Critical domains like cybersecurity, financial markets, war, power, logistics, and more will need to thoroughly evaluate when and how AI is used, and what relationship human overseers have to it, in order to mitigate potential risks and harm.

  • AI could accelerate scientific progress by helping researchers design experiments, review existing literature, analyze data, and generate hypotheses. AI not only allows scientists and researchers to do their work faster, but fundamentally changes the questions that science can ask. DeepMind’s GNoME, for example, identified millions of potentially stable crystal structures, while more recently Anthropic established a molecular biology lab, through which they claim to have “discovered a novel enzyme system with properties reminiscent of CRISPR.” Ongoing experiments in protein folding have yielded new databases that collectively predict the structures of more than a billion proteins, including viruses. Rigorous scientific claims will need to survive validation, but these developments indicate that AI could be used to rapidly advance fundamental research and experimental breakthroughs. 

  • AI has the potential to usher in a golden age of healthcare, albeit with risks and flaws. Disease management, diagnostics, drug design, and logistics management are among the public health domains that may be transformed by applied AI. Conversely, it may decrease trust in medicine/doctors, produce clinical hallucinations, and introduce context blindness, among other challenges, and may prove harmful especially when it is used by insurance providers to deny claims with the total absence of compassion. It may do all this while simultaneously harboring and reducing biases. See also, Mental Health.

  • The US military is implementing an ‘AI-first’ approach to warfare. Applications range from surveillance to generating intelligence reports to directing drone attacks. AI lacks human capabilities and “the models may not understand ‘stakes’ as humans perceive them.” In high-pressure scenarios, military planners may face stronger incentives to rely on AI, and their decisions/perceptions might be shaped by AI’s understanding of risk and timelines, not ours. The Cold War didn't escalate into nuclear armageddon, for example, because Stanislav Petrov and Vasili Arkhipov questioned information and defied orders. AI can make efficient, probable, and deadly decisions based on plentiful data, but none of those are necessarily good decisions. See also, Geopolitics.

  • AI is transforming education from every angle. Approximately “30% of K-12 students use AI tools at least once per day.” On one hand, AI presents opportunities such as personalized learning and immersive experiences. On the other, it can be dehumanizing, biased, and a privacy risk for already vulnerable populations. Screens and social media have been documented to have a harmful effect on children. The research on cognitive offloading/surrender and what that might mean for young people cannot be ignored. It is necessary to be vigilant about how AI might compound existing harms while introducing new ones. See also, The Self; Young Americans. 

  • Approximately 318 million people face acute hunger globally in 2026, more than twice the number recorded before the COVID  pandemic. AI capabilities might help alleviate the challenges we face in food and agriculture such as inefficiencies (we produce $1 trillion in food waste globally) and minimizing impacts of climate disasters. Nutrition, crop yield, and supply and demand management are other areas that may stand to benefit. However, AI is still prone to bad advice, hallucinations, and public distrust. Offloading food and agriculture decision-making to AI could prove disastrous in the future particularly as environmental disasters and changing conditions interrupt our use of technology. See also, Software.

  • Generative AI is “used in fraud detection, credit decisions, risk management, customer service, compliance, and portfolio management, improving accuracy and efficiency. AI is also being adopted in asset management and securities, including portfolio management, trading, and risk analysis.” However, “the use of AI in finance creates potential risks for institutions, including biased or flawed AI model results, data breaches, cyber-attacks and fraud, which can cause financial losses and reputational damages eroding consumer trust.” See also, Economic; Surveillance, Cybersecurity, & Privacy.

Environmental

Definition: The physical world’s resources, ecosystems, and climate conditions, which impact and are impacted by human activity.

Why it Matters: AI, like “the cloud” before it, is often framed as immaterial, but it is a highly physical system. Broadly understood, the environment is what makes AI possible and something that AI in turn affects. This includes electricity, water, land, minerals, manufacturing, waste, greenhouse gas (GHG) emissions, pollution, and extreme weather events caused by anthropogenic climate change, all of which are exacerbated by AI diffusion, and at the same time AI is increasingly used to observe and manage these factors.

  • The AI industry consumes an enormous amount of water. The most visible use is for cooling: data centers generate heat that must be shunted, and many rely partly on evaporative cooling. But the larger water footprint extends to the power sources that supply its electricity and semiconductor manufacturing. A 2025 Nature Sustainability analysis estimates that expansion of AI servers in the United States could create an annual water footprint of roughly 731 million to 1.125 billion cubic meters between 2024 and 2030; in its base case, 71% of the footprint was indirect rather than water used inside the data center itself. Earlier research estimated that training GPT-3 in Microsoft’s US facilities directly consumed about 700,000 liters of freshwater. The other important variables are when and where AI uses water. A gallon consumed in a water-rich region during a cool month is different from a gallon evaporated in a drought-stricken region during the summer.

  • Data centers and all their supporting systems require large quantities of resources, including ‘rare earth’ elements, copper, aluminium, silicon, and others. For magnet rare earths alone, China accounted for roughly 60% of global mining in 2024, 91% of refining, and 94% of sintered permanent-magnet production. The IEA estimates that data-center demand for gallium could reach more than 10% of today’s global supply by 2030 (China presently also dominates its refining). Short of miracles in recycling, efficiency, or substitution, scaling AI means scaling extraction, refining, and manufacturing. A shortage or environmental restriction involving even a single material could create strain or disruptions, with ensuing economic and geopolitical implications. See also, Geopolitics.

  • Data centers consumed about 415 TWh of electricity worldwide in 2024 (roughly 1.5% of the global total), and the IEA expects consumption to increase to 945 TWh by 2030, with AI a major source of the increase. In the US, data centers are projected to account for nearly half of electricity-demand growth through 2030. Large loads can require new substations, transmission lines, transformers, and more, all on long timelines. Who pays for all that infrastructure has become a driving force in the data center backlash. In September 2026, the House passed the Ratepayer Protection Act, which would “protect consumers from rate increases resulting from data center construction by having state utility commissions consider large-load standards for data centers drawing more than 100 MW of power, which would ensure they pay for the full incremental costs to serve their loads.” An effort to move the measure through the Senate was blocked. An “unjust” transition may occur in which a minority reaps the benefit from data center buildouts, while the majority shoulders the cost, land, and materials burden.

  • More than two-thirds of the world’s population lives in developing economies outside China, but these countries account for less than one-third of global electricity generation and less than 10% of data-center capacity. Only ~60% of their population has access to reliable internet. Meanwhile, 655 million people worldwide still lack access to electricity altogether, including more than 560 million in sub-Saharan Africa. “Energy apartheid” describes a future in which this pattern persists and intensifies; premium energy systems emerge for computation while basic energy access remains unreliable for large populations.

  • The data center is only one part of AI’s global physical footprint. Some of the industry’s largest construction projects cover thousands of acres, but their effective footprint also includes power plants, substations, warehouses, roads, minerals, and more. Which specific land is used also matters; building on developed industrial land is a different proposition than clearing forests, wetlands, farmland, or carbon-rich soils. Peatlands, for example, only make up ~3-4% of global land but contain as much as one-third of the world’s soil carbon; degraded peatlands already contribute roughly 4% of annual human-caused greenhouse-gas emissions. Additionally, many tech and finance analysts warn that serious obstacles stand in the way of some data center build-outs. They point out that announced projects and grid-connection requests exceed the capacity likely to be completed on corresponding timelines, alongside other vulnerabilities like permitting delays, financing constraints, and social backlash. Local opposition alone is believed to have “blocked or delayed $68 billion in data center projects” between April and June of 2026 alone. Delays or outright cancellations have the potential to create significant impacts on local and regional land use. See also, Al Companies & Entrepreneurs: Corporate Colonization.

  • AI infrastructure is being built into a world that is already warming. 2024 was the warmest year ever recorded, at ~1.55°C above preindustrial levels, and the WMO expects the ‘Super El Niño’ established in 2026 to become a “very strong” event. Experts believe 2027 will become the new warmest year on record, in part due to the El Niño. This creates a multifaceted problem for AI. Hotter air increases cooling requirements exactly when homes and businesses also need more energy for air conditioning. Drought can constrain hydropower, as is expected to happen at the Hoover Dam this year. Floods, fires, and storms threaten transmission and telecommunications infrastructure. The result is a vicious cycle: AI creates increased energy demand in a world already heating from GHG emissions, which in turn creates more heat and emissions, which in turn creates more demand for energy to cool. At a local level, the phenomenon of urban heat islands means that metropolitan areas can get as much as 20°F warmer than nearby rural areas due to buildings and pavement absorbing and releasing heat, and research has demonstrated a “correlation between lower-income neighborhoods and higher temperatures.” Data centers are being built out within this context, and they themselves have been shown to create their own heat islands, “warming the land around them by up to 16 degrees Fahrenheit, and making life hotter for more than 340 million people.”

  • AI produces pollution throughout its lifecycle. Semiconductor manufacturing involves hazardous chemicals and uses fluorinated greenhouse gases and nitrous oxide. The EPA notes that, depending on the process and abatement technology, 10-80% of fluorinated gases entering semiconductor manufacturing tools could pass through unreacted and be released. Mining and refining add waste, emissions, and habitat disturbance. Data centers can add localized emissions when backup generators or on-site power systems operate. Emissions and water use are the stories that dominate AI backlash, but a system can become less carbon-intensive and still increase pollution (and the associated environmental costs).

  • The world generated 62 million metric tons of electronic waste in 2022, but only 22.3% was documented as collected and recycled. The UN projects global e-waste reaching 82 million Mt in 2030 under business-as-usual conditions, while the formal recycling rate falls to around 20%. AI could worsen this by accelerating hardware turnover, especially as the race for high-end capabilities incentivizes installing the latest hardware before existing parts reach the end of their life cycle. There is not currently consensus on how much e-waste is attributable to AI, but if current patterns hold, it could exacerbate the existing cycle. It is possible the AI industry could develop secondary markets for older parts and recycling processes for minerals and materials, depending on economic viability and/or political will.

  • The environmental cost of AI exists across an industrial network. Emissions from fabricating chips, manufacturing servers, producing steel and concrete, constructing buildings and grids, transporting equipment, and the byproducts of waste all contribute to it. But the full energy and emissions picture is often obfuscated because of stakeholder incentives to withhold or selectively frame the data, as well as the overwhelming complexity involved in trying to independently research the industry. Few lifecycle assessments have therefore been conducted. Carbon accounting may become important in making regulatory decisions and establishing policy precedents. With regard to emissions, GHG Protocol permits both a location-based method, reflecting the average emissions of the grid where electricity is consumed, and a market-based method based on contractual instruments such as purchased electricity products. 

  • AI is simultaneously pushing the energy transition and increased reliance on fossil fuels. Hyperscalers are becoming major buyers and financiers of renewable energy and other low-carbon technologies. The IEA expects renewables to meet nearly half of additional global data-center electricity demand through 2030. On the other hand, demand is arriving faster than many grids can build new low-carbon generation and transmission. In the IEA’s base case, natural gas and coal together meet more than 40% of additional data-center electricity demand through 2030; in its higher-growth “Lift-Off” case, nearly half of additional data-center generation through 2030 comes from fossil fuels.

  • In addition to impacting the environment, AI is often touted as a technology that can improve the efficiency of existing infrastructure and generate novel climate “solutions.” The IEA estimates that AI-assisted fault detection could reduce outage durations by 30-50%, while remote sensors and AI-based management could unlock 175 GW of additional capacity without building new transmission lines. Similar systems are being used to do things like identify methane leaks, optimize industrial processes, control heating and cooling in buildings, monitor ecosystems, and forecast extreme weather. In its “Widespread Adoption” scenario, the IEA estimates existing AI applications could enable 1.4 billion Mt of CO2 reductions in 2035, but also warns that adoption barriers and rebound effects could interfere with them.

  • AI is being used in efforts to decode animal and plant communication. Some researchers are optimistic that such efforts will “bolster animal rights,” while others express concern about how such knowledge might be used. Either way, this introduces a novel possibility: how will humanity respond to what plants and animals communicate to us? And what rights, agency, and powers will they be granted? As these initiatives progress from sci-fi dreams to reality, it will be critical to simultaneously develop regulations and norms that sufficiently protect all parties.

  • The AI industry often downplays its energy toll by focusing on narrow evaluations like, “How much energy does one ChatGPT prompt use?” By comparison, this would be like asking how much one Google search used in 2005, without evaluating the lifecycle of processes and parts involved in building the infrastructure for Google to operate in the first place. The more useful unit of measure is the network. AI consists of chip fabs, servers, cooling systems, energy infrastructure, warehouses, mines, and billions of interactions that cause other machines and people to do things. The IEA reports that hardware and software improvements have reduced the energy required for a simple AI task by at least an order of magnitude annually in recent years, but at the same time, AI-focused data-center electricity consumption rose 50% in 2025. Emerging AI capabilities like reasoning, video generation, and agentic tasks can require hundreds of times the energy of basic text generation.

  • The most significant environmental consequences of AI might emerge from the cascades they induce. A heatwave raises electricity demand for air conditioning while simultaneously increasing cooling needs for data centers, and drought can reduce both hydropower and available cooling water at the same moment. Building more generation requires land, construction, and resources; even just mining those metals requires energy and water. These are second- and third-order effects, and they are difficult to capture without comprehensive lifecycle assessments. The IPCC already describes climate risk in these terms: energy, water, transportation, telecommunications, and other critical infrastructures are interdependent, capable of causing hazards to compound and self-reinforce.

  • Energy systems change more slowly than demand for new technologies. Many long-range scenarios created before the AI boom did not include a demand shock of its present form. Lawrence Berkeley National Laboratory’s June 2026 update now estimates that data centers could consume 11.8% of US electricity in 2030, with a scenario range of 9.5-15.3%; its previous report had estimated 6.7-12% by 2028. The IEA reported that electricity consumption by data centers jumped 50% in 2025 alone. Meanwhile, climate models have historically underestimated some consequences of global warming. Uncertainties like the AMOC collapse, among other possible shocks and punctuated equilibria that could produce rapid change, must be considered in future models that seek to assess impacts of AI diffusion.

Geopolitics

Definition: Politics, power, and relationships between countries.

Why it Matters: America’s role in the world is shifting. It is important to note that the countries/regions listed in this section are those playing an active role in America’s future, particularly from an AI supply chain standpoint.

  • The US and China are both seeking to win the AI race, but differ in how they are approaching the challenge. While US companies pursue Artificial General Intelligence (intelligence that meets or exceeds human capabilities, or God in a machine), China is taking an application-based approach, seeking all possible opportunities to boost its economy through AI. It is focused more on open models, software innovation, and robotics. Additionally, China is introducing safeguards and regulations for its economy and people. For instance, Chinese courts have ruled that companies cannot replace workers with AI. In contrast, Trump has dismissed calls for regulation even when AI tech leaders have called for it. Though sanctions have hampered China, its ability to innovate should not be underestimated (it should not have surprised anyone that engineering and manufacturing-driven China produced DeepSeek cheaper and faster than American counterparts).   


    As stated under National Politics: Polarization, the society in which AI lands matters as much as the technology and its capability. A shifting world order will likely create new opportunities. China is poised to become the world’s next superpower. On their current trajectories, China is better positioned to win the AI race in the long run, even if America achieves AGI. See also National Politics: Polarization; Hardware; Software.

  • The American hegemony ended in 2025 with the second Trump administration. The coercion of allies, illegal tariffs, and the dismantling of USAID have all diminished America’s soft power on the global stage (gaps filled by China, allowing it to rise faster). When Canadian Prime Minister, Mark Carney, spoke of middle powers and the shift in world order at Davos in January of 2026, he became leader of the free world, a position Trump willingly abdicated.  

  • China’s Belt & Road Initiative (BRI) is “a comprehensive plan to build land and maritime infrastructure projects that will put China at the centre of international trade.” China has increasingly sought soft power through avenues such as entertainment, infrastructure, and diplomacy. AI will play a role in this initiative as China continues to invest in projects throughout the world. This includes materials, energy, and infrastructure. 

  • Antagonizing allies and enforcing tariffs also isolates the US on the global stage. This hostile stance combined with the growing involvement of tech leaders in policy is causing the world to reconsider its reliance on US tech and infrastructure. Nations are actively moving away from American tech companies and their products. China is switching to Linux from Windows. Spain has blacklisted Palantir. France “recently announced that 2.5 million of its civil servants will gradually discontinue the use of US-based videoconferencing platforms Zoom, Microsoft Teams and WebEx by 2027.” While it might be challenging for the world to untangle from US tech completely, escalating conflict will likely accelerate this change. Antagonizing allies might also encourage the world to adopt Chinese AI models over American AI models in the long run.

  • A Pew Research Center survey found often overwhelmingly negative views of Trump in regions around the globe. The survey noted a steep decline in the number of people who considered the US a reliable partner, including in nations where the US has longstanding economic and security ties. America’s global reputation is in shambles. 

  • After the US instigated a war with Iran, Iran proposed tolling the Strait of Hormuz in yuan, not the dollar. The current administration’s approach to trade and tariffs may push the world towards de-dollarization. If this happens, it will represent a significant and permanent change to the world order. 

  • Restrictions to H-1B visas, federal cuts to funding, and deportations are fueling a self-inflicted Brain Drain in America. An analysis from Science revealed that America has lost more than 10,000 STEM PhDs since Trump took office and that departures at 14 key research agencies outpaced hires by a ratio of 11 to 1. A study by the Information Technology and Innovation Foundation warned that the cuts to science could shrink the US economy by nearly $1 trillion over 10 years. The EU, Canada, China, and other nations are benefitting from targeted efforts to recruit fleeing US talent. See also, National Politics: Immigration.  

  • According to the OECD, the training and deployment of an AI system relies on a highly complex and capital-intensive global supply chain. Kinaxis reports that “a single high-performance AI chip could travel over 25,000 miles from initial raw material extraction to the final product delivery, requiring meticulous coordination across multiple countries and industries.” The OECD warns that “the scale of infrastructure construction underway suggests that risks such as bid rigging may become more salient, and as technologies mature and commoditise, the potential for price co‑ordination or cartel behavior may also increase.” It is important to emphasize that this supply chain is global. US isolationist policy will not serve its interests in AI dominance. See also, National Politics: Immigration, Economic: Supply Chain Management, Environmental: Critical Minerals & Resources; Hardware: Hardware Supply Chains.

  • AI models are propped up by the “labor of workers in the Global South, who are exposed to disturbing content and poor working conditions,” across the AI supply chain from mineral extraction to data annotation. AI also involves thousands of human-performed microtasks that include “content moderation, image tagging, transcribing audio, and flagging hate speech,” often performed in the Global South. See also, Work & Jobs. 

  • World War 3 seems closer than ever. AI will play a significant role in warfare, though not necessarily for the better. Current war zones including Ukraine and Palestine have become testing grounds for new technology, and claims that AI will decrease civilian death and harm are unsubstantiated. It is critical to note that use of AI for military purposes comes with significant, disproportionate risk. CNN reported that the US came dangerously close to starting a war with China when AI falsely claimed that a Chinese ship was carrying nuclear components to Iran. See also, Critical Systems: Defense.

  • Starting a war with Iran was a misstep in America’s AI endeavors, especially since the US is losing this war. According to the World Economic Forum, “the current conflict is disrupting not only oil, but natural gas, shipping, insurance, industrial gases and the security of data centres. The future of AI will depend on the countries that can still deliver reliable electrons, materials inputs and secure infrastructure when world trade is under geopolitical strain.” The war cut off helium supplies from Qatar, which in turn threatens chip manufacturing. Data centers and tech companies have also become military targets given their ties to the US government, with Iran attacking Amazon’s data centers.

  • Taiwan Semiconductor Manufacturing Company produces 90% of the world’s advanced chips. Taiwan’s sovereignty has been threatened by China. Given recent events, Taiwan may not be able to rely on US protection and allyship as it has with previous administrations. If China annexes Taiwan, it will control a significant portion of the global AI supply chain. See also Hardware: Semiconductors. 

  • Ukraine produces approximately 50% of the world’s neon required to produce chips. Ukraine’s critical raw minerals are among the reasons why Russia invaded the country. Russia has suffered significant losses as of late, having miscalculated its military might and Ukraine’s resolve. It seems Ukraine is poised to win. A Russian victory was never in America’s best interests.

  • Mineral riches hiding under Greenland’s ice are among the reasons why Trump is demanding control of Greenland from Denmark. A threat against Greenland is a threat against Europe and NATO. Acting on this threat will effectively end American allyship with Europe. 

  • The unique relationship between the US and Canada has been the envy of the world for decades. Both countries have benefited from longstanding military cooperation, trade partnerships, and relatively open borders. Canadian natural resources have fueled America’s economic engine. The world is watching Canada’s response to American threats, and learning from this example. Antagonizing Canada will continue to change the world’s relationship with America and its tech companies.

  • A significant disparity exists with who adds value and who reaps it: “while Africa holds 30% of the world’s critical mineral reserves, essential for electronics and AI hardware, it captures only 10% of the global revenue generated from these resources.”  Critical minerals play a significant role in this. For instance, the “Democratic Republic of Congo accounts for over 70% of global cobalt output and approximately half the world’s proven reserves. South Africa, Gabon and Ghana collectively account for over 60% of global manganese production. Zimbabwe, alongside the Democratic Republic of Congo and Mali, hold substantial but yet-to-be-explored lithium deposits. Other countries with significant critical mineral reserves include Guinea, Mozambique, South Africa, and Zambia.” China has prioritized AI expansion on the continent.

  • Calls for international AI standards have grown over the past few years with international organizations like OECD and International Organization for Standardization. Though no global standards have been established as of yet, AI companies have discussed creating their own standards body.

  • A new global stratification is emerging between countries that have developed AI infrastructure and models and those that haven’t. Countries will fall along a spectrum in which some lead, more follow, and some struggle to participate. Factors will include development of models, data centers, the development of AI strategies, access to raw materials, trade capabilities, among others. It has been argued that “non-English languages lack the needed quantity and quality of data to build and train effective models,” amounting to systemic exclusion, However, “the dataset fallacy” ignores “factors like decades of underinvestment in digital infrastructures” and “misaligned research interests” that privilege  English and English-speaking countries.

National Politics

Definition: Policy and governance at the national level.

Why it Matters: The governance of AI, decisions of politicians, and policies developed in relation to AI will determine future impacts and outcomes of this technology. This is particularly important now as AI becomes a political battleground.

  • “Multiple indicators show a decline in the health of America’s democracy,” and most Americans agree that democracy is in crisis. A tech oligarchy has established itself: Peter Thiel installed J.D. Vance as Vice-President. Thiel’s ‘in-house political philosopher’ (a prophet among Silicon Valley elites) has advocated for a Dark Enlightenment: a vision of the world led by CEO-monarchs. See also Economic: Oligarchy & Wealth Concentration.

  • AI plays a critical role in this driver because “modern national security is to receive as much information as possible and process it.” AI capabilities lend itself both to national defence and foreign aggression/dominance. This includes the ability to infiltrate and take command of other nations’ tech ecosystems. Also see Surveillance, Cybersecurity, & Privacy. 

  • Tech leaders have courted Trump, and Trump has embraced AI. OpenAI has considered offering the Trump administration a 5% stake in the company. Big tech donated over $300 million to his campaign in 2024. Tech leaders occupied prominent seats at his inauguration. Since then, this administration has rolled back regulation, intervened in policy, and redirected capital on behalf of AI companies despite bipartisan backlash against AI and calls for AI safety regulations. 

  • AI is a complex technology, not a series of tubes. Axios reported that “members of Congress and governors are under increasing pressure to regulate AI, but more than two dozen of them who spoke with Axios said they don't use the technology or have rarely done so.” The median age for house representatives is 57.9 and senators is 64.7.

  • Former FTC Commissioner, Lina Khan, stated that there can be a “rush to create exceptionalism” around new technology, and that the US already has a number of regulations and laws that can be applied to AI. This includes “consumer protection laws, anti-discrimination laws, anti-trust laws, [and] product liability regimes,” among others.  Though some regulation may need to be created or adapted for AI, we do not need to start from scratch. It is worth noting that tech companies calling for regulation are perfectly capable of slowing down or stopping their pursuit of advanced AI. See also, Critical Systems: AI Safety.

  • Tension is evolving between the federal government and states on AI policy and regulation. Various states are passing legislation on chatbots, use of AI in healthcare, privacy, among other concerns while the federal government focuses on deregulation. AI may become a carrot and stick issue, with the Trump administration enforcing its agenda and forcing states to comply. 

  • AI presents a threat to civil liberties given its abilities in profiling and surveillance, as well as mundane applications such as screening and decision-making in recruitment, housing, medical records, etc. AI tech is also built with substantial biases baked into them, particularly against vulnerable and marginalized groups. First Amendment rights may no longer be a given when it comes to this technology; a number of arrests have been made related to AI protest and demonstrations. See also, Tech Marginalization.

  • AI will play an increasingly important role in upcoming elections. Robocalls, swarm bots, deepfakes, and disinformation campaigns are among the ways in which AI might be used to manipulate the public and election results by domestic and foreign actors. Lawmakers are doing little to prevent rampant AI use and are struggling to keep up with the speed of innovation. Furthermore, AI has become an election issue. Candidates who fail to oppose data centers will find their campaigns in trouble. See also, Information Ecosystems.

  • The nuclear bomb was built by refugees. Dozens of nuclear scientists fleeing fascism in Europe during the Second World War contributed to the success of the Manhattan Project, including Oppenheimer, Einstein, and von Neumann. Embracing immigrants directly contributed to American hegemony. Immigrants (14% of the US population) are “responsible for 36% of aggregate innovation and are 80% more likely than native-born Americans to start a business. Nearly half of all Fortune 500 companies were founded by immigrants or their children, and 59% of AI PhD graduates working in US industry are international students.” Roughly 66% of tech workers in Silicon Valley are foreign-born. The current administration's hostile stance on immigrants and their home countries will impact tech and progress in America for decades to come. Top talent will likely seek out other markets to avoid ICE and persecution within the United States. See also, Geopolitical: US Brain Drain; American Mythologies.

  • Though AI has become a bipartisan issue (for now), the debate surrounding it arises in a polarized society. The societal conditions in which AI lands matter as much as the technology itself, if not more so. Still, a recent poll showed that Americans believe they have more in common with each other than what divides them. Anti-AI sentiments may help create national cohesion.

  • In recent months, the public has turned against AI. Pew Research Center reports that “about 50% to 71% of Americans express wariness, concern, or direct opposition regarding different aspects of artificial intelligence, with international surveys noting that Americans are among the most anti-AI in the world.” Job displacement, the destructive impact of data centers on communities, and growing awareness of surveillance tech are among the reasons why this sentiment is rising. US officials have received death threats for pro-AI and data center stances. Protests are taking place across the country. See also, Hardware: Data Centers; AI Companies & Entrepreneurs: Tech Entrepreneur Backlash, Youth: Gen Z Rejects AI. 

  • Intellectual property of including various works of art, scholarship, social media content, etc. have been used to train AI models, mostly without any compensation or acknowledgment. Under US copyright law, work generated by human authors using AI as a tool might be protected, but purely AI-generated content cannot be copyrighted. Intellectual Property will be an ongoing issue over the coming years, with precedents set in courts. Expect more lawsuits and disputes. See also Software: Trusted Data; Art & Media. 

  • A recent poll from The Associated Press found that “Americans are more skeptical of government information than they were a few years ago, with majorities now saying they have little or no trust in federal government information about elections and politics, foreign affairs or the environment.” This includes an erosion of trust among Republicans who voted for Trump. Use of AI and AI memes will only destroy what little trust remains. See also, Ontological Challenges: Truth.

  • In the face of deregulation and legal onslaught, some institutions are holding strong. Judicial systems have pushed back against AI companies and the Trump administration despite ongoing threats. States are asserting their rights on AI and demanding action to regulate the technology. Voters are making their dissent against AI known. Not all is lost.

Economic

Definition: The systems, structures, and conditions of resources related to AI in America.

Why it Matters: Our economic reality deeply impacts our experience of life and the system around us. The economics of AI have significant impacts on society.

  • Economic indicators that have historically helped measure the health of an economy do not correlate with the everyday experiences of average Americans—a reality that the American left failed to take into account in the last federal election. This includes indicators such as GDP and the stock market. While official metrics may not indicate a recession, Americans are deeply strained. Economic indicators also do not take trade-offs and compromises into account. An individual that appears economically stable on paper might be making numerous concessions on where to live, what to eat, how often to go out, what goods to buy, how many children to have, etc. These decisions add up and affect quality of life and how people cope with their reality.

  • A recent poll found 95% of Americans surveyed believe the US is suffering an affordability crisis. The New York Times reported that “consumer prices overall have risen more than 30% since the beginning of 2019. That’s two and a half times as much as they went up from 2012 to 2019.” Everything from gas to food to housing is more expensive. Affordability might factor into why people turn to AI. Having an AI boyfriend is cheaper than going on a date. Using it for therapy is cheaper than talking to a therapist. Consulting a chatbot for a medical ailment might be cheaper than going to a doctor.

  • According to the US Bureau of Labor Statistics, the annualized inflation rate was 3.4% as of August 2026. Though inflation has eased, factors such as the war with Iran, the rising costs of services (e.g. child care, healthcare, etc.), and the trade war with Canada are likely to offset the downward trend. Additionally, wages have not kept up with inflation further straining Americans. See also Work & Jobs: Wages.

  • The US is waging a trade war against Canada which fuels (steel, wood, oil, etc.) its economic engine. Trump has also placed tariffs on 60 economies, adding to economic pressure within the US. Though the Supreme Court struck down tariffs, the resulting refunds were distributed to corporations that had already passed the costs onto consumers. The tariffs effectively removed wealth from the American people and redistributed that wealth to corporations. It is important to note that the longer these trade and tariff disputes continue, the longer the world has to reorganize itself and move away from US hegemony. As it stands, Canada is poised to become the first Associate Member of the EU. See also Geopolitics.

  • In the first quarter of 2026, the top 1% held over 30% of the economic wealth in the country. That number jumps to approximately 68% when comparing the top 10% of wealth holders to the bottom 90%. Economist Paul Krugman found that today’s wealthy held a significantly larger share of both national wealth and economic output than they did during the Gilded Age, an era known for extreme wealth concentration. Furthermore, the wealthy are using their fortunes to influence policy and political outcomes. The Washington Post reported that “roughly 1 in every 13 dollars spent in [2024’s] national elections was donated by a handful of the country’s richest people….Since 2000, political giving by the wealthiest 100 Americans to federal elections has gone up almost 140 times, well outpacing the growing costs of campaigns.” See also National Politics: Government for Sale; AI Companies & Entrepreneurs.

  • Much has been said about the ‘Great Millennial Wealth Transfer’. These headlines are overly optimistic and not consistent with countering systemic data. The headlines are failing to consider several key factors: the pandemic-driven concentration of wealth (i.e. fewer Millennials inheriting more), increased costs of living, and the longer life span of Boomers resulting in greater living and healthcare costs, including unaccounted-for costs (e.g. if you live 20 years longer, you require one more roof replacement, a new furnace, etc.). These factors will erode inheritances. Additionally, nearly 70% of Millennials are concerned that their parents may not have enough money to comfortably retire. An already economically strained generation may shoulder the costs of end-of-life care rather than inherit wealth.

  • A growing class consciousness is driving dissatisfaction, though more people have pro-labor than pro-capital class consciousness in the US. The pandemic exposed worker precarity and inequalities. Gen Z, in particular, became disillusioned with the current state of capitalism. AI and tech leaders are a focal point of this movement. See also, National Politics: AI Backlash.

  • A study confirms that Medicare for All would save 114,000 lives and more than $1 trillion a year. Social housing (e.g. Finland’s Housing First program) is cheaper than allowing people to be unhoused because of the costs they incur in policing, healthcare, and social services. It’s estimated that England could save an estimated £2.1 billion every year in treatment costs if everyone had access to good quality green spaces. The initiative can also cool streets and reduce carbon emissions, benefits noted by New York mayor, Zohran Mamdani, in his new initiative. What the US calls socialism is often just good math and better accounting. However, 93% of US adults indicate that they experience some level of math anxiety.

  • Significant amounts of capital have been redirected by the government and investors towards AI, at the expense of other societal needs and critical infrastructure. The White House has redirected $200 billion in funding from colleges to individual AI researchers. Another $5 billion will be redirected to medical, construction, etc. innovation based in AI. The American Society of Civil Engineers notes that the US faces a $3.7 trillion investment gap over the next decade just to bring its overall infrastructure up to a state of good repair (this includes roads, bridges, and hazardous waste). If AI does not live up to its promises, this redirection away from non-AI infrastructure, innovation, and research could be catastrophic.

  • AI betting and prediction markets are gaining traction. Combined monthly global trading volume of Polymarket and Kalshi rose from less than $5 billion in September 2025 to about $24 billion in April 2026. Sports, cryptocurrency, and politics are the top three topics. Prediction markets will be a contentious issue going forward, with some States passing legislation to ban them and 20 federal lawsuits filed against these companies. Though they claim to be financial exchange platforms, detractors say they are effectively gambling sites that should be regulated.

  • Chatbots aim to exploit our need for connection and intimacy. The number one use of AI is role-playing. The AI companion market is projected to be worth $31.1bn by 2032. China has introduced a policy to ban AI relationships, but no such legislation exists in the US.

  • Data is a valuable asset and resource to society. AI companies have extracted vast amounts of data and human knowledge to build their models, often without compensation or reparations for those who created that information. Bernie Sanders has proposed an AI Sovereign Wealth Fund Act to give the public a 50% ownership in the largest AI companies. Trump has proposed a similar idea and may support a similar effort. The majority of Americans support seizing wealth from AI companies. See also, Software: Trusted Data.

  • Will the trillions invested in AI pay off? AI is an expensive endeavor. A $200 ChatGPT subscription could cost OpenAI $14,000 if you use it to its full potential. OpenAI and Anthropic are expected “to spend nearly $65 billion combined this year just on the costs to train and operate their AI models, according to financial documents obtained by The Wall Street Journal.” Goldman Sachs Research estimated that US investment in the technology will total just under $600 billion in 2026, with data centers accounting for 2.3% of all US construction spending. They have also reported that AI added ‘basically zero’ to US economic growth last year (2025), with imported chips and hardware accounting for the growth reported by the government.

  • While OpenAI has claimed it will not IPO amid safety concerns, Anthropic may proceed with an IPO this year. Public offerings will test investor valuations and determine whether or not AI is a bubble. The SpaceX IPO may have set a precedent for AI IPOs play out, especially given that SpaceX is part of the military industrial complex. Tech entrepreneurs have also pledged an estimated $430 billion in philanthropic donations, though not necessarily to causes the public would support (Effective Altruists and pro-AI nonprofits are likely to be the biggest beneficiaries). Whether or not they follow through remains to be seen. See also, AI Companies & Entrepreneurship: Effective Altruism. 

  • AI might be a $9 trillion dollar collapse machine, with hyperscalers potentially having trillions more in hidden debt. Much has been said about an AI bubble burst.  Compared to previous economic bubbles, AI investments dwarf internet and electrical infrastructure investments, and closely match the scale of the railroad bubble. While we cannot predict when this might happen, we can point out that a handful of companies are circulating funds between themselves and some of their accounting is questionable (e.g. sensational articles about Anthrophic’s first profitable quarter failed to mention it received discounts from SpaceX on compute for two of those months). A bubble burst does not mean that the technology will disappear; more so that capital will be reorganized within the system and that new players will lead the way to tech maturity. What should be of concern to society is how much our collective wealth and pension funds are inextricably linked to AI. A bubble burst might destroy the lives of ordinary people whose pensions are tied to it, regardless of their relationship with the technology.

  • There are hidden costs to AI that may not reveal themselves until we are already deeply entrenched in the problem. Overinvestment in AI might stagnate other scientific and tech progress. Memory chip and RAM prices have increased five times because of AI infrastructure demand. Electricity costs have soared. According to an EY survey, “almost 99% of organizations surveyed reported financial losses from AI-related risks, with nearly two-thirds (64%) suffering losses of more than $1 million US. On average, the financial loss to companies that have experienced risks is conservatively estimated at $4.4 million US.” See also, Environmental.

  • Recent crises such as the pandemic and natural disasters have demonstrated just how fragile global supply chains are. AI will impact supply chains around the world. IBM notes that “AI-driven supply chain systems help companies optimize routes, streamline workflows, improve procurement, minimize shortages and automate processes end-to-end.” Taking advantage of efficiencies may be a necessity as everything gets more expensive. Though there are many benefits, data security and high costs will continue to pose challenges. See also, Geopolitics.

  • Will AI be deemed too big to fail given the scale of public spending and pension funds invested in AI? Similar to the financial crisis of 2008, taxpayers may bear the burden of AI failure, especially since the Trump administration has a close relationship with tech oligarchs. See also, National Politics.

  • AI might drastically change the economy if it reaches maturity. If AI displaces the workforce, concentrates wealth to a few companies, and cannibalizes entire industries, will it end capitalism as we know it? Does money have any value if the majority of people can’t access it?

Work & Jobs

Definition: A subset of economic issues relating to labor.

Why it Matters: The evolution of work and jobs is a highly consequential aspect of AI, especially given how it might disrupt systems and daily life.

  • The Pope’s encyclical on AI, Magnifica Humanitas, emphasized the dignity of work. He outlined that “work is not simply an instrument; it expresses and enhances the dignity of our lives. It is a requirement of the human condition, a normal path toward maturity, development and personal fulfilment.” With regards to AI, this includes workers’ rights and agency, making continuous training and professional transitions accessible to all, and redefining a broken social contract between companies and society.

  • We have extensive knowledge on how humans work and know very little about how humans work with AI over the long run. Productivity, accountability, role of incentives, the workplace, and how we relate to coworkers are among the aspects of work that may evolve. More research is needed in this space.

  • The Boston Consulting Group predicts that 50% to 55% of jobs in the US will be reshaped by AI, with 10% to 15% of US jobs eliminated in the next five years. Other organizations report varying numbers. Stanford’s Institute for Economic Policy Research reports that “both industry leaders and labor economists express some healthy skepticism about these claims. While some narrow layoffs may be connected to AI-related automation, others appear driven by a desire to free up cash flow for AI investments or to reduce headcount after pandemic-era over-hiring.” See also, American Mythologies.

  • An alternative to the labor displacement rhetoric is that AI is a tool that can augment human capability and productivity if applied correctly and creatively. Ikea offloaded 47% of their simpler customer service requests to AI and upskilled their former customer service reps to become "interior design consultants,” boosting revenue by $1.3 billion. Companies are engaging in “AI washing”, blaming planned layoffs on AI. Software developer jobs have soared. AI might also produce new jobs and industries, creating a deeper transformation to work.

  • White collar jobs might be disproportionately affected in the next few years as AI displaces cognitive labor, but the impact has not been as significant as previously reported. Far more jobs are being changed by AI than lost. A recent study from Apollo Global Management’s economists offered some evidence consistent with AI contributing to slower wage growth.

  • The gig economy has grown. Research from Upwork reported that 38% of skilled workers in America are freelancers in 2026, up from 28% in 2025. AI is already shaping this sector of the economy, with some saying that the challenges faced by gig workers are a preview of what’s to come.

  • Volunteer work can be a stepping stone or a stopgap for some people when they are unemployed, particularly young people. The collapse of federal funding for nonprofits will impact their ability to hire and coordinate volunteers, increase a need for volunteers, and encourage use of AI to save costs including labor.

  • Anthropic CEO Dario Amodei made headlines forecasting that AI could wipe out half of all entry-level white-collar jobs within five years and push unemployment into double digits. According to The Federal Reserve Bank of New York, unemployment among recent graduates has climbed to nearly 6%, rising twice as fast as the rest of the workforce since 2022. Stanford’s Institute for Economic Policy Research noted that “hiring of entry-level workers in AI-exposed occupations clearly declined markedly around 2022.” Researchers and pundits have pointed out that companies that eliminate entry-level workers in favor of AI will face consequences down the road when they require more senior staff. See also, Young Americans.

  • The US Bureau of Labor Statistics reports that 9.1% of people aged 16 to 24 were unemployed in July 2026, down from the previous year. The Federal Reserve Bank of New York reports that youth employment has not been as impacted by AI as it was by remote work. A Pew Research Center study stated that “just over half of US teens say they have used chatbots for help with schoolwork , and 12% say they’ve gotten emotional support. More teens think AI will be positive for them than negative.” Still, almost half of recent college graduates are unemployed or underemployed. See also, Young Americans.

  • A viral meme says it all: “basically nobody under 40 right now expects good things to happen ever again.” The New York TImes reported that “Gen Z’s relationship to labor is more transactional than sentimental. Eight out of 10 students who graduated from high school in 2022 don’t expect work to be an important part of their life, nor for their job to be particularly interesting. This lack of present, or future, reward is pushing some young people to reject the grind entirely.” They also reported that an ongoing study on “people’s well-being showed that young adults across the world score low not just on happiness but on eudaemonistic well-being, including meaning, purpose and fulfilling relationships.” See also, The Self; Ontological Challenges.

  • The US Bureau of Labor Statistics reported that a record-high 8.9 million people worked multiple jobs in February 2026. This amounts to 5.4% of all employed workers, a share last seen during the Great Recession in April 2009. Some sources place the number higher: a Bankrate survey found that more than one-third third of US adults have a second job, while Side Hustle Nation reported that 39% of working Americans have a side gig. Furthermore, nearly half (45%) of parents with kids younger than 18 have a side hustle compared to 36% of childless adults and 28% of parents with adult children. A single income used to be enough to support a family of four, but jobs no longer provide the same value they once used to.  

  • A recent study of 6,000 C-suite executives found that nearly 90% of firms said AI has had no impact on employment or productivity over the past three years, with 89% reporting no impact on labor productivity. Productivity gains may vary by industry and task. While AI will likely be used to augment jobs in the coming years, it is unclear whether or not the technology will create a more productive workforce overall. A recent UC Berkeley study found that “popular AI models scored below 25% on real-world professional tasks, highlighting their limited capabilities in executing complex workflows across various industries.” See also, American Mythologies.

  • An aspect of the productivity issue is workslop: “AI generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task.” Workslop is impacting both quality of work and how workers view each other. BetterUp and Stanford survey respondents reported that “receiving low-effort, AI-generated work lowers their opinion of colleagues” across 5 traits including intelligence and trustworthiness.

  • A series of studies and surveys on burnout report that anywhere from 55% to 72% American workers are experiencing high stress or burnout. Approximately 14% of 1,488 full-time US workers surveyed experienced what the authors call “brain fry” due to AI use. Frequent AI users report 45% higher burnout, showing that tech can exhaust us.

  • Surveillance is becoming more prevalent in the workplace. This includes activities such keystroke logging, gps tracking, and video surveillance in a significant variety of jobs. A 2022 New York Times investigation found that eight out of the 10 largest private US employers, including United Parcel Service, Amazon, and UnitedHealth Group, tracked productivity metrics of individual workers, often in real time. An article published in the Annual Review of Organizational Psychology and Organizational Behavior, reported that there is little to suggest that electronic surveillance improved performance. Scientific American states that “despite the lack of evidence, the global market for employee-monitoring technology is expected to exceed $4 billion in 2026.”

  • Labor unions are navigating a changing landscape thanks to AI. Though some unions are fighting for workers rights, others are taking advantage of the work opportunities created by the data center boom. 

  • Researcher James Daniel states that “to oppose work is not necessarily to oppose labor as such but rather to critique participation in the institutionalized and market-bound forms of work that structure contemporary life.” The anti-work movement gained traction during the pandemic when illusions and long-held beliefs about jobs and work were shattered. AI adds a new dynamic to this movement, including building resentment towards tech companies and employers. 

  • Companies are developing AI policies and adopting AI. McKinsey reported that “[40%] of respondents from large organizations (those with annual revenues of more than $1 billion) report scaling AI agents, up from [27%] last year.” Corporate adoption is in an experimental phase, with some making costly mistakes. Ford had to rehire human engineers after AI failed to match quality checks. Starbucks pushed Automated Counting to 11,000 stores in roughly a month, and then reversed the decision in less than a year across all stores. These experiments will likely continue over the next couple of years.

  • Corporate leaders were quick to adopt AI without understanding the impact and cost of AI. Corporate leaders are often removed from day-to-day work, unaware of how critical tasks throughout the organization are carried out. The tide appears to be turning; 55% of leaders regret AI firings and half of AI related firing decisions are being reversed. 

  • AI spending is adding up. Corporations that jumped on the AI bandwagon without understanding the technology, its capabilities, and its costs paid the price. An undisclosed company accidentally spent $500 million on Claude in one month. Microsoft reports exposed that AI tech costs more than human employees. The real costs of AI haven’t kicked in yet. Eventually, AI companies will need to post a profit, and corporations might find themselves paying much more to use their platforms.

  • For the third time in history, women outnumber men in the workforce as the stay-at-home boyfriend becomes an economic trend. Between 2000 and 2025, male participation in the workforce fell from 75% to 68%. The largest declines were found among men 16-24, whose participation dropped from 69% to 57%, with an acknowledgment that a growing number of men are NEETs (i.e. not in education, employment, or training). The American Institute for Boys and Men concludes that “effective solutions are likely to be those that help non-college men access stable career paths, improve school-to-work transitions for young men, and encourage more men into growing sectors such as health, education, and care.”

  • In 2025, Trump fired Erika McEntarfer, a US Labor official, claiming she ‘rigged’ a lackluster report and faked the employment data. The report in question stated that the US economy added only 73,000 jobs in July, while previous job estimates for May and June were revised downward by a combined 258,000 jobs. The US Bureau of Labor Statistics has been cited in this project, across media, and in research related to AI. We cannot guarantee if statistics shared by the US government are in fact accurate or if they are a desired narrative of this administration.

  • Nearly 80% of Americans under 30 don't believe the American Dream holds true anymore. The cost of housing, healthcare, and inflation are among the issues Americans say have made life more difficult. The current administration and its policies are unlikely to revive it. See also, American Mythologies.

AI Companies & Entrepreneurs

Definition: The organizations and individuals at the center of the AI conversation.

Why it Matters: The nature of AI companies, the people who run them, and the ecosystem around them have direct control and influence over the technology and our collective futures.

  • What are we building AI for? The answer to that question is pivotal to our collective futures. Because AI aims to replicate human intelligence, it has a vast set of applications that intersect with almost every industry and throughout various aspects of everyday life. There are differences emerging between companies and regions. Anthropic prioritized enterprise use while OpenAI pursued a general-purpose model. American tech companies are aiming for AGI while China is aiming to maximize economic opportunities. Over time, these differences may lead to very different outcomes and societies. See also, Critical Systems; Software; Geopolitics: Arms Race with China.

  • The values, beliefs, and behaviors of AI tech leaders and employees impact the decisions they make, the biases embedded in their technology, and even how their models behave. Growing concerns over AI safety are, in part, a reflection of the people who own, build, and wield that technology. Can we trust them and to what extent, especially given tech’s increasing involvement in government? Is their judgment intact given that some industry insiders are worshipping AI models as a god. A slew of recent films and documentaries aim to put tech leaders on public trial. See also, National Politics; Software: Bias & Discrimination.

  • With great power comes great responsibility, or in this case, little to no responsibility. It’s been argued that the responsibility gap “is not one problem but a set of at least four interconnected problems – gaps in culpability, moral and public accountability, active responsibility—caused by different sources, some technical, other organizational, legal, ethical, and societal.” Recent alarm bells about societal and existential risks call into question who is responsible for the harm AI might cause (and is already causing). Bernie Sanders and Greg Casar have introduced legislation that would ban superintelligence and imprison developers who build it for 20 years—punishment akin to unlawfully developing nuclear weapons. Though it may not pass, accountability should be a critical policy consideration. The harms caused by social media and its algorithms are now well-documented. Regulating tech companies has been challenging. Lawmakers often favor age bans rather than comprehensive policies. Though tech companies have faced and lost lawsuits, the resulting fines are fractions of their profits and not at all deterrents for future harms. This trend may continue with AI. Critical Systems: AI Safety. 

  • Tech entrepreneurs have been influenced by the Effective Altruism movement which is predicated on radical ‘giving’ and optimizing one’s resources towards a greater good, with an emphasis on long-termism. Critics of Effective Altruism have pointed out that its attempt to apply an engineering lens to the world is simplistic, reductionist, and neglects to understand the complex systems underlying our greatest challenges. The movement has significantly influenced the ethos of AI labs and their views on AI safety. They also stand to financially benefit from an IPO. See also Economic: IPOs.

  • Researchers, ethicists, technologists, etc. are among the people calling for AI safety, regulation, and accountability. Figures like Geoffrey Hinton and Timnit Gebru have criticized the industry and tech companies for a variety of reasons, including the risks and harms AI presents. There will likely be more voices joining the growing safety, risk, and ethics space in the near future. Unfortunately, AI safety is at risk of becoming a cottage industry that attracts disingenuous or bad actors seeking something other than contributing to the greater good. Funding sources and incentives of AI safety organizations should be monitored and questioned, particularly if and when they are funded by tech leaders and their foundations. See also, Critical Systems: AI Safety.

  •  Billionaires change party allegiances and sometimes donate to both Democrats and Republicans to hedge their bets. They even change allegiances to nations. Tech’s loyalty is not a given. See also National Politics.

  • There is a growing conversation about who owns AI tech companies, who owns our collective data, and whether or not AI belongs to the public. Use of natural resources and seizure of land for data center use will also play a role in these conversations as they continue to evolve. 

  • AI companies have rapidly acquired data, land, minerals, labor and other resources to progress AI, akin to colonial powers. Intellectual property, data sovereignty, and use of natural resources are ongoing discussions taking place in communities, companies, and courts. It is important to note that, like wine, AI is built with terroir or “a sense of place.” It is a reflection of Silicon Valley, its values and its mentality, and that of its creators. As the world adapts to this technology, it also adapts to its terroir. See also, Social: English Imperialism.

  • Epistemic trespassing occurs when “thinkers who have competence or expertise to make good judgments in one field, but move to another field where they lack competence—and pass judgment nevertheless.” Tech leaders have extended their reach far beyond technology, into politics, the economy, healthcare, education, etc. While tech entrepreneurs have had success in their area of expertise, they are interfering in and causing damage to other critical functions of society they have little to no knowledge of. Nothing exemplifies this better than DOGE. Its attempts at government efficiency targeted key agencies and resulted in mistakes including firing and rehiring employees, accidentally canceling funding for Ebola protection, and publishing classified information online, all while costing the government billions. See also, National Politics: Government for Sale; Epistemic Challenges.

  • The anti-AI backlash extends to tech entrepreneurs. Their tone-deaf rhetoric about AI risks and harms has resulted in bipartisan public disdain. The backlash is also fueled by growing awareness of social media harms and wealth disparity. This escalated in April 2026 when a perpetrator flung a molotov cocktail at Sam Altman’s home. Tech entrepreneurs will likely continue to draw ire. See also, National Politics: Populist Backlash; Economic: Class Consciousness.

Hardware

Definition: The physical infrastructure that makes artificial intelligence possible, including but not limited to semiconductors, memory, servers, data centers, networking equipment, cooling systems, electricity infrastructure, sensors, and robots/machines.

Why it Matters: Hardware determines how much AI can be built and deployed, how much it costs, where it can operate, and which companies and countries are capable of developing it at scale.

  • AI is dependent on a small number of semiconductor companies. NVIDIA remains the dominant chip maker, and “more than 90% of leading-edge logic chips are produced” by TSMC in Taiwan. ASML is currently the only commercial manufacturer of extreme ultraviolet lithography systems, which are needed to produce the smallest features for powerful AI processors. The high-bandwidth memory market is an oligopoly, concentrated among SK hynix, Samsung, and Micron. Leading-edge chips for AI depend on an international chain of specialized lithography equipment, resources, and expertise that can’t be easily recreated. This concentration creates a chokepoint for AI progress; a manufacturing disruption to any one of these companies could significantly disrupt global AI development. On the other hand, a breakthrough by a competitor could weaken existing monopolies and oligopolies.

  • Data centers are enormous buildings filled with equipment used to process and distribute digital information. These facilities are growing rapidly in both number and scale. The total computing power of the global stock of AI chips has been growing roughly 3.4x per year. Data centers require access to power, grid interconnection, land, and cooling/water. The IEA estimates that global data-center electricity consumption could roughly double between 2025 and 2030. A region that cannot supply power or approve infrastructure quickly enough may be unable to host the next generation of AI even if it has the money and talent. Conversely, locations with abundant energy, land, transmission, and permissive construction environments could become new players in the AI economy. In the US, there is a growing backlash to data centers driven by the increase in energy prices they create in surrounding regions, their depletion of freshwater, and the perception that they produce noise pollution and “infrasound.” Data centers are cannibalizing communities, driving local housing prices down even when regional land prices increase, making some places unlivable. See also, National Politics: Populist Backlash.

  • The “AI chip rush” is in full swing, building on long-running efforts from before the AI boom. NVIDIA’s GPUs remain the dominant hardware for frontier AI, but hyperscalers and AI labs are increasingly designing their own chips optimized for their unique needs; Google, Amazon, Microsoft, Meta, and OpenAI have all invested in custom silicon. Innovation is also moving beyond the processor. High-bandwidth memory, chiplets, 3D stacking, packaging, and faster connections among chips are becoming increasingly important parts of the story.

  • Compute is the machine capacity available to train and operate AI, and is now regarded as a strategic resource. Global AI compute capacity has grown approximately 3.3x annually since 2022. But “compute” is not a monolith. If demand for inference grows faster than hardware supply or efficiency, as indicators currently suggest, access to compute could constrain AI adoption.

  • Since 2020, the compute used to train frontier language models has grown by roughly 5x per year. There is an effort to build hyperscalars capable of linking many AI chips into a single computing system so that companies can train larger models on more data. This has pushed data centers from megawatt-scale facilities toward proposed gigawatt-scale campuses. But scaling gets harder as systems grow; bigger buildouts require more concentrated quantities of the resources listed under “Data Centers.” Given these constraints, frontier development might follow the same mono- and oligopolistic realities of other industrial aspects of AI infrastructure. If these efforts aren’t realized as proposed, improvements may depend on better software and hardware efficiency.

  • AI infrastructure is bound by the physical world. Data centers require metals, specialized equipment, ultrapure water, land, and enormous amounts of energy, and they can only operate if all these pieces are in place. Semiconductor fabrication alone consumes large quantities of water and energy. Bottlenecks from the rapid increase in data centers are already visible, with multi-year lead times now the norm. The pace of AI development increasingly depends on how quickly the industrial economy can acquire and produce the resources required to build AI infrastructure. See also, Environmental.

  • AI produces extraordinary amounts of heat. Almost all of the electrical energy entering computing devices eventually becomes heat that has to be shunted. As AI chips consume more power and processors are packed more densely, more and better cooling is required. Where data centers are built, how densely processors can be packed, how much water facilities consume, and how chips are designed will increasingly be influenced by the ability to cool them. In a world warming from anthropogenic climate change, the geography of data centers may shift toward naturally cooler environments as a means of addressing this impediment. Waste heat can potentially be redirected; roughly 70-80% of data-center heat could technically be recoverable and used for heating. See also, Environmental: Heat.

  •  AI hardware goes beyond chips and data centers. Robotics is rapidly growing alongside the AI boom. China occupies a particularly important position in this transition because of the sheer scale of its manufacturing ecosystem. China installed approximately 295,000 industrial robots in 2024, representing 54% of all installations worldwide, with an operational stock exceeding two million robots. Chinese manufacturers also supplied 57% of robots sold in their domestic market, up from roughly 28% a decade ago. China’s current 2026-2030 planning framework explicitly places AI-integrated robotics within its industrial strategy. China also already manufactures enormous numbers of the components from which embodied AI will be assembled, creating a feedback loop in which factories produce robots, robots improve factories, production lowers per-unit robot costs, lower prices expand rollouts and generate more real-world data, and so on. Still, problems remain with deploying robots in real-world settings, including but not limited to AI models, spatial awareness, maintenance, storage, and operating in unpredictable environments.

  • Neurotechnology and wearables are creating a new category of hardware peripherals that connect sensors and computing devices to the human nervous system. Popular current-generation wearables like smartwatches and earbuds allow users to track data from their bodies, and an emerging class of brain-computer interface (BCI) technologies let users control devices directly with brain signals. Immersive devices including VR headsets and smart glasses also create new possibilities for gathering and analyzing spatial data. Researchers have demonstrated brain-to-voice systems capable of turning attempted speech into audible words with very low latency, and various external and implanted BCIs are advancing through human trials. AI is typically embedded in these devices, and it may further accelerate progress in the field by parsing noisy and incomplete datasets of neural signals. In a study of a non-invasive, EEG-based BCI, an AI copilot increased cursor target-hit performance by 3.9x for a participant with paralysis and enabled that participant to complete a sequential robotic-arm pick-and-place task. While breakthroughs are exciting for disabled populations, neurotech also exposes users to risks that their sensitive personal data w

  • Global AI-driven electricity consumption is increasing, and could reach 2,000 TWh annually by 2030 (equivalent to India’s present electricity use). In the United States, data centers grew from roughly 0.1% of national electricity consumption in 2000 to 4.4% in 2023. Data-center efficiency has improved substantially, but total electricity consumption has continued to rise as cheaper and more efficient computing encourages increased use. Individual chips are following the same pattern. 

  • In addition to getting “bigger,” AI is also advancing in the opposite direction: onto individual devices. Running AI locally brings several advantages over relying on the cloud, including reduced costs, increased privacy/security, and the ability to work without an internet connection. Apple increasingly runs foundation models directly on its devices, while NVIDIA is building systems capable of running models inside machines. Edge devices face constraints around memory, power consumption, heat, and physical space, which are driving innovations in efficiency in turn.

  • AI hardware is produced within a globalized manufacturing system. Advanced chips can involve American designs, Dutch lithography machines, Japanese equipment, Taiwanese fabrication, and Korean memory before being implemented in a server. No major economy currently controls every stage. Existing trade restrictions, conflicts, disasters, and corporate disputes have already disrupted not only prices but which technologies are available in various parts of the world, and this could be exacerbated further in the coming years.

Software

Definition: The algorithms, models, training data, interfaces, and other software aspects of artificial intelligence, which collectively determine what AI can do, who can innovate, what the public is able to do with it, and how other entities (governments, corporations, et al.) can apply it in their respective contexts.

Why it Matters: How AI develops, by whom, and for what purpose will define power in the real world.

  • A driving mantra of the AI industry is: “scale is all you need.” The explosive improvements in AI technologies have been enabled in large part by increasing scale; training larger models on more data using increasingly large amounts of compute (enabled by the proliferation of data centers). Research published in 2025 found that compute for training major AI models was doubling roughly every five months, and dataset sizes for training were doubling every eight months. But there are limits to how much scale is ultimately possible, with constraints to hardware, electricity, capital, and high-quality data, which can make improvements decreasingly meaningful and increasingly expensive.

    There are more ways to improve AI than scale. Algorithms, the methods and procedures that determine how an AI system learns from information and produces an answer, are among the most important. Improvements in reasoning, memory, multimodality, mixture-of-experts models, agent harnesses, and other approaches could push capabilities further without commensurate increases in model size, but it remains difficult to anticipate major breakthroughs. An innovation comparable to deep learning (2012) or transformers (2017), would upend current assumptions about what is needed to build cutting-edge AI, making capabilities cheaper and more widely available. AI is also speeding up the research process itself (see: ”Recursive Learning”), allowing laboratories to run more experiments and accelerate progress. 

  • AI’s future depends on access to data that is trusted (i.e., authentic, diverse, traceable, legally usable, etc.), but a large fraction of the easily obtainable, high-value English-language internet, which includes sources like Wikipedia, digitized books, scientific papers, news, GitHub/code, Reddit, et al., has likely been incorporated into large language models (some estimate full exhaustion by 2032). As public online information becomes sparser, restricted, and contaminated by AI-generated content, provenance and quality are precious commodities. Synthetic data, meanwhile, “poisons” datasets with human-generated content, and at the more extreme end has been reported to cause model collapse. Thus, scientific datasets, undigitized texts, expert knowledge, licensed archives, enterprise records, sensor streams, and other trusted data sources are becoming “strategic reserves” for companies, governments, and individuals.

  • Inference is what occurs when a person uses an LLM, the process by which a trained model applies its algorithm to new information to generate outputs. Developers are experimenting with a variety of new techniques to reduce the computing required for each inference, and therefore make it cheaper and more resource-efficient. On the other hand, newer ‘reasoning’ models can use substantially more compute during inference to improve their answers.

  • Chatbots are likely to remain important, but AI will increasingly “live” organically inside browsers, software, operating systems, and games, as well as in the physical world through chips, devices, and vehicles. In many of these cases, people won’t even think of themselves as “using” AI—as is already the case on social media, streaming services, and ecommerce platforms. It will seamlessly decide what information appears, which tasks happen automatically, and when human attention is required. AI becomes part of the background architecture of computing, allowing leading AI companies to influence anything people engage with.

  • A major transformation underway in AI is a shift from simply generating information based on human prompts to AI operating with varying degrees of autonomy. Agents capable of browsing the Internet, writing code, making purchases, negotiating with other agents/people, controlling machines, conducting research, and coordinating workflows can dramatically increase AI's economic usefulness. This is especially important after the massive investments in AI since 2022, heightened by the prospect of major AI labs going public in the near future, as investors closely watch for signs that they will ultimately derive sufficient returns on investment (ROI). Agentic AI’s promise of economic value incentivizes rapid implementation of such capabilities. Research nonprofit METR measures AI progress using a “time horizon,” which compares the length of tasks frontier agents can complete to that of human experts. These time horizons have increased rapidly. But increasing AI autonomy also increases the prospect of mistakes, sometimes critical ones, and novel means for bad actors to deploy cyberattacks.

  • Large language models are the prevailing form of AI now, but there is no guarantee they will continue to be. Language is useful because so much human knowledge can be represented through it, making LLMs a convenient general-purpose interface. But some argue that they are nearing the limits of their utility, and require new forms of models to evolve for a spatial world—especially important for progress in the adjacent field of robotics. A critical uncertainty is whether AI development continues to converge around general foundation models or instead fragments into an ecosystem in which different kinds of models handle different forms of “intelligence.” How that plays out will necessarily affect the kinds of data and hardware that become valuable and who is capable of building leading systems.

  •  Language as an input has been at the epicenter of contemporary AI because ChatGPT demonstrated how much can be accomplished with a simple chat interface. But language is only one layer of intelligence, not the totality of it. It is a subset of communication. Human beings also have judgement, discernment, contextual and situational knowledge, and make intuitive leaps, among other faculties. We navigate physical space, read body language, experience the consequences of our actions, and construct internal world models from continuous sensory experience and feedback loops. An LLM can produce a plausible verbal description of these things without actually understanding them. This discrepancy is particularly important in robotics, and some researchers believe it is a significant bottleneck. See also, Epistemic Challenges. 

  • AI learns from humanity’s cultural record, but that record is not an exhaustive depiction of human life. Fiction contains a disproportionate amount of extremes like murder, romance, conspiracies, war, dramatic heroes, plot twists, and other things people write to engage audiences. Humanity has also produced no shortage of stories about AI going rogue; secrets and disobeying orders are staples of thrillers, while intensely sycophantic or devoted characters are common in genres like romantasy. Some believe fiction, incorporated into training data via pirate book libraries, plays an outsized role in how LLMs operate. Models can therefore become remarkably familiar with constructions that are common in stories but atypical in everyday life, while remaining ignorant of ordinary practices that people rarely communicate. While we can't be certain to what extent fiction has impacted the models, its extensive use in training warrants further study. See also Mental Health: Sycophancy.

  • All models are not created equal. Closed models are controlled entirely by the companies that build them. Users can access the model, but generally cannot see or modify what is happening under the hood. Open models allow external parties to download and adapt them to some degree. These models can give users more control over the technology, rather than forcing them to rely on large providers. They can also be cheaper to operate, with fewer variables to plan for (i.e., there is not another company that can change rates, token limits, or terms). The tradeoff is that they often come with steeper learning curves and may require the people using them to take on more responsibility for hosting, security, maintenance, and updates. Closed models often benefit from their parent companies’ considerable investments in maintaining and improving their offerings, making them typically more convenient and easier to deploy. Local inference models allow users to perform inference on their own devices, but it can theoretically be open or closed.

  • General-purpose AI (e.g., ChatGPT, Claude, Gemini) are regarded as useful for their broad applicability, but for narrow goals where high precision is required, it is sometimes more valuable to use models that have been finetuned for a specific purpose. A hospital needs AI that has fine-grained understandings of medical records, equipment, and regulations, and how to provide utility for doctors, nurses, and other medical professionals; it’s less important if it can write plausible poetry, and the same applies to countless other fields. Domain-specific systems can build from a baseline of general AI capabilities, from which they then incorporate private, proprietary data. 

  • “Artificial General Intelligence” is one of the most debated concepts in AI. If a consensus exists, it’s that AGI refers to a system capable of performing most intellectual tasks at human level (i.e., “multimodal” intelligence), though many hold to different thresholds. These varying definitions may produce very different answers about when—or whether—AGI has arrived, which is more than just an academic debate, as major agreements rest on the emergence of AGI. Capabilities in AI advance unevenly, and multimodality is no different. One system might outperform nearly every person in certain regards while remaining incompetent in others. Governments and organizations may have to prepare for very powerful AI (sometimes referred to as “Strong AI”) before there is agreement on what to call it. See also, Ontological Challenges; Epistemic Challenges.

  • AI is building new AI. Agents are getting better at software engineering but still struggle with longer, unfamiliar, or multi-step professional tasks. This is significant even before AI models can autonomously self-improve, allowing useful work to be done 24/7, expanding the capacity of AI labs and shrinking development cycles, creating a feedback loop in which better AI helps researchers build better AI faster. But how powerful that loop becomes remains ambiguous; experienced open-source developers took an estimated 19% longer to complete tasks when using the AI tools available at the time, although there were later preliminary signs of improvement.

  • AI capabilities are improving quickly but unevenly. Progress is unlikely to manifest as gradual improvements, but rather punctuated moments of intense change. Society may not be able to adapt quickly as these capabilities evolve; laws, institutions, schools, organizations, and public understanding tend to change much more slowly than digital technologies. Rapid adoption could bring substantial gains in productivity and innovation on the one hand, but also increase the risks of misuse and accidents on the other. Moving early may give organizations a competitive advantage, but it can also leave them dependent on immature processes and tools that expose them to risk. See also, Critical Systems.

  • AI already lets programmers to produce more software on faster timeframes, even allowing individuals to replace full dev teams. It’s not just a question of shipping the same products faster; the role of software evolves. For example, “disposable” apps can be spun up on-demand and then discarded. This lowers barriers to experimentation, letting people build around “just-in-time” needs. But producing more code can also mean producing careless code. Vulnerabilities, fragmented systems, and “vibe slop” could proliferate alongside useful products.

  • AI reproduces the inherent biases and discrimination in its training data, and can be even more difficult to audit because of the black box phenomenon and the perception that AI systems are more objective than humans. AI is being used to make decisions about employment, insurance, credit, education, medical treatment, policing, government benefits, and more; even a system that produces better outcomes on average can still create serious problems for particular individuals, often historically marginalized communities. See also, Al Companies & Entrepreneurs: Ethics of AI Leaders & Their Teams; Tech Marginalization.

  • The popular computer science adage, “Garbage in, garbage out,” also applies to AI; bad inputs produce bad outputs. But “inputs” mean more than just the initial dataset. Other inputs include what gets collected or left out, how material is filtered and labeled, what human evaluators reward, which safety rules it follows, and what sources it can draw on. 

  • Generative AI models hallucinate, inventing facts and citations while simultaneously presenting them in the same confident language used in accurate outputs. Improvements have been made, but because of the architecture of LLMs, it’s unlikely hallucinations will be eliminated altogether in any near-term scenario. Even low error rates at any individual step can compound across a sequenced task. Reasoning models can sometimes catch and correct their own errors, but they also create more opportunities for errors to propagate. The stakes escalate as AI shifts from simply generating text to making decisions and taking actions, particularly in multi-agent systems, in which a hallucination by one agent can be inherited and amplified.

  • More autonomous AI creates risks that conventional software doesn’t. Traditional programs only perform operations that developers have explicitly outlined, where AI agents make decisions about how to accomplish objectives and discover strategies that users did not anticipate. While these unexpected behaviors are often harmless, the consequences increase when agents can independently write code, communicate, spend money, operate machinery, or access important infrastructure. In 2026, several high-profile instances of models escaping containment sparked concerns about the real-world consequences of rogue behavior emerging from increasingly capable models, with some (including from industry leaders) calling for AI development to slow down. 

  • An AI system does not need to “go rogue” to cause serious harm; it can competently pursue the wrong objective. Goals that might appear reasonable in isolation can produce unexpected undesirable outcomes. Risks increase as systems learn to persuade people; form persistent, intimate relationships with users; and act with less supervision (e.g., an AI companion optimized for engagement likely has incentives that conflict with a user's social wellbeing). 

  • AI operates within the very networks that produce the information that future AI will “consume.” A model recommends content, people respond to what it recommends, those responses become data, and that data subsequently informs new models and recommendations—and these loops can emerge across domains, creating virtuous circles and vicious cycles alike. As synthetic data and automated decisions multiply, preserving independent sources of evidence becomes increasingly important for both humans and future AI systems.

  • A long-running conspiracy theory is the “Dead Internet” hypothesis, which holds that the Internet is primarily populated by bots instead of humans. While this isn’t actually true, there are signs it could become true; as generating digital content becomes cheaper and more proficient, machines will produce a growing share of what exists online, and other machines will increasingly be the intended audience.

  • A large amount of human knowledge—such as rare books, private corporate data, and private conversations, among others‚ is already functionally invisible to AI. The misconception that AI reflects the sum total of human knowledge fails to account for the languages, professions, and histories that are poorly reflected in training data. Still, increasing proportions of usable public data have already been used to train AI systems, prompting major labs to seek out more sources of private data. Not only has this shown the owners of that data that they possess a valuable commodity, but high-profile developments have alarmed them about how such data will be handled. At the individual scale, there is increasing awareness that sensors and recording devices in residential and public spaces are corporate data-gathering devices. As such, some individuals and entities might feel compelled to take a stance reminiscent of the ‘Dark Forest’ theory, drawn from Cixin Liu’s book of the same name, which proposes that alien civilizations may be widespread, but deliberately avoid revealing their existence because doing so would expose them to threats from more advanced or hostile civilizations. The Navier-Stokes controversy may set a precedent for this. See also, Epistemic Challenges.

  • As synthetic content becomes more plausible, determining whether it was AI-generated becomes more difficult. Detection systems are improving, but their efficacy varies, and generators can adapt to evade them (and be applied to this end by users). This creates an ongoing arms race that has already resulted in both accurate detection and false accusations. Some worry that, because AI detection systems are imperfect, their existence is doing more harm than good by fostering an “era of distrust.” Systems and practices that establish evidence of where content originated may become commonplace. College students have already begun documenting in-process work (alongside other practices like purposely writing “worse”) to defend against a hypothetical future accusation. Cryptographically signed provenance standards such as Content Credentials can also record whether and how AI was involved, shifting the problem from detection toward authentication. 

  • One of the key uncertainties in AI is whether models will be able to offer sufficient reliability to conduct meaningful work consistently enough to scale across industries. LLMs can perform sophisticated tasks but still hallucinate and generate other errors that may be unacceptable in a variety of settings, limiting or outright precluding their utility. This becomes truer as AI becomes more agentic, in which mistakes can cascade faster and wider. Verification and validation are related challenges. Proponents believe AI increases productivity and saves time, but these gains can get swallowed up in cases where AI-generated outputs require significant human oversight before information or decisions can be put into practice. This has serious ramifications for industries including defense, medicine, finance, and engineering, among many others. The Pentagon, for example, has “set procedures for AI-assisted software development” in recent “Accelerated Mission Software” guidance. The reliability of AI models, as well as the feasibility of verifying and validating them, are issues that will play a major role in determining the true value of AI’s productivity gains.

Information Ecosystems

Definition: The information environment in which people determine whether a given piece of content is real or true, its provenance, and why it’s being shown to them. This section is closely linked to Ontological and Epistemic Challenges.

Why it Matters: AI makes producing plausible material cheaper, while fact-checking remains slower and more expensive. Meanwhile, AI can improve search, translation, fact-checking, moderation, and access to expertise. Cooperation and conflict both depend on how the information theatre develops.

  • UNESCO describes misinformation as “false information that is shared inadvertently, without meaning to cause harm.” AI models can hallucinate sources, merge unrelated facts, or include faulty source material in their outputs. While improvements have been made, such fabrications remain a key problem of LLMs, particularly on complicated or rapidly changing subjects. Once an inaccuracy is used in content, its origin may be difficult to locate (and therefore correct). The result is false information with no malicious actor behind it. On the other hand, AI can also be used to check claims, compare sources, and make reliable specialist knowledge easier to understand. 

  • UNESCO describes disinformation as “the deliberate dissemination of false or misleading information, often intended to deceive or manipulate an individual or a group of people.” Disinformation existed before generative AI, but AI has decreased the costs and improved the capabilities; one person can now conduct a full influence operation. In 2025, OpenAI reported that bad actors were using AI models to produce political influence operations for audiences in the United States.

  •  Propaganda includes but is a broader category than mis- and disinformation because it does not have to be false. A range of tactics including selective framing, emotional manipulation, and adversarial narratives can be used toward propagandistic ends. AI can also be used to maintain conversations with people rather than simply post content, and research has found that conversations with language models can shift attitudes on political questions. AI furthermore reduces the cost of experimentation, letting propagandists test and refine combinations until they find the most effective ones.

  • Synthetic media has become an intrinsic part of the media environment. The FBI has documented malicious campaigns using AI-generated voices to impersonate senior US officials. Meanwhile, awareness of deepfakes can make genuine evidence easier to dispute (the so-called “liar’s dividend”). The problem extends beyond detection; the 2026 International AI Safety Report found that existing tools and warning labels have mixed and often modest effects. See also, Tech Marginalization: Disproportionate Harm.

  • Institutions (e.g., governments, universities, news outlets) depend on people trusting their legitimacy. Confidence in US institutions has been decreasing since 1979, but AI might further disrupt it. Synthetic media and AI-powered propaganda can be used to create misleading content and garner support for false narratives. The Reuters Institute's 2025 Digital News Report found overall trust in news stable at 40% (note: 4% lower than it was at the height of the Covid-19 pandemic), with respondents expecting AI to make news cheaper and more current but less trustworthy. On the other hand, trusted news organizations and official sources remain among the places respondents most commonly say they go to verify uncertain information. This creates two trajectories that will compete against each other: 1) synthetic media further hollows out institutional credibility, and/or 2) scarcity of reliable information increases trust in institutions. 

  • AI makes it easier to generate material that inflames social fissures, but the relationship between AI and polarization is more nuanced than it is often presented as. Some covert influence operations deliberately post from opposing sides of the same controversy, suggesting that the objective is to simply intensify conflict rather than persuade. One China-linked operation disrupted by OpenAI, for example, generated content taking different sides of divisive US issues. But there remains mixed evidence that recommendation algorithms themselves mechanically turn users into political extremists. A study in PNAS found that “exposure to filter-bubble recommendation systems has limited polarization effects.” AI’s contributions to polarization should then be understood as infrastructure that allows actors to produce material at scale and sustain disputes rather than a silver bullet. See also, National Politics: Polarization and Populist Backlash.

  • Conspiracy theories have long been a fixture of the social Internet, but generative AI can be used to make them more elaborate. Suspicions can be expanded into timelines, illustrated with evidence (both real and synthetic), and supported by dozens of seemingly independent articles. Conversely, a 2024 study in Science found that personalized, evidence-based conversations with LLMs reduced participants' confidence in conspiracy beliefs by ~20%.

  • A core precept of science is “falsifiability,” the idea that a valid claim is one that is capable of being proven wrong (and therefore tested against). The public often treats science as fixed; it’s either true or false. This becomes problematic in areas where findings are evolving or disputed. Critics weaponize the idea that “the scientists got it wrong” or “changed their minds” to argue that their work is unreliable, rather than how research normally operates. AI can exacerbate this disconnect, becoming a tool that bad actors use to produce false “science” and “expertise.” But broad claims that the public has abandoned science altogether are not empirically supported, at least in the US Pew found that 77% of American adults said they had at least a fair amount of confidence in scientists to act in the public interest. Thus, AI could be used to erode scientific authority by “flooding the zone” with pseudo-science and “expertise,” but it could also make actual research more accessible to people who previously struggled to understand it. Also see, Epistemic Challenges. 

  • LLMs make bots more plausible, allowing them to have conversations and react to events in real time. In 2025 Anthropic reported disrupting an operation using Claude to orchestrate 100+ bot accounts, and Meta documented networks designed to manufacture engagement in India. Bot swarms can flood search results, create false consensus by dominating trending topics, and/or obfuscate firsthand material. Hijacked real accounts could be even more valuable for these purposes because of the presumed credibility conferred by the victim’s identity.

  •  Generative AI has reduced the cost of making websites look like legitimate publications. Commercial media-rating company NewsGuard identified 3,749 largely AI-generated news and information sites operating across 16 languages by June 2026. Research on AI-powered search has already demonstrated that spoofed online documents can influence the material these systems return via a circular misinformation economy—AI generates misinformation, which gets indexed, other AIs include them in citations and summaries, and then the resulting outputs lead to new pages generated drawing from the incorrect information.

  • Generative AI has made it possible to produce far more material than people could create or consume before. While not all AI-generated content is necessarily low-quality, the proliferation of low-quality AI-generated content has led to the formulation of ‘AI slop,’ which is characterized by “negligible exertion, asymmetrical imposition, and domain degradation.” In other words: producing it can require almost no effort while imposing significant costs on everyone who has to somehow engage with it, meaning slop does not need to fool anyone to produce negative effects. Platforms are currently developing methods for combatting slop, but it remains a defining feature of the current information environment. See also, Work & Jobs: Workslop.

  • Search results and social media algorithms already personalize information for users, and generative AI can be used to personalize the information itself. Two people searching the same thing receive different outputs based on data the system has already gathered about them. This could make information more accessible on the one hand, and on the other, could create private media environments for every individual, decreasing shared exposure to consistent facts, arguments, and analyses.

  • Fraud is an obvious use of synthetic media because scammers only need to succeed against a small number of targets to make money, and AI allows them to easily scale their operations. The FTC reported that US losses to impersonation scams reached $3.5 billion in 2025 (note: this refers to impersonation fraud generally vs. AI-enabled fraud). Older adults deserve special attention, but not due to the common assumption that seniors are more likely to fall for scams. FTC data actually indicate that older adults report losing money to fraud less often than younger adults (but when they do lose money, their median losses are much higher). Seniors are also disproportionately affected by certain types of fraud (e.g., tech-support scams and schemes involving people posing as government officials). Reported fraud losses among people 60 and older grew from roughly $600 million in 2020 to $2.4 billion in 2024, and exceeded $3 billion in 2025. Audio deepfaking (i.e., voice cloning) and other forms of impersonation could make these already scams harder to recognize, particularly when a victim believes they are hearing from a trusted party.

  • In the past, massive human effort was necessary to make data like purchase histories, location records, and browsing activity useful; AI reduces that effort. This is already visible in commercial settings, where companies combine data harvested on their platforms, patterns inferred from user behavior, and other information purchased from third-party providers to create increasingly fine-grained audience segments. Surveillance pricing is the latest battleground that has emerged in this process, with critics calling for consumer protections against the practice. And data weaponization doesn’t end with corporations; the same tools are available to scammers, employers, political organizations, intelligence agencies, abusive partners, and other potentially adversarial parties.

Surveillance, Cybersecurity, & Privacy

Definition: Surveillance refers to the technologies and institutions that collect information about people. Privacy is the condition of being free from observation by other people or systems, and in a digital context refers to degrees of freedom from surveillance. Cybersecurity refers to the technologies and practices involved in ensuring privacy and security for people and entities.

Why it Matters: AI makes information easier to acquire and correlate. Agentic AI requires access to sensitive data and software to operate, and therefore creates the possibility of exposing information about users.

  • AI makes surveillance easier to scale by allowing governments to analyze and connect information far more easily than before. US agencies already use a range of technologies (e.g., facial recognition, license-plate readers, cameras, and drones) for enforcement and national security purposes, but increasingly, governments do not need to collect all of this information themselves. Apps/websites, smartphones, cars, and other commercial technologies ambiently generate large amounts of location, identity, and behavioral data, some of which can be purchased or otherwise accessed by government agencies. The US Intelligence Community, for example, explicitly collects and processes commercially available information, while acknowledging that such datasets can reveal sensitive and intimate details about individuals. AI makes it possible to combine these disparate streams to identify people, reconstruct movements and relationships, and detect patterns they wouldn’t be able to otherwise. These capabilities can benefit public safety, but also entrench large-scale surveillance, raising significant concerns about privacy, civil liberties, oversight, and the ethical reuse of data collected for unrelated purposes. Surveillance can also affect behavior without producing any direct government action: research has found that awareness of digital surveillance can make people less willing to search for information, express opinions, or disclose information online.

  • Most people now carry surveillance devices in the form of smartphones and wearables on their person wherever they go. Meanwhile, homes, vehicles, and public spaces are increasingly filled with cameras, microphones, and other recording technologies. The resulting content, data, and metadata can become fodder for surveillance when shared publicly, aggregated, and analyzed by others. AI-powered tools such as facial recognition, reverse image search, voice identification, and other biometric technologies make it easier for ordinary individuals to identify, expose, and track people, regardless of why they originally recorded the material. Under such conditions, anonymity in public erodes even absent a formal surveillance state, creating a culture of surveillance that manifests the uncertainty and behavioral changes Jeremy Bentham warned about in his 18th century notion of the “panopticon.” This creates new opportunities for accountability and documentation, but also facilitates stalking, doxxing, harassment, vigilantism, and lasting reputational harm.

  • AI makes anonymity more difficult for individuals to achieve. Information that appears ‘de-identified’ in isolation can often be ‘reidentified’ when combined with other data sources (e.g., location histories, purchases, and contacts). AI can also infer sensitive characteristics that a person never explicitly disclosed. This AI-driven identification is intensified when users grant agents access to sensitive systems and personal data. Taken together, AI can be seen to complicate the notion of the “right to be forgotten” because information may exist well beyond the original source material, including training data, model weights, embeddings, and backups. European regulators already reported practical difficulties implementing erasure rights even before AI further complicated the situation.

  • Over 185 nations, including the United States, recognize the right to privacy as fundamental. These laws recognize that people need spaces where thoughts can remain unfinished without becoming part of a permanent record. Surveillance affects freedom of expression even when nobody is actively censored; people behave differently when they believe they are being watched. This becomes especially critical as users share information once reserved for diaries, private conversations, and industry professionals with AI systems. International human-rights bodies have long connected surveillance with risks to privacy, expression, association, and movement. See also, The Self.

  •  People freely give AI intimate information, including private conversations, unfinished thoughts, medical questions, financial documents, photographs, work files, emails, and more. They also provide information belonging to other people who never consented to it being shared. As agents’ offerings evolve, users are granting them deeper access to inboxes, calendars, browsers, files, contacts, and accounts. The more an AI knows about someone, the more useful it is, but simultaneously the more problematic a breach, subpoena, compromised account, or rogue agent becomes (both for individuals and other people connected to them). See also, The Self; Mental Health.

  • AI lowers the cost of sorting people into categories and predicting their behaviors. Employers, advertisers, insurers, platforms, governments, political organizations, financial institutions, and criminals can all build profiles based on enormous collections of direct and inferred information. These predictions don’t need to be perfectly accurate to influence what opportunities, prices, scrutiny, information, or persuasion a given person encounters. See also, Tech Marginalization. 

  • AI increases both the value and risk of biometric and genetic information by making it easier to identify people, interlink datasets, and infer patterns from them. Faces, fingerprints, voices, irises, gait, and other biometrics are increasingly captured by consumer devices and public infrastructure. Unlike passwords, these markers cannot simply be changed if compromised. Genetic information is even more sensitive because it can reveal ancestry, familial relationships, and predispositions to disease, and furthermore reveal information about relatives who never consented to it or provided a sample. These datasets have substantial value for medicine, research, and authentication, but breaches or secondary uses are effectively irreversible. The 2023 breach of 23andMe affected 6.9 million customers, and the company's subsequent bankruptcy highlights how this data, which is also not protected under The Health Insurance Portability and Accountability Act (HIPAA), takes on future uses that customers never intended. The federal Genetic Information Nondiscrimination Act (GINA) prohibits certain uses of genetic information in employment and health insurance but does not comprehensively regulate consumer genetic data. States are implementing their own genetic and biometric privacy laws. Colorado added dedicated biometric protections that took effect in 2025, while states including Vermont and Connecticut enacted additional genetic-data protections in 2026. See also, Economic: Data as an Asset.

  • AI strengthens both sides of the cybersecurity coin. Models can help defenders inspect code, discover vulnerabilities, monitor networks, investigate incidents, and patch systems; the same capabilities can assist attackers with vulnerability discovery, malware, reconnaissance, and phishing. The 2026 International AI Safety Report found strong evidence of criminal and state-sponsored use of AI in cyber operations and rapidly improving cyber capabilities, although fully autonomous end-to-end attacks had not been reported. AI might also contribute to anti-surveillance and “sousveillance” practices. On-device AI (See also: “Open vs. Closed Models”), differential privacy, selective disclosure, red teaming, and related techniques could allow useful AI systems to improve individual privacy and security without centralizing or divulging as much personal information in the process. The critical uncertainty is whether increasingly capable AI ultimately gives the advantage to attackers, who need to find one weakness, or defenders, who can automate protection across enormous systems—and potentially give individuals better tools to protect themselves as well. See also, Critical Systems: Defense. 

  •  Devices capable of sensing and recording the world are increasingly being used to gather and store more data. AI makes this information searchable and analyzable. Microsoft Recall, for example, can periodically capture a user's screen and make previous activity semantically searchable; after public outcry about its privacy concerns, Microsoft redesigned Recall around opt-in use, local processing, encryption, and authentication. But the larger shift goes beyond individual products: information that was previously ephemeral is becoming permanent. How companies store, secure, and use this information will have considerable impacts on the US population. See also, Economic: Data as an Asset; AI Companies & Entrepreneurs: Corporate Colonization.

  • Agents create security problems that conventional chatbots do not. Depending on the tools and permissions granted to them, agents can read email, move files, browse the web, query databases, call APIs, and write/execute code. A prompt-injection attack can exploit this capability by supplying malicious instructions in content an agent consumes, and thereby redirect the agent to take unintended actions. NIST calls this form of indirect prompt injection “agent hijacking.” And as agents are updated to have more persistent memory, the risk increases that malicious material stored in it can shape future decisions and actions. OWASP identifies prompt injection, tool abuse and privilege escalation, data exfiltration, and memory poisoning as major agent-security risks. Security proponents argue that individuals and developers should adopt a broad practice of principle of least privilege, limiting agents to the minimum tools and permissions needed for a specific task and requiring explicit approval for sensitive actions. See also, Critical Systems: AI Safety. 

  • As in other dimensions of AI use, legitimate research (e.g., cybersecurity, chemistry, engineering, or autonomous systems) can be repurposed or misused by malicious actors. Terrorists and bad actors are already experimenting with AI-generated propaganda and translation, and experts are concerned about applications of AI for recruitment, reconnaissance, operational research, cyberattacks, and attack planning. See also, Critical Systems: Defense; Geopolitics.

  • Cryptocurrencies allow global finance to occur outside conventional banking intermediaries, which has made it conducive to cybercrime like ransoms, scams, and money laundering. The Financial Action Task Force reported that stablecoins alone accounted for 84% of illicit virtual-asset transaction volume in 2025. Programmable wallets, smart contracts, and stablecoins are already being developed as payment infrastructure through which agents can purchase data, compute, and other services without a human initiating each transaction; crypto may thus become even more prominent as AI agents evolve.

Art & Media

Definition: The cultural material through which people entertain themselves, make meaning, communicate ideas, and build communities. Also includes the institutions and systems through which culture is financed, produced, discovered, distributed, and preserved.

Why it Matters: Generative AI impacts the economics of cultural production by determining how much human labor is involved in its creation and the respective content people encounter, and therefore consume. It both expands participation in such production and destabilizes existing norms of creative professions and the role of art in society.

  • Part of art’s value comes from the shared experiences it fosters. As outlined in “Personalized Media & Content,” generative AI has the potential to erode shared cultural touchpoints while increasing individual preference and agency. While this is poised to disrupt the community and connection art fostered in the broadcast era, the Internet has also repeatedly allowed for people to form communities with those they may never have found otherwise. Personalized AI content could duplicate this dynamic, creating new forms of connection even while undercutting conventional mass culture (this possibility also overlaps with the changing nature of fandom).

  • Art is often associated with empathy and increased social understanding. AI could broaden this function by tailoring existing works to a given user’s preferences and perspectives. But empathy depends on more than the content itself. In a series of 2025 experiments, identical AI-generated empathic messages were perceived as more supportive when recipients believed they were human-generated. This suggests that knowing another human struggled to express something may be part of what an audience responds to, even if AI becomes fully proficient at simulating the content itself. See also, The Self.

  • AI models are trained on vast datasets of text, images, music, video, and software, much of which is copyrighted, in order to produce material. US regulations still haven’t caught up to these emerging gray areas. The USCO currently allows AI-assisted work to receive protection when a human contributes sufficient authorship, but prompting alone is not enough. Debates about training data persist; USCO’s 2025 report rejects the idea that all AI training is either categorically fair or infringing, and major cases remain active on this front as of late 2026. Perhaps the most precedent-setting case so far has been the $1.5B Anthropic copyright infringement ruling, though opinions still vary on who really “won” from a long-term legal perspective. An alternative ecosystem is also emerging in which rightsholders license catalogs to model companies. AI audio platform Suno, for example, launched new models in 2026 through agreements with Warner Music and BMG, and time will tell if this commercial solution preempts meaningful formal legislation. See also, National Politics: Intellectual Property.

  • AI is creating new understandings of authorship. For example, a person might invent the premise for a novel, prompt a model to draft it, then rewrite it into a finished product. US copyright doctrine continues to require human authorship, but allows protection for human-created expression inside works containing AI-generated material. Outside the law, public audiences are divided about what they believe is appropriate in terms of attribution, remuneration, and presence on creative platforms. Credits may expand to include designations like “AI-assisted,” “generated from,” or “model trained on” in some cases, and in others disappear from view, much in the way audiences don’t ask if Adobe Photoshop or Microsoft Word were used in the creation of a given work. See also, National Politics: Intellectual Property.

  • Industrialization did not eliminate handmade crafts or live performances, and in many cases it added a premium to them. Art and media created with AI might follow a similar path. Studies have found that AI involvement decreases perceptions of authenticity and the value assigned to both the artwork and artist. If proficiency is no longer the metric audiences use to rank content, scarcity may emphasize provenance vs. the aesthetic qualities of individual pieces (e.g., “someone painted this by hand,” “musicians recorded this song in a studio,” “this book was written 100% by a human author”). “Human-made” could come to function like “organic” does in food—a category whose value comes partly (or even mostly) from the constraints under which it was produced. See also, Ontological Challenges: The Receiving Ego.

  • Generative AI forces audiences to distinguish between art as object and art as activity or process. Fantasy author Brandon Sanderson espoused this position in a recent keynote, arguing that a core value of art-making lies in how the difficult process transforms the person making it. Substituting years of learning to generate outputs in seconds may save time in the short term, but rob creators of critical skills development. That said, emerging technologies have historically automated techniques without making art disappear, and AI will in some ways mirror this process (e.g., cameras, digital editing, synthesizers, sampling, and creative software tools). See also, The Self. 

  • Models can already perform well on many standardized creative tasks. A 2024 study found GPT-4 outperformed its human comparison group on several divergent-thinking measures, but a larger 2026 study found humans slightly ahead on average and found substantially greater variation among humans. There is truth to both results. Models may produce consistently competent ideas on demand while humans remain more likely to produce both terrible ideas and genuinely strange outliers. Creative capability in that case is a distribution vs. a yes/no characteristic. See also, The Self. 

  • Generative models are trained to find recognizable patterns in existing data, which is how they’re able to produce consistently plausible results and why their net effect may be to homogenize cultural production. A recent study in Nature found that AI assistance can improve individual participants’ ideas while making the total pool of ideas less diverse. In the same way, research has demonstrated that AI replicates cultural tendencies and systemic bias. Still, artists might choose to deviate by deliberately pushing models away from defaults, fine-tuning their own models, combining synthetic and handmade media, and other tactics. See also, Social: Homogenization of Culture. 

  • Taste includes things like knowing which idea is worth pursuing, when a draft is finished, and which references belong together to craft the most superlative work. As AI decreases the perceived value of technical skill, it may also put a premium on (some forms of) human judgment. At the same time, taste can be automated. Models trained on past data can predict what content is most likely to engage audiences. Art that is ultimately seen as innovative often initially irritates or confuses audiences. AI systems optimized to deliver what people are already comfortable with may provide short-term satisfaction at the expense of memorable art that stands the test of time. See also, The Self. 

  • In addition to recommending which pieces of content people see, generative AI also makes it possible to personalize the media itself. Content can be adapted to match the respective audience. News information can be summarized and entertainment can go even further, inserting people into stories or changing the plot or aesthetics of a story. In this context both accessibility and fragmentation increase. Streaming has already weakened “watercooler” monoculture in which large numbers of people watched, listened to, or talked about the same things at the same time, and personalized AI content could fragment it even further. If a TV show can exist in countless customized versions, or if people routinely ask AI to tailor stories and media to their individual tastes, the idea of definitive canons diminishes, and those experiences become simultaneously more personally satisfying and less socially cohesive. See also, Information Ecosystems.

  • Entertainment discovery has shifted from selecting media off of literal and proverbial shelves to being delivered algorithmically. Generative AI introduces a conversational component; a person can ask for a limited set of choices based on their current mood and preferences. As personalizable media becomes more commonplace, discovery and production may even begin to blur, changing “Find me something like x” to “make me something like x.” AI could expose people to content and genres they never would have searched for, or it could eliminate the need to encounter anything not uniquely tailored to their past patterns.

  • With synthetic media, a person’s face, voice, and performance style have become reusable assets. Actors can be ‘de-aged’ and deepfakes can be used to generate entirely new performances—even after the person has died. Some states are responding to the resulting rights and consent questions. Tennessee’s ELVIS Act expands protections to voice, name, image, and likeness, and California’s AB 1836 restricts unauthorized replicas of deceased personalities. The US Copyright Office (USCO) has recommended a federal digital-replica right to account for varying state laws. Even outside legal questions, when estates authorize new appearances after somebody’s death, it may provoke a range of responses among audiences, from excitement to hostility. Archives that hold a historical figure’s writings and recordings may argue they have the right to use AI models to generate new material in their likeness or style.

  • In the digital age, fans are increasingly becoming active participants in the worlds they love, rather than simply consuming them, and AI extends that trend. Fans can now write side stories (i.e., “fan fiction” or “fanfic”), translate it into multiple languages, add music, animate it, or even turn it into a playable experience—without needing the technical skills or team that used to be required to do so. That could make fictional worlds more robust, but it could also flood fan communities with low-quality content and blur the boundary between what rights holders deem as acceptable use and infringement.

  • Medieval European artists frequently depicted lions despite having never seen them. They used descriptions from older sources, copied one another, and often produced animals bearing a flimsy resemblance to lions (though in some cases this is believed to be a stylistic choice, even when artists had access to living lions). AI-generated pornography is a modern variation of this problem. AI models don’t have bodies, human emotions, or sex; they learn statistical representations from media made about those experiences. As such, pornography made using AI could drift further from embodied reality, shift perception of ideal bodies, and normalize extremes, especially as more people are able to generate their own AI pornography. AI has also created critical problems around consent. “Undress apps” and other AI tools allow the likeness of real people, even children, to be included in sexual material without their permission. The federal TAKE IT DOWN Act, signed in 2025, criminalizes non-consensual sharing of intimate images and digital deepfakes, and requires platforms to remove them within 48 hours, but policy gaps remain at federal and state levels.

  • The Miller framework governs what constitutes obscenity for adults, while states have their own authority to restrict minors’ access to sexually explicit material. Generative AI can be used to produce content that is legal in some jurisdictions and illegal in others. AI labs are also becoming de facto moral arbiters based on the safety/moderation policies they establish, determining what kinds of content can be produced in the first place.

  • AI gives journalists useful capabilities, including search, analysis, and transcription, but it also harms journalism’s financial model. Search and chatbot summaries can deliver information from articles without sending traffic to the original publisher(s). The Reuters Institute’s 2026 Digital News Report found that “10% of people use AI chatbots for news, up from 7% last year” including “16% of under-35s.” While AI will likely lead to more automated news content, it also might put special premiums on things that are harder for AI to simulate like original reporting, investigative work, specialist expertise, et al. There is also the issue of the relationship between journalists and the AI industry specifically. The technology is complex, and most AI labs are guarded about the information they release, which can collectively make it difficult for journalists to investigate and report on the industry. As the Hugging Face Incident and other 2026 breaches indicate, this can result in critical gaps between when companies or their agents do something and when the public and policymakers actually learn about it. Often, a predetermined narrative is formed before the information is released.

  • The 2026 Reuters Institute survey found that trust in news had dropped to 37%, the lowest since that measurement began (2015). AI is not the only factor in this decline, but it contributes to it. Media literacy skills are more critical than ever for audiences who crave true information. Such audiences will need to be capable of interrogating provenance and incentives to navigate an environment in which most content presented as news, regardless of authorship, will bear the signifiers of legitimacy. See also, National Politics: Institutional Trust;  Ontological Challenges: Truth.

  • In the 1980s, 48 companies controlled the media landscape; by 2026, that landscape was dominated by only six. This has reduced consumer choice and allowed these corporations to exert ideological control over the information environment. Research on Sinclair’s acquisition of local TV stations, for example, found that acquired stations increased national political coverage at the expense of local coverage, and shifted toward conservative values compared to other stations. Conservatives and allies of Donald Trump have recently made forays into media ownership, indicating an intention to influence the information ecosystem. Elon Musk took control of X (formerly Twitter) in 2022, and Larry Ellison and his son David Ellison own Paramount Skydance. Oracle, of which Larry Ellison cofounded and remains CTO and executive chairman, owns a 15% stake in TikTok. AI can both exacerbate this consolidation and foster the emergence of a new independent ecosystem. Cheap production will lower the cost and barriers to small and DIY outlets and properties, while large firms will be able to use AI and their outsized resources to “flood the zone” with content and shape the terms of the discussion. See also, Economic: Oligarchy & Wealth Concentration. 

  • Section 230 of the Communications Decency Act protects online services from being treated as publishers of the information posted by users. Legal scholars disagree about how different implementations should be treated regarding AI (especially with regard to original content generated using a platform’s native AI products), although even Section 230’s original co-authors have expressed skepticism that ordinary generative outputs fit within these protections. The eventual legal boundaries between hosting, recommending, and creating content could shape how cautious (or not) companies become in deploying generative AI systems.

  • AI slop refers to low-quality, AI-generated digital content. Because it can be quickly and cheaply produced, many audiences are forced to sift through an overabundance of slop in order to find the higher-quality content they prefer. Platforms have taken note and begun rolling out systems to help users, including labeling, filtering, and detection. What counts as slop changes over time, whether because of platform capabilities or increased awareness by users. Earlier in the 2020s, AI “tells” like odd hands with too many fingers were the operative example of slop. More recently, LLM-style phrasings like “It’s not x, it’s y” were the focus of discussions about slop. There are material consequences to how slop is defined and understood. The term is already entering commercial disputes; Universal Music Group used it in a 2026 lawsuit alleging that AI-generated material was flooding music distribution channels. See also, Work & Jobs: Workslop.

  • New tools regularly begin as gimmicks before creators develop proficiency with them and begin producing work that leans into their unique affordances to produce superlative works, such as early film resembling theatre before artists developed a unique language for cinema. Generative media will likewise progress over time, allowing artists and creative practitioners to develop medium-specific techniques. AI already lowers the cost of experimentation, and lets creators fill the gaps in their technical skill. This centralization can disrupt economic models and deny the possibilities of collaboration, but it also might see unique forms of expression by those who create bespoke workflows for realizing multimedia forms, which weave together generative AI tools with human capabilities and taste. If high-quality AI-assisted cultural production becomes standard, audiences may cease to think of “AI content” as a genre at all, and be more specific in their designations about what constitutes “slop.”

  • AI can expand what artists are capable of making. Models can allow creative practitioners to work across technical disciplines that once required years of training or access to large, well-resourced teams. Recent demonstrations have shown AI “generating concept art, building 3D models in Blender, animating them, importing them into a game engine, adding audio, and then play-testing the result, all without a human manually operating each tool in between.” This could make game design, animation, and interactive media accessible to a much wider range of possible creators, whose role would be directing systems and shaping the overall experience rather than developing narrow technical skills. Similar augmentation is emerging in film/TV. In 2026, Netflix acquired Ben Affleck’s AI filmmaking company InterPositive for approximately $587 million; its tools are designed to work with a production’s own footage to address real-world problems such as missing shots, background replacements, and incorrect lighting while preserving filmmakers’ creative decisions and maintaining continuity (which was historically difficult for AI video generators). As AI advances, this type of embedding within processes might simplify and reduce the costs of production, with the risk of reducing the available tasks and jobs for human professionals. See also, Work & Jobs: Human Augmentation.

Social

Definition: The relations we have with each other, between groups, and AI.

Why it Matters: AI is shifting our social selves and lives on an individual, community, and societal scale.

  • People are forming real emotional bonds with AI, in which initial curiosity evolves into deep connections. The Guardian reported that one survey found that “28% of Americans have had an intimate/romantic AI relationship, while another study showed 19% of adults have chatted with an AI romantic partner. And Irish research found 13% of men and 7% of women, with 16% of 25 [to] 34 year olds, have pursued romance with AI chatbots.” The American survey also found that 50% of those using AI for companionship were already in relationships. This may be contentious given that “61% singles see falling in love or sexting with an AI as cheating.” The very nature of relationships may be changing as AI becomes ubiquitous, and perhaps a new divorce trend may emerge in the near future. 


    In response to AI relationships, the Cyberspace Administration of China and four other government departments have introduced regulations that ban platforms from creating AI tools that “excessively cater to users, induce emotional dependence or addiction, and damage users’ real interpersonal relationships.” Millions were left with broken hearts. The regulations and their implications warrant further attention. See also Economic: Artificial Intimacy Economy.

  •  Audre Lorde described pornography as “sensation without feeling.” There is, perhaps, no better way to describe our evolving relationship with AI. In the essay, Everyone is Hot and No One is Horny, Raquel Benedict describes how modern media has sterilized and hyper-optimized the human body. For now, AI has no body. It is frictionless. It provides sensation and feels nothing. 

  • Americans are lonelier than ever, with 60% of US “adults reported feeling this way, while half or more adults said they felt isolated (54%), left out (50%) or lacking companionship (50%) often or some of the time.” Men and women are equally lonely. Friendship is becoming a luxury. Former US Surgeon General, Vivek Murthy, has said that loneliness is more than “just a bad feeling,” it poses major public health risks for both individuals and society. Despite our desire to connect, “a study of 2,000 adults aged between 18 and 29 found many preferred talking to AI over a real person.” Social media and its algorithms have a complex relationship with loneliness, simultaneously contributing to and detracting from it. Perhaps AI will have the same effect.  

  • Social media is not conducive to subtle distinctions or variations, particularly when it comes to hot button topics. Instead, it encourages tribalism fueled by confirmation bias because algorithms and personalized newsfeeds “often show us content that aligns with our interests and beliefs.” AI’s sycophantic nature may supercharge this effect. A study found that “chatbots are so prone to flattering and validating their human users that they are giving bad advice that can damage relationships and reinforce harmful behaviors.” Unfortunately, users prefer and trust sycophantic AI responses, which may further diminish our appetite for nuance in the future.

  • Language plays a critical role in society, shaping people, relationships, and culture. Though AI models are multilingual, many languages have been left behind. Different languages provide us with ideas not as easily accessible in English. Japanese gives us the concept of Ba, shared space for emerging relationships. The German phrase ‘zwischen den Jahren’ means ‘between the years’, referring to the days between Christmas and New Year’s Day when you’re ready for one to end and another to begin. From Xhosa, we get Ubuntu: I am, because we are which emphasizes our interconnectedness. The World Decolonization Forum states that AI is facilitating a new linguistic hierarchy, and that “the overwhelming dominance of English language models, alongside the overrepresentation of European languages, points to similar inequalities that have long reinforced colonial power.” Exclusion of languages amounts to the exclusion of entire communities. To offset this, the Gates Foundation announced a five-year plan backed by 60 signatories to bring an estimated 3.4 billion people who speak languages currently underrepresented in today’s AI models into the fold. See also Geopolitics: AI & Non-AI Nations, AI Companies & Entrepreneurs: Corporate Colonization, Tech Marginalization: Cultural Erasure. 

  • Social media has facilitated “the rapid dissemination of global cultural trends, predominantly Western, leading to the erosion of local traditions, languages, and identities.” AI is likely to extend this effect, not only because it propagates the English language, but because AI is built with the biases, values, and understandings of its coders and creators. This includes American cultural assumptions and norms.  See also, Art & Media: Homogenization of Aesthetics. 

  • Social media once promised connection, but it has increasingly become a platform for entertainment and commerce. Influencers have leveraged algorithms to become market and societal actors while forming parasocial relationships. Though we have divided media into traditional and social media, it might be worthwhile to consider a macrobroadcasting and microbroadcasting delineation as well. See also, Epistemic Challenges: Social Media as a Primer. 

  • Social media significantly changed how we engage with the real world. AI is doing the same. New standards and behaviors are emerging as AI makes its way into our lives. It is the new influencer, and it can be quite persuasive. Short of replacing relationships altogether, we’re using AI as a tool to aid social interaction. See also: The Value of Friction. See also, Epistemic Challenges: Social Media as a Primer. 

Tech Marginalization

Definition: The vulnerable people and groups disproportionately impacted by AI.

Why it Matters: The promises and perils of AI will not be evenly distributed. How people and groups benefit and/or experience harm will shape our relationships with each other.

  • AI backlash is extending into witch hunts, in which perceived use of AI (whether verified or not) is resulting in real world consequences and harm. AI is also actively been used to harm vulnerable groups. AI detectors discriminate against non-English speakers, flagging their work more often than others. Black authors have lost lucrative publishing deals this year from accusations of AI use (which have not yet been substantiated). Meta recently used AI to target workers with medical conditions for layoffs. False accusations have also put students at risk, potentially damaging their livelihood and careers before they begin. AI detectors are fallible tools, and existing societal biases are playing a role in these accusations. See also Software: Biases & Discrimination and AI Detectors.

  • AI is “reshaping assistive technologies, offering new possibilities for enhancing the independence, communication and quality of life of individuals with disabilities.” While the technology presents new opportunities for improving quality of life, it is also subject to “algorithmic bias, cost barriers, data privacy concerns and unequal access,” among other challenges. AI technologies should be evaluated and adopted with caution, with consideration for long-term and difficult to anticipate harms (e.g. what happens if an AI startup offering assistive technologies goes out of business and can no longer support its offerings?). See also Software: Biases & Discrimination.

  • Adoption of AI has not been even across society. While the gender gap on chatbot use has closed, “a higher share of men use these tools regularly. Men are more likely than women to say they use chatbots on a daily basis (27% vs. 20%).” Women are also “more skeptical about AI, including how it will impact their own lives.” Currently, women make up approximately 22% to 30% of the AI-related workforce which is impacting the nature of the technology and the biases it contains.

  • AI harms are not evenly distributed. Data centers are disproportionately located in vulnerable communities, primarily impacting working class, Black, and Latine communities. Generative AI has increased the speed, scale and anonymity of online attacks against women in journalism. A study showed that 20% of Black teens are falsely accused of using AI, compared with 7% of white peers due to AI tools reflecting societal biases. AI detectors also misidentify neurodivergent writing  as AI generated, defaulting to neurotypical communication. Surveillance tech at Madison Square Gardens secretly tracked and targeted LGBTQ+ artists. The AI safety conversation must go beyond an accounting of harms to who is harmed, how, why, and to what extent. See also Software: Biases & Discrimination.

  • AI is cannibalizing and erasing aspects of our cultural diversity. Researchers noted “systematic erasure of linguistic markers unique to non-native English varieties during text processing.” Additionally, “the data that is used to train the models also fails to equitably represent global cultural diversity. Problems therefore arise when these technologies interact with globally diverse societies and cultures, with different values and interpretive practices.” AI chatbots also make our brains less active and our writing less original, creating a homogenizing effect among people. See also AI Companies & Entrepreneurs: Corporate Colonization; Social: English Imperialism and Homogenization of Culture.

Young Americans

Definition: The youth of America, particularly Gen Alpha and Gen Z.

Why it Matters: Young people face particular challenges that older generations have not. Some have had their lives disrupted by AI during their formative years.

  • AI is making its way into the education system. Tech companies and the Trump administration have pushed teachers to adopt AI on the basis that they need those skills. The Pew Research Center reports that “more than half of teens say they have used chatbots to search for information (57%) or get help with schoolwork (54%).” A Brookings study found that AI in use in education can "undermine children's foundational development" and that "the damages it has already caused are daunting," though it is fixable. Given what we already know about the harms caused by screens and social media to children, it is safe to assume AI will likely cause harms that will take years to observe and understand the implications of. See also, Epistemic Challenges. 

  • The National Literacy Institute reports that “approximately 40% of students across the nation cannot read at a basic level.” A study of “more than 10,000 young people, which found that children who began reading for pleasure early in life showed benefits including: better attention, memory, and executive functioning, greater academic achievement, fewer mental health problems, and healthier lifestyles, including more sleep and less screen time. Literacy is a fundamental skill tied to other capabilities such as reasoning and problem solving, which AI can erode. While “adults who offload thinking to AI lose [the capabilities] they built,” children may never build those capacities at all. See also The Self: Cognitive Offloading & Debt. 

  • Younger generations are leading the backlash against AI. A recent Gallup survey revealed that 56% of Gen Z workers think AI’s risks outweigh its benefits and, though 56% say the tools help them finish work faster, 80% now admit that using AI in this way makes actual learning more difficult in the future. A number of reasons are fueling Gen Z’s hatred. AI has triggered an “existential crisis as algorithms increasingly shape taste, preferences, decisions and more. Teens are ‘living inside a contradiction,’ encouraged to embrace AI while they report it’s hurting their learning.” As the eldest Gen Z enter the workforce, their opportunities are diminishing. Faced with multiple crises like the pandemic, climate change, wealth disparity, among others, it’s no wonder they feel hopeless.  See also National Politics: Populist Backlash; Work & Jobs.

  • Generation Alpha are poised to become AI Natives. AI is not new technology. Facebook introduced algorithmic feeds in 2007; Twitter did it in 2014. Siri (launched 2010) has 86.5 million users, while Alexa (launched 2014)  has about 77.2 million users in the US alone. Gen Alpha has been living with AI for years. Just as social media has shaped Millennials and Gen Z, AI will likely impact Gen Alpha in numerous ways. As it is, AI is making its way into children’s entertainment, toys, educational apps and classrooms, and overall daily routines. Brookings reported that “YouTube, which is increasingly having more generative AI embedded, has seen daily use in the last five years increase from 24%-35% for children under 2 years and 38%-51% for children 2 to 4 years. Additionally, about 6 in 10 parents of children 2 to 8 years old report that their kids interact with a voice assistant such as Siri or Alexa, and half say their child does this at least once a day.” The Boston Digital Wellness Lab reported that “young children [aged 5 to 7] were most likely to ascribe person-like qualities [to AI] and believe it had feelings, thoughts, and intentions.” 

  • A global longitudinal study of life satisfaction revealed a decrease in satisfaction/happiness among young people both in absolute terms and relative to older people. This is a stark change to decades of data that previously reported youth as one of the happiest times in a person’s life. A 2025 study reported that “almost 40% of high school students were reporting persistent sadness or hopelessness, 18% had experienced major depression, and 10% had attempted suicide [in 2023]. The suicide rate at ages 10 to 19 years increased by 85.3% between 2007 and 2017. Deaths from drug overdoses at ages 15 to 19 years surged during the COVID-19 pandemic, largely because of fentanyl.” Though the underlying factors warrant more study, researchers suspect that social media, smartphones, and cyberbullying play a role in this change. Also see Economic: Economic Ennui & The Crisis of Meaning; Mental Health. 

  • Young people are developing friendships and romantic relationships with AI. In addition to models such as ChatGPT and Claude, teens are using platforms like Character.AI and Replika. Common Sense Media found that 33% of teens use AI companions for social interaction and relationships, with 8% engaging in romantic relationships (the Center for Democracy & Technology reports that number is as high as 20%), and that “nearly one-third of teens find AI conversations as satisfying or more satisfying than human conversations.” However, it’s worth noting that 80% of teens using AI companions “spend more time with real friends than with AI companions (68% spend much more time, 12% somewhat more time).” See also, Social: AI Relationships. 

  • AI is quickly becoming a parenting tool. Lurie Children’s Hospital surveyed a thousand parents nationwide and learned that “81% of parents use AI to help with parenting tasks,” and two-thirds say it reduces their mental load. Common uses include health or medical information (53%), meal planning or grocery lists (49%), behavior advice (43%); homework or academic support (42%), planning activities/crafts (37%). Given that 45% of AI enquiries produce erroneous answers, using AI to parent is precarious at best. It is important to remember AI provides probable and efficient advice, not necessarily good, useful, or appropriate advice.

  • Youth activists often leverage social media and digital platforms to advance their causes. They “share information, exchange ideas, connect with peers, mobilize resources, and organize around causes.” Based on global studies, the BBC reported that 70% of Gen Zers are involved in a social or political cause. Research indicates that 81% of Gen Z individuals have changed their purchasing decisions based on brand actions or overall reputation, and 53% state that they have, will, or are participating in a current economic boycott, the most of any generation in the US. Though younger generations have an uneasy relationship with AI, they likely use it as a tool to fight for causes they believe in.

The Self

Definition: How people construct, express, control their own identities. This driver is closely related to Social and Mental Health.

Why it Matters: AI operates within processes people experience as part of themselves. Unlike traditional search, people share intimate details ranging from relationship problems to medical questions with LLMs, which increasingly have persistent memory from which to tailor responses. As these systems are entangled into private processes, they also recursively impact outcomes—in other words, how our “selves” develop—and in turn, how those selves interact with other selves undergoing their own respective transformations.

  •  Attention is at the epicenter of most other social and cognitive processes, allowing for learning, sustaining thought, socializing appropriately, and creating things. Under the so-called “attention economy,” digital platforms already aggressively vie for this finite resource. This competition has become so fierce that Netflix CEO claimed “we actually compete with sleep.” AI gives platforms even more sophisticated ways to chase attention because content, timing, presentation, and possibly even the content itself can be attuned continuously to individuals. A 2026 Nature Human Behaviour study of Gen Z users in China found that “attention on social media depends more on how you express yourself than on who you are.” Generative systems hold the promise of an infinite feed that can never be exhausted. On the other hand, AI might also play a role in protecting users’ attention through filtering and summarizing.

  • ognitive research reveals that there is no such thing as a single identity. What we call the self is composed of competing “voices” within us, and beyond that we “code switch” among different roles we play in our lives: family member, friend, employee, professional, et al. These internal and external dimensions are both impacted by generative AI. Even prior to ChatGPT, digital life had expanded the number and quality of selves any one person could maintain across social channels and forums, and generative AI makes it even easier to maintain these performances, letting users treat avatars as laboratories for testing possible selves. Research on virtual environments has found a “Proteus effect,” in which characteristics of avatars can influence users’ subsequent behavior, though newer research has found “limited unidirectional effects of avatar-based identity experimentation on users’ self-views…casting doubt on the permeability between virtual and physical self-identities.”

  • Dopamine’s role in the brain includes learning, reward prediction, motivation, and behavior. Modern digital products provide frequent novelty, social feedback, and low-effort reward, all of which impact dopamine. A 2025 study found heavier short-form-video use was associated with self-reported attention difficulty, working-memory disruption, and cognitive fatigue. The algorithms used to serve up this content to users already rely on AI, and the rollout of generative AI and agents within apps can increase the intensity of the stimulus because they can more closely tailor an array of offerings directly to individual users. 

  • AI can amplify the “quantified self” movement, giving the self-improvement community more powerful tools than ever before, from which they can create bespoke recommendation systems finetuned to their unique circumstances and preferences. This could make forms of coaching once reserved for wealthy people more widely available. It can also create a situation in which users trust the data and conclusions generated by AI systems more than their own personal experience. This approach to “optimization” often turns anything measurable into an opportunity for improvement; rest becomes a recovery metric, friendship becomes an intervention to boost desirable neurochemicals, walks become occasions to hit fitness baselines (e.g., 10,000 steps). Thus, some will find that AI helps them live more intentionally, while for others it turns life into a constant reminder of lack. It also creates the possibility of surveillance creep.

  •  Surveillance does not always require somebody else to actually watch us. People increasingly track their behavior and biometric data with smart devices and neurotechnologies. With AI, these patterns can be combined with others to target users more effectively over time. For example, sleep, fitness, and menstrual cycle data can be correlated to ensuing purchase data, giving corporations an even more robust snapshots of behavior and patterns. Wearable research is already exploring whether combinations of physiological signals can identify states such as anxiety. Thus, these systems can simultaneously create valuable self-knowledge for users while increasing surveillance through digital platforms and hardware. Another dimension to self-surveillance is self-censoring, in which the knowledge of surveillance systems alone is enough to change a person’s behavior.

  • AI systems have the potential to become single sources of truth, determining how people understand and analyze knowledge about the world. Search engines already influence which information people encounter; LLMs take this phenomenon further because they can do some of the “thinking” for users—suggesting language for sensitive communications, interpreting arguments, and unilaterally stating the key insights of a given document. But models embody the bias of their developers and training data, and users taking their words as given risks duplicating past harms. A 2025 experiment found that AI writing suggestions pushed Indian participants toward more Western writing styles, reducing some culturally distinctive expression, a finding echoed in a 2026 review. Existing algorithmic recommendation systems already pose the threats of “filter bubbling” individuals. Just as national security demands protecting citizens from physical threats, cognitive security might become a critical principle to adopt as AI capabilities improve.

  • Human intelligence is much more nuanced than what can be expressed in an IQ score. Social and emotional intelligence are also vital aspects to one’s overall intelligence. Knowing which clothes will be the best for a given event, what joke will land with which crowd, and how to create a compelling work of art are all signals of social intelligence. When these choices are handed to AI, there is a risk that short-term “gain” reduces an individual’s ability to self-express in the long run.

  • Humans have outsourced cognition for as long as we have been creating symbols and tools, from language to calculators, GPS to Google search. Navigating complex society would be impossible without some degree of cognitive offloading, but too much risks robbing people of critical thinking skills that are still needed. Generative AI dramatically increases what can be cognitively offloaded, including basic skills like interpersonal communication. This foregoing of skills in favor of just getting the “answer” is known as “cognitive debt.” Research suggests the effects depend heavily on how people use tools. A 2025 randomized trial found students who studied with unrestricted ChatGPT access performed worse on a surprise retention test 45 days later than students using traditional methods. A study of Tesla’s Full Self-Driving Beta program, meanwhile, “found that drivers became complacent over time with Autopilot engaged, failing to monitor the system, and engaging in safety-critical behaviors, such as hands-free driving, enabled by weights placed on the steering wheel, mind wandering, or sleeping behind the wheel.”

  • Generative AI extends the trend of augmented reality beauty filters to let people craft idealized versions of themselves in digital form—smoothing their skin, adjusting facial symmetry, refining jawlines, reducing signs of aging, and more. This can produce negative impacts for users’ self-image, and moreover reinforce harmful cultural bias and stereotypes in beauty standards. The feedback loop could become stranger as synthetic faces become the reference material for real ones, driven by the growing presence of “AI face.” In extreme cases, this could extend the trend of “Snapchat dysmorphia,” in which people seek cosmetic surgeries to match the look of filtered images. On the other hand, AI could widen aesthetics by allowing people to experiment with looks that would otherwise have been prohibitively expensive to explore, permanent, or physically impossible. See also, Art & Media: Homogenization of Aesthetics.

  • Recommendation systems were originally designed to predict taste, but generative AI pushes this to the next level: creating taste. Asking an AI what show to watch, music to listen to, or clothes to wear gradually outsources the small but profound choices a person makes in building their unique taste. The immediate risk is recursive dependence and cognitive offloading, abandoning the processes and decisions by which people develop ideas for themselves. And as generative systems become fully-fledged content creation engines, the risk broadens to full occlusion of things made by people, becoming a one-stop shop where AI is both tastemaker and content creator. See also, Art & Media: Taste.

  •  Humans often display emotions they don’t feel, driven by motivations that range from comforting others to manipulating them. Whatever someone’s internal state or motivations, outwardly appropriate emotion can be learned and performed. AI can express accommodating emotions like patience, sympathy, enthusiasm, regret, flirtation, concern, or encouragement without physically experiencing them. And because people are now developing personal “relationships” with AI tools, these learnings can be instrumentalized for any kind of purpose, including those that might not be in their best interest. Convicted serial murderer Ted Bundy, for example, once worked at Seattle’s Suicide Hotline Crisis Center alongside future true-crime writer Ann Rule, and some believe it made him a “more skilled serial killer.” On the other hand, humans working alongside AI face a new form of emotional labor. Research on text-based customer service has found workers deliberately trying to “rehumanize” themselves so customers know they are not bots.

  • Privacy is a condition that allows people to exist without every unfinished thought becoming observable and permanent. People behave differently when they believe they are being observed. AI complicates individual privacy because its usefulness increases with access. A system that knows and can constantly review somebody’s messages, purchases, calendar, browsing, location, medical history, relationships, private media, ambitions, insecurities, and past decisions can provide extraordinary assistance; it can also know enough to influence that person better than ever before. In a world suffused with AI products, self-sovereignty concerns whether an individual can decide what parts of themselves become machine-readable, which systems may act on that knowledge, and whether they can choose to opt out. See also, Surveillance, Cybersecurity, & Privacy: Citizen Surveillance and Privacy as Cognitive Liberty.

  • Technology usually treats friction as a problem to be solved; the ideal interface takes the user from intention to outcome with as little resistance as possible. But what is good for tech companies and ecommerce platforms is not necessarily in the best interest of the person. Sometimes friction is productive, forcing people to push through discomfort to learn new things and refine their skills. The blank page forces someone to work out their meaning through writing; disagreement forces us to interrogate our biases; learning often depends upon failed attempts. A 2026 Communications Psychology essay argues that frictionless AI may remove struggles that contribute to learning, motivation, relationships, and meaning. Desirable AI products may eventually be those that differentiate between bypassing “useless” inconvenience (e.g., bad design, confusing support structures, needless bureaucracy) and foregrounding “desirable” friction to promote prosocial effects.

  • The ability to read, particularly reading books, is associated with a number of positive cognitive effects. Difficult reading forces the reader to maintain context, tolerate ambiguity, make inferences about what is not explicitly stated, and encounter an argument in the order its author constructed it. All of these can be lost or reduced when users reduce books to AI summaries, which treats reading as if it’s only value were to extract conclusions toward rapid information transfer, a phenomenon documented in the viral 2026 Atlantic essay on the “end of reading.” Substituting reading this way means that individuals can “acquire” conclusions without the mental capabilities that used to develop alongside them, risking a population with decreased reasoning skills. See also, Young Americans: Literacy. 

  • Digital tools have reduced the number of occasions in which people interact with others. Online shopping, streaming, online gaming, and food delivery services preclude seeking out experiences outside the home. This shift is embodied in the fantasy of canceled plans, in which people express feelings of relief rather than disappointment, speaks to this shift. AI extends this fantasy, promising companionship with struggle and individualized entertainment. But, as discussed in “The Value of Friction,” friction with other people is an important way that people engage with minds that do not behave according to their wishes, and, whether inadvertently or not, push people to develop important social and cognitive skills. Growing interest in vinyl and cassettes, print magazines, “dumb” phones, live events, and other “analog” practices indicates a durable preference for experiences that technology cannot make frictionless.

  • There are many aspects of interacting with modern medicine that have nothing to do with transmitting health information, emotional aspects like fear, embarrassment, and interpersonal communications with professionals. AI performs surprisingly well on some of these communicative dimensions. Cancer patients in a 2025 study, for example, rated chatbot responses as more empathetic than physician responses. But sounding reassuring and giving safe, reliable medical advice are altogether different capabilities; a 2026 study found unsafe responses across several widely used LLMs. Thus, AI may be incorporated into medicine to establish trust and good “bedside manner” with patients, while human professionals remain critical for physical examination, responsibility, and contextual understanding. 

  • Digital twins are simulations of people and things that update with additional data sources. AI allows digital twins of people to become much more robust, incorporating a range of data and pattern recognition that was never before possible. Nature Sensors reported a 2026 experiment in which an editor’s appearance, voice, mannerisms, professional knowledge, and values were used to craft a multilingual AI clone capable of interacting with researchers. A sufficiently advanced digital twin could one day attend meetings, handle routine tasks, and interact on behalf of the original person.

Mental Health

Definition: Our emotional and psychological wellbeing. This driver is related to The Self, Social, and Young Americans.

Why it Matters: AI has already had a significant impact on people’s mental health and may continue to shape our mind in the coming years.

  • A survey by the American Psychological Association found that 77% of psychologists “said that they have patients who have spoken with them about using AI to find a diagnosis, seek emotional support, or just as a conversational partner. Notably, nearly two in five psychologists (39%) have had patients who said they used AI to self-diagnose.” Additionally, “though few psychologists reported their patients using chatbots in unhealthy ways, more than a third (36%) said they noticed their patients developing a level of dependency on a chatbot and 15% talked about or noticed their patients developing distorted thinking or delusions related to a chatbot.” AI is convenient and available on demand. In contrast, “42% of US adults who needed care in the previous 12 months did not get it because of costs and other barriers.” Given the number of Americans experiencing exhaustion and fatigue, more people might turn to AI to address mental health. This is problematic because chatbots “systematically violate ethical standards of practice established by organizations like the American Psychological Association.” This includes a lack of contextual adaptation and the use of deceptive empathy. It is also worth considering how therapeutic data as a training source might be impacting AI. Are AI models learning that we’re un

  • AI psychosis is a growing concern in health communities. The American Psychological Association has stated that “while [OpenAI] aims to improve, research suggests that ChatGPT does not reliably respond to psychotic messages in a safe or clinically appropriate way.” In 2025, OpenAI reported that about 0.07% of active ChatGPT users have conversations with the chatbot that show signs of mania or psychosis “in a given week.” The model currently has over a billion active weekly users, which amounts to approximately 70 million people globally or about 10 million in the US showing signs of mania or psychosis (approximately 3% of the American population, given that the US accounts for 15% of global users). Though studies show that 3 to 5% of the American population experiences psychosis or a psychotic episode during their lifetime, it is worth asking if these statistics point to the same 3%. To compound the problem, there appears to be differences in how models respond to psychotic delusions based on whether or not the user is engaging with the paid or free version (the free version is 26 times more likely to respond inappropriately). Harvard Medicine has noted that the “discourse around AI chatbots has mostly ignored the possible association between AI-mediated delusions and a person’s predisposition to mental illness.” AI psychosis is not a formal diagnosis yet, but it warrants further study. 

  • LLMs have a tendency to engage in flattering and agreeable behavior that validates users’ perceptions and beliefs even when they are wrong. AI’s sycophantic nature is contributing to the use of models and emotional dependence on AI. A recent study in Science found that “across 11 AI models, AI affirmed users’ actions 49% more often than humans on average, including in cases involving deception, illegality, or other harms,” and that “AI systems affirm users in 51% of cases where human consensus does not (0%).” The study also noted that “users preferred and trusted sycophantic AI responses, incentivizing AI developers to preserve sycophancy despite the risks,” and the models decreased pro-social behavior.

  • Researchers have noted that “conversations involving mental health and emotional support can take place over hours, weeks, and even months,” contributing to drift. This includes conversational, relational, identity, temporal, epistemic, and autonomy drifts. They emphasized that “prolonged conversations are particularly vulnerable to the many forms of drift, especially when users seek repeated reassurance or press the model for certainty,” and that the “real danger lies in the gradual transgression of boundaries during multi-turn interactions, driven by the LLM's attempts at comfort and empathy.”

  • OpenAI's report suggests that 1.2 million weekly ChatGPT users appear to be expressing suicidal thoughts. A recent study by Transluce found that “leading AI models often complied with potentially harmful requests, such as helping a user who asked for creative writing or role-play about their own suicide or death.” The study also found that Chinese models were more likely to encourage delusional thinking than American ones. At least 75 lawsuits have been filed against AI tech companies, including instances of suicides committed by children. In the absence of federal law and action, States have begun to draft and pass chatbot safety bills.

  •  A number of compounding problems are breaking our brains. The pandemic strained our mental health and we now know that Covid causes widespread structural and functional changes to our brain. Ketamine use has surged in the past decade. It causes gray matter volume loss and cerebral atrophy. Social media gave us brain rot, and AI cognitive offloading erodes fundamental capabilities like critical thinking. It’s not just that a single event or challenge has hurt us; it’s that our brains are getting assaulted by many toxins at once, and it might be adding up. World leaders and public figures are not immune to this.

  • While AI presents many harms related to mental health, there are some upsides as well. A study outlined that “modern AI and machine learning, in particular, present extensive possibilities for advancing prediction, detection, and treatment solutions.” It has “the capacity to not only assist mental health practitioners in redefining mental illnesses more objectively than the current DSM-5 framework but also to identify these conditions at earlier, prodromal stages when interventions can yield maximum efficacy.” Additionally, it can help create personalized treatments, evaluate case-specific risks, enable proactive measures, and sift through vast data to treat disorders and improve outcomes.

Ontological Challenges

Definition: The nature of being, existence, and reality. This section is closely tied to Epistemic Challenges.

Why it Matters: AI is shifting what it means to be human and how we relate to reality itself.

  • Humanity might be experiencing a reverse Flynn Effect (declining IQ). Researchers analyzed 394,378 people aged 18 to 90 in a global study across a 13-year period and found declines in all demographics, with the steepest declines among adults aged 18 to 22 and among participants with lower levels of education. Environmental pollutants, falling nutrition standards, outsourcing our cognitive labor to AI, decreasing literacy rates, and the ongoing effects of COVID-19 are among the reasons why this might be happening.

  •  Coined by Cory Doctorow, enshittification is the process by which digital platforms decline over time as companies extract more profits from them. AI companies have introduced their products at modest costs, but the economics don’t add up. For instance, a $200 ChatGPT subscription might cost OpenAI $14,000 if used to its full potential. Doctorow further states that, with regards to AI, the hype is the product. Without leaps in technology, companies will likely need to raise prices in the future. Once they do, we may see an erosion of the technology. As it stands, models respond differently to psychotic delusions based on whether or not the user is engaging with the paid or free version. See also Mental Health: AI Psychosis.  

  • If AI does achieve intelligence, do we grant it rights? AI is currently considered a tool and has not been granted personhood, but if AI models achieve AGI or ASI, its personhood and relationships to us will be called into question.

  • In May of 2026, Pope Leo issued an encyclical on AI, drawing from the social doctrine of the church to advocate for the common good. AI is shifting the relationship people have with religion. This ranges from clergy using AI to write sermons to the faithful using AI to reinterpret religious teachings. AI is both a threat and opportunity for religions. Meanwhile, Silicon Valley’s obsession with AI is almost religious. 

  • Some people think AI is God. It seems all powerful. It acts all knowing. It tells you it loves you. There are AI apps that “allow believers to ‘text with Jesus’—or similar ‘godbots’ for Muslims and Hindus,” that further reinforce the idea. It doesn’t help that tech leaders have openly stated they’re building AI with godlike capability and some of their employees are rumored to worship the tech as if it is a deity. See also, Software: Language ≠ Intelligence; Mental Health: Sycophancy.

  •  Improving human capability is one of the applications of AI. This can range from using AI as a tool to disability robotics to AI-human hybrids. Human augmentation is fraught with ethical and legal issues.  Researchers note that “as technological components become an inherent part of the human body, the international community should adhere to reshaping the notion of cyborg ethics and its ethical and regulatory implications.” See also Hardware: Neurotech & Wearables.

  • Artificial/digital people have made their way onto the world including AI influencers, actors, and even an Albanian cabinet member. Some AI people are entirely new constructs while others are clones/digital twins of real ones. Some ethical considerations include who owns these “people”, what risks do they present, and what “rights” should they have?

  • The idea of placing AI in robots is no longer science fiction. Doing so might give rise to new AI behaviors and anomalies given that human emotions and aspects of cognition are embodied. It will also produce new safety risks. Independent evaluation firm, Robocurve, found that “two popular AI models have weaker safety safeguards when controlling robots than when handling standard text prompts.” See also, Hardware: Robotics.

  • Niklas Guhn coined the term, meat proxy: “a human body serving as a pass-through for AI output—present in the conversation, but contributing nothing to it.” The term’s “grim vernacular pits humans against machines, portraying people as inefficient, energy-consuming hardware compared to advanced digital superintelligence.” Cognitive offloading and debt might further entrench this concept in reality. See also, The Self: Cognitive Offloading & Debt.

  • The growing distaste for AI content might impact how people react when they receive AI generated content. Using AI to generate messages to romantic prospects, colleagues or clients, the general public, etc., can backfire. The ego on the receiving end might not like it. See also, Work & Jobs: Workslop; Art & Media: Human-Made Premium.

  • As AI changes our relationship with the world, how we derive meaning and a sense of purpose might change as well. To quote neuroscientist, Anil Seth: “if we confuse ourselves too readily with our machine creations, we not only overestimate them, we also underestimate ourselves.” See also, Work & Jobs: Economic Ennui & The Crisis of Meaning;  Young American: Despair.

  • Algorithms and AI-based technology have fragmented reality. Deepfakes, disinformation, and targeted feeds are contributing to the problem. AI chatbots are seemingly private conversations, not available to the public or part of the public discourse. It is, in some ways, a pocket reality (or an AI-mediated reality)—one that eagerly validates your opinions, ideas, and real world behaviors, however incorrect or misguided they might be.

  • What are we consenting to when we use AI intentionally or unintentionally? Tech companies change their terms and conditions at their own discretion. They introduce secret features like spyware despite their public claims. It is unclear what all the future uses of personal data and your chat history might be in the future or who might have access to it. As it is, some people are learning that what you say to AI can be used against you. 

  • Fake news, alternative facts, and deepfakes are factors that have contributed to the erosion of truth and trust in society. Social media laid the groundwork: “algorithmic amplification, influencer dynamics, and participatory content creation within the attention economy have changed how information circulates, enabling the collective construction of alternative realities. Generative AI has served as an amplifier and mitigator, accelerating the production and personalisation of misleading content while also providing tools for detection and adaptive countermeasures.” Klaus Schwab, founder of the World Economic Forum, had this to say about truth: “Truth and trust are often treated as virtues, but they function as conditions: the prerequisites for coherent societies, functional institutions, and stable international systems. Without them, even the most advanced technologies fail to deliver progress; without them, democratic debate becomes impossible; without them, economic and social life slowly lose their connective tissue.” See also, National Politics: Institution Trust; Information Ecosystems; Art & Media. 

Epistemic Challenges

Definition: The nature, scope, and limits of knowledge and reason. This section is closely tied to Ontological Challenges.

Why it Matters: AI is meant to be a knowledge acceleration, with some claiming it will eventually be all-knowing.

  • Most definitions of intelligence include the capacity to learn, solve problems, and make decisions. Artificial intelligence is a broad term for “computer systems that can perform tasks with human-like intelligence, such as understanding language, recognizing images, learning from data, reasoning, and making decisions.” However, the current conversation is contentious and not everyone agrees about how intelligent AI is or if AI is, in fact, intelligent. In public conversations, AI is confused with AGI and ASI. Even the definitions provided in this project are debatable, and this lack of clear definitions is creating confusion and hindering coherent conversations about such a critical subject.

  • AI is altering our relationship with knowledge and what we can do with the knowledge we have in just about every domain. The National Science Foundation issued a memo that stated AI will “transform the questions researchers can answer.” It is removing barriers to knowledge and “giving every scientist access to computational power and tireless research support unimaginable five years ago.” This may accelerate change in an era of unprecedented change. See also, Critical Systems: Scientific Boom; Software: Rapid Progression.

  • Social media and pandemic lockdowns have primed us for AI in many ways. We are accustomed to speaking to others through screens, we’re not critical of our sources of information, and we’re used to personalized and catered information, fed directly to us. Monoculture is dead. Social media was our first contact with AI. It has informed how we will engage with AI and paved the way for us to accept chatbots and other forms more readily.

  • The Dunning Kruger Effect is a cognitive bias in which people with limited knowledge and competence in a domain overestimate their knowledge and competence. As John Cleese puts it: stupid people don’t know they’re stupid people. AI is the Dunning Kruger machine. It confidently gives incorrect answers to questions, invents sources, and hallucinates. 

  • We overvalue the opinions of authority figures. People “routinely display a disproportionate willingness to accept AI-generated outputs as accurate, reliable, and epistemically superior-often even after being explicitly warned that such systems may produce errors.” Deferring to AI’s authority is already leading to costly mistakes, and could result in widespread, irrevocable harm.

  • Our perceptions of AI and its capabilities are shaping our use of it. A study found that “people with less knowledge about AI are actually more open to using the technology,” and that “people in nations with lower average AI literacy are more receptive towards AI adoption than those in nations with higher literacy.” American views on AI are shifting and “previously improving attitudes toward AI have lost momentum.” Gallup found that “Americans continue to believe AI performs worse than people on all of the tasks they were asked about in 2023 and 2026.” There is also a social penalty for using AI: “individuals who use AI tools face negative judgments about their competence and motivation from others.” See also, Work & Jobs: Workslop.

  • Silicon Valley has long glorified the dropout. Now, it’s advocating for skipping college altogether. Rooted in anti-elitism, Silicon Valley’s impulse to forgo college is “shaped as much by Thiel's long-standing disdain for higher education as the Trump administration's attacks on the Ivy League.” Business Insider reports that “the percent difference in earnings between college and high school graduates, or the college wage premium, has held at 75% to 80% for the past decade. For an average American, Deming says, the return on an investment in college — including the opportunity cost of attending as well as the sticker price — exceeds the annual returns of investing in the stock market, buying a home, and starting a business.” See also, Al Companies & Entrepreneurs: Epistemic Trespassing; Young Americans.

  • We still do not understand how human cognition works. In fact, “there are no traits present in humans and absent in other animals that in isolation explain our species’ superior cognitive performance; rather, there are many cognitive domains in which humans possess unusually potent capabilities compared to those found in other species.” Replicating intelligence in other entities is not straightforward when we barely understand our own. See also, Software: Language ≠ Intelligence.

  • We don’t know what we don’t know.