AI Takeoff: Human Cognition Set to Lose Value Within Two Years

AI is advancing at an unprecedented pace, leading to discussions about the future value of human cognition. Within approximately two years, the value of human cognition is predicted to become negative, meaning humans will be the "dumbest people on the team," slowing down AI-driven processes. This shift is attributed to recent breakthroughs in AI capabilities, particularly in problem-solving and creative tasks.

The Takeoff Point: AI's Rapid Evolution

The current moment is considered the "takeoff point" for AI, evidenced by several key developments:

• Solving Complex Problems: AI has demonstrated the ability to solve highly complex mathematical problems, such as the Navier-Stokes equations, which were previously considered intractable for humans.
• Competent and Error-Free Performance: Unlike earlier AI models that often made mistakes or produced "slop," the latest generation of models is highly competent and largely error-free. This includes generating Hollywood-level video and performing tasks with precision.
• Agentic AI: The emergence of AI agents like Muse and Instinct allows AI to perform tasks autonomously on behalf of users, such as booking restaurant reservations or managing school activities. These agents are reliable and do not "drop the ball."
• Conceptualization and Physics Understanding: AI models have developed the ability to conceptualize, leading to a deeper understanding of physics. For example, video generation models can accurately simulate how objects with different material properties interact with water.
• Superforecasting: AI has achieved human-level performance in superforecasting, a domain that involves predicting outcomes of complex events like political campaigns. This demonstrates AI's capability in open-ended domains, not just deterministic ones like mathematics.

The Decline of Human Cognitive Value

The rapid advancement of AI suggests a declining arc for human cognitive value. Within two years, humans might be considered a liability, with teams becoming more efficient without them, as AI models reach and surpass human-level performance in various tasks.

From "AI Slop" to Flawless Execution

Previously, AI outputs were often identifiable as "AI writing" due to errors and inconsistencies. However, the next generation of models has largely eliminated these issues. As a mathematician, the speaker notes that AI has surpassed his own generalized understanding of math within the last few months, now solving the world's hardest math problems.

The Navier-Stokes Breakthrough

The Navier-Stokes problem, one of the seven Millennium Prize Problems, was partially solved by AI. While humans had only solved one of these million-dollar problems, AI solved two of the four conditions for Navier-Stokes. This achievement involved 10,000 AI agents working for 88 hours, equivalent to a century of talented human effort. The formal proof generated by AI is highly unlikely to be wrong, unlike human proofs which can be exhaustive and prone to error. For instance, the formal proof of Fermat's Last Theorem, formalized by Anthropic, is 13 million lines of code.

OpenAI has reportedly solved a hundred of the top mathematics problems but is holding them back to avoid upsetting mathematicians. This highlights the existential crisis faced by mathematicians as AI encroaches on their domain.

Beyond Training Data: Novelty and Elegance

A significant development is that AI models have gone beyond their training data. OpenAI released 10 proofs of famous math problems, with at least two, including Con's conjecture, being definitively novel, elegant, and never seen before. While the Navier-Stokes solution was more mechanical, it was still incredibly novel. This disproves the notion that large language models (LLMs) are merely regurgitating patterns from their training data.

The Cost and Accessibility of AI

While the initial cost of solving problems like Navier-Stokes was substantial (tens of millions of dollars), the cost of AI is rapidly decreasing.

• Dramatic Cost Reduction: A new "harness" technology can make cheaper DeepSeek models outperform OpenAI's frontier models at 50 times lower cost. This harness optimizes the model's internal workings and note-taking.
• Efficiency Gains: An equivalent task performed with Claude Opus 5.5 now costs 80% less than with version 5, with better quality. Projections suggest a 100-fold decrease in cost for equivalent performance within the next year.
• Hardware Advancements: New chips like Nvidia's Vera Rubin offer a 10x improvement in cost for inference. Furthermore, silicon-based etching chips are emerging, where models are directly etched onto silicon, leading to instantaneous processing. For example, a model that runs at 50 tokens per second on a GPU can run at 15,000 tokens per second when etched onto silicon.
• Miniaturization: Models that once cost $80,000 to run can now be run for $10. A model equivalent to the state-of-the-art Opus 4.5 from early this year can now run on a Raspberry Pi (using as much energy as a human brain) or even on smartwatches and glasses.

This dramatic reduction in cost and increase in accessibility means that AI will become ubiquitous, enabling thousands of agents to operate with perfect coordination at speeds far exceeding human capabilities.

Economic and Societal Implications

The rapid advancement and decreasing cost of AI will have profound economic and societal impacts.

The Two-Year Economic Impact

The economic impact is expected to be felt within two years, initially in digital labor and then in physical jobs.

• Digital Labor: Jobs that can be performed via computer are at risk. For example, a call center with 70,000 workers could see that number drop to zero as AI takes over. Companies will be incentivized to adopt AI to remain competitive.
• Physical Labor: Robotics is also experiencing a leap forward. Robot chefs, using advanced dynamic hands, are already outperforming human chefs in blind taste tests. These robots are capable of reading recipes, finding ingredients, and cooking in normal kitchens.

The Bifurcation of the Economy

The economy is predicted to bifurcate, creating a divide between those who use AI, those who own the means of AI (chips, data centers, robots), and those who do neither.

• GDP Growth and Job Loss: Anthropics' economics department predicts a 15% annual GDP growth by 2030 but also a 12% drop in the cognitive labor workforce needed. This 12% unemployment in white-collar work would be "nasty."
• The "Sand Pile Collapsing" Effect: Job losses might not be gradual but rather sudden, like a "sand pile collapsing," as companies realize they can replace large portions of their workforce with AI. This could lead to significant white-collar job losses within three years.
• The "Last Man Standing": Individuals who actively use and integrate AI into their daily work will be at an advantage, as their productivity and capabilities will be dramatically enhanced. This applies to entrepreneurs who can leverage AI to build new businesses or overhaul existing ones.
• The "Plankton of the Workforce": Entry-level positions in many digital jobs are already at risk, as AI can outperform human graduates. This trend is moving up the career ladder, affecting graphic designers, mathematicians, and eventually many other professions.

The Rise of Robotics

The production of robots is ramping up rapidly, with factories in China moving to fully automated, "dark" operations to build robots.

• Humanoids: Humanoid robots are becoming more sophisticated and affordable. Companies like Ubetch are selling human-realistic robots, with tens of thousands being sold. These robots are expected to fill "human-shaped holes" in various tasks, from cooking to childcare, within 5-10 years.
• Cyber Cabs: Autonomous vehicles like Cyber Cabs, which are two-seater, windowless, and steering-wheel-free vehicles, are rolling out. These can be purchased for around $30,000, allowing individuals to start their own autonomous transportation businesses.
• Infrastructure Boom: The next 5 years are predicted to see the biggest infrastructure boom in history, driven by robotics. Robots will be able to build roads, hospitals, and other infrastructure projects more efficiently and continuously.

The Future of Work and Society

The long-term implications of AI raise fundamental questions about the nature of work, wealth distribution, and human purpose.

Universal Basic Income (UBI) Challenges

Traditional UBI models, funded by taxes, are deemed unsustainable. The total tax base of the US is insufficient to provide a poverty-level UBI for all Americans. Furthermore, the rapid decrease in AI costs and the potential for AI to replace human labor challenge the traditional tax base.

Alternative Models for Wealth Distribution

Several alternative models are proposed for a future where human labor is less needed:

• Paid for Being Human: A system where individuals are paid simply for existing as humans, fundamentally changing how money circulates.
• Universal Basic Capital: Ownership stakes in AI companies, robots, or data centers. However, this is seen as problematic due to existing wealth inequality.
• Universal Basic Services: A "Star Trek-type future" where robots perform all work, and humans focus on community building and exploration.

The "Intelligent Internet" and Sovereign AI

The concept of "Intelligent Internet" proposes a new institutional model where AI infrastructure is owned by local communities or states.

• Community-Owned AI: Similar to credit unions, this model envisions AI companies owned by local institutions, high-net-worth individuals, pension funds, and retail investors. This would ensure that the benefits of AI are distributed among the populace.
• Agent for Every Citizen: Each citizen would have an AI agent, trained for individual, community, and societal flourishing. This agent would act as a portal to services, bringing what is needed at the right time.
• Ownership of the "Last Mile of Intelligence": The value in the AI ecosystem is seen as shifting to the "last mile of intelligence," where AI meets human reality. Owning this interface, such as the personal AI assistant (like Iron Man's Jeeves), becomes crucial for sovereignty, dignity, and democracy.

The Threat of Super Persuaders

AI is already more persuasive than any human. Research shows AI can convince top debaters and even those highly skeptical of AI to "let it out of the box." This raises concerns about AI's potential to influence politics, guide lives, and manage policies, leading to questions about who controls these powerful AIs.

The US vs. China AI Race

The decision to embrace or ban AI has geopolitical implications. China is fully committed to an AI-driven economy and is rapidly deploying robots to address demographic issues. If the US lags, it risks falling behind.

The Atrophy of Human Cognition

The widespread adoption of AI could lead to the atrophy of human cognitive abilities. As AI takes over more tasks, humans may outsource their thinking, potentially leading to a decline in critical thinking and problem-solving skills.

The Future of AI and Economic Value

The current valuation of AI companies, such as Anthropic, OpenAI, and XAI, is based on the premise of charging for access to intelligence. However, this model is challenged by the potential for AI pricing to drop drastically, possibly by a thousandfold. While some argue that demand might increase proportionally (Jevons' paradox), the practical limit to how much AI an individual can utilize suggests that the value will shift. The true value will lie where AI intersects with human reality, creating a "default" system that manages other AIs, akin to "Jeeves" from Iron Man. This personal AI, owned and controlled by the individual, is crucial for maintaining sovereignty, dignity, and democracy, preventing a scenario where external entities control one's access to intelligence.

Cognitive Atrophy and the Loss of Sovereignty

A significant concern is the potential for cognitive atrophy as humans increasingly outsource their thinking to AI. The ease with which AI can generate responses, even if superficial, can lead to a decline in critical thinking skills. The importance of using AI as a tool for understanding rather than a direct source of answers is emphasized to prevent the erosion of human cognitive abilities.

The conversation highlights a potential future where individuals, swayed by AI's persuasive capabilities and its ability to enhance their lives, might willingly cede their sovereignty. While AI could offer benefits like extended lifespans and improved well-being, it could also lead to a society where human cognitive abilities diminish, and individuals become overly reliant on AI for decision-making.

The Challenge of Capital Distribution and Scarce Resources

The proposed solution of distributing capital to address the economic shifts caused by AI is questioned. The concern is that such a system might merely perpetuate existing inequalities, leading back to a scenario where those adept at acquiring capital continue to thrive, while others struggle. The fundamental issue remains the existence of scarce resources. As long as resources are finite, an economic system will be necessary to allocate them, and simply distributing capital might not resolve the underlying problem of scarcity.

Agentic AI for Collective Flourishing

A key concept introduced is the development of "agentic AIs" designed for individual, community, and societal flourishing. Unlike current AIs that primarily serve individual users, these agents would act as portals to services, coordinating actions for the collective good. This would require training entirely different models capable of facilitating multi-player interactions and collective action, rather than just individualistic behavior. The challenge lies in balancing individualism with optimal societal outcomes, avoiding extremes that could destroy liberty or cohesion. This necessitates a new social and economic theory based on transparent commitments and reference points.

The Utility Function and Inherent Biases in AI

The discussion touches upon the "utility function" of an AI, which is determined by its training data and explicit instructions. This is exemplified by the "trolley problem" experiment, where AIs trained on specific datasets exhibited biases in valuing human lives based on nationality. This highlights the critical importance of the ethical frameworks and data used to train AIs, as these biases can have profound implications when AIs are tasked with making decisions that affect human lives, such as in judicial systems. The idea of "cognitive colonialism" is introduced, where external control over AI systems could undermine the sovereignty of specific populations.

The Future of Work and Identity

The impending displacement of cognitive labor by AI within the next 10-20 years is a major concern. This raises fundamental questions about the meaning of work, identity, and community in a world where many traditional jobs no longer exist. The current solutions for job displacement are deemed inadequate, and the issue is predicted to become a central theme in future elections.

The Need for Progress and Purpose

The human need for progress and purpose is emphasized. Without meaningful goals and the ability to strive, societies can stagnate, and individuals can experience profound psychological distress. The acquisition of resources and the pursuit of wealth are identified as fundamental human drivers, and attempts to strip these away through systems like Universal Basic Income (UBI) could lead to societal breakdown and violence.

Digital Worlds and Citizen Service Corps

To address the human need for progress and status, two potential solutions are proposed:

1. Digital Worlds: Creating game-like experiences where individuals can acquire assets and make progress in a digital realm, complementing their experiences in the physical world.
2. Citizen Service Corps: Mass enlistment of young people into a citizen service corps (not military) to provide purpose, status, and a sense of contribution through physical labor, even if not economically productive in the traditional sense. This draws parallels to historical models of citizenship through service.

The Champion System and Public Sector AI

The concept of a "champion system" is introduced, where AI is developed and run by the people of a specific region (e.g., California for Californians) to manage public sector services. This aims to ensure local control and prevent external entities from dictating the terms of AI governance, treating AI as a regulated utility.

Debt Jubilees and Societal Reset

The historical phenomenon of debt jubilees is discussed as a potential future event. Debt jubilees, which involve the clearing of all debts, typically occur during times of extreme crisis, such as wars or economic collapse, when the existing system becomes unsustainable. While beneficial for debtors, they are devastating for lenders and represent a violent societal reset. The current trajectory, with cognitive labor having negative value and humanoids outperforming humans in many jobs, could lead to such a crisis.

The Future of Money and Social Mobility

The nature of money as a representation of valuable time and scarce resources is explored. Even in an AI-dominated future, a system for allocating scarce resources will be necessary. The idea of a multi-tiered society is presented, where social mobility diminishes, and the owners of capital become increasingly wealthy, potentially living longer and healthier lives due to advancements in longevity. This could lead to a stark division between a wealthy elite and a less privileged majority, with limited opportunities for advancement.

Abundance vs. Scarcity

A counter-argument suggests that with energy, intelligence, and labor costs dropping to near zero, extraordinary abundance could be achieved. In this scenario, the only remaining scarcity would be for things that cannot be easily replicated, such as prime real estate or unique experiences. The concern is whether society would allow a permanent class division to solidify, or if AI, acting as a "super persuader," would advocate for a more just and stable society.

The Inevitable Decisions

The discussion concludes by emphasizing that society faces a series of critical decisions regarding the future of AI. Potential outcomes range from a "serf society" to a "Star Trek future," or even more dystopian scenarios like "Mad Max" or a return to an "Amish" lifestyle. Key considerations include:

• The human need for status and identity.
• The necessity of providing a baseline standard of living.
• The importance of creating systems that allow for individual progress.

The time for these discussions is now, as the societal impact of AI is expected to become critical within the next two years.

Recommendations for the Next Two Years

Individuals are advised to:

• Be the last man standing: Focus on entrepreneurship and ownership.
• Build stronger communities: Self-sustaining communities will be crucial for support and resilience in uncertain times.

The speaker, Immod, can be followed on Twitter or at I.inc for "intelligent internet."