The Cognitive Revolution: Summary of "When AI Will Outthink Humans"

This article is a summary of "When AI Will Outthink Humans" by David V Gilmore.
For over 70 years, artificial intelligence (AI) has fascinated and challenged our perceptions of humanity and the limits of machine intelligence. Predictions that AI will radically transform society have circulated for decades, and while significant progress has been made, the true scale of its impact remains ambiguous. A key question is no longer if AI will transform society, but when it will outthink humans — specifically in terms of cognitive output, or the sheer volume of thought.
In this exploration, we propose a new framework to quantify AI's intellectual power in terms of "thought-hours." Just as the industrial revolution saw innovations like steam engines measured in horsepower, the AI revolution could benefit from a similar intuitive unit of measurement. By using “thought-hours,” we may be able to more clearly juxtapose human cognitive capacity with that of AI. Though imperfect and subject to assumptions, this model can help frame an ongoing discussion about AI’s rapidly increasing influence.

The Thought-Hour: Measuring Cognitive Work
To begin reasoning about AI's cognitive work, we introduce the concept of a "thought-hour," an intuitive approximation of human intellectual output. One thought-hour is roughly defined as the average number of tokens (or words) a human can process in an hour, encompassing both reading and writing tasks. On average, a human reads at 200 words per minute, translating into 12,000 tokens per hour. To account for productivity lags, a conservative estimate pegs the human thought-hour at 10,000 tokens.
This thought-hour measurement is crucial in comparing human and AI capacities, where AI systems such as large language models (LLMs) operate at vastly different token speeds. AI’s ability to process tokens, akin to human thought processes, presents a framework to estimate AI’s cognitive labor in relatable terms.

The Rise of Synthetic Thought-Hours
As AI adoption scales, so too does the demand for greater cognitive output. Enter the concept of synthetic thought-hours, a term used to quantify AI’s cognitive output based on the tokens it processes. Take GPT-4, for example. This model outputs tokens at an impressive rate of 67 tokens per second, or 241,200 tokens per hour—24 times faster than the average human thought-hour.
The comparison doesn’t stop there. In terms of cost, AI is exponentially more efficient. While human thought-hours cost approximately $13, synthetic thought-hours produced by GPT-4 cost just $0.15. This creates an 86-fold cost advantage for AI, emphasizing its scalability and financial feasibility in an increasingly digital economy.

The Growing Cognitive Capacity of AI
The growth trajectory of AI's cognitive capacity is determined largely by the number of GPUs, such as Nvidia's H100 chips, dedicated to processing tasks. Current estimates place the global AI cognitive capacity at 85 gigathought-hours (GTh) annually, compared to humanity’s 1.9 trillion thought-hours (TTh) produced by the global knowledge labor force.
However, AI’s capacity is growing rapidly. With a compound annual growth rate (CAGR) of 36% for global GPU production, the volume of synthetic thought-hours doubles approximately every two years. Given that growth rate, AI could surpass human cognitive output within the next decade. If we factor in improvements in AI algorithmic efficiency, AI may outthink humans in as few as four years.

What Happens When AI Outthinks Humans?
As AI approaches the threshold of outthinking humans, it becomes critical to consider the societal and economic implications. At current rates, AI’s intellectual capacity may soon eclipse that of the entire global knowledge workforce. This cognitive shift holds tremendous promise for industries dependent on cognitive labor—research, education, healthcare, and more—but it also raises existential questions about the value of human labor in the knowledge economy.
The use of AI in cognitive tasks could create massive shifts in employment, requiring human workers to adapt or transition to roles AI cannot easily replicate, such as those requiring emotional intelligence or creative problem-solving. Additionally, as AI systems improve and become increasingly embedded in societal functions, they could reshape the balance of global power, allowing nations and corporations with advanced AI systems to dominate industries and geopolitics.

Challenges to Scaling AI Thought-Hours
Despite the promise of AI's cognitive output, several limitations could constrain its growth. For one, the availability of advanced chips like Nvidia's H100 units is a bottleneck. Current geopolitical tensions, particularly around Taiwan and semiconductor supply chains, could dramatically impact AI growth. Moreover, the energy demands of AI are substantial. While data center infrastructure can keep pace for now, increasing compute demands could strain energy resources as AI continues to scale.
Another challenge is the economics of AI itself. Though AI is capital-intensive, the return on investment (ROI) must be sustained to keep momentum in the industry. As with all technology-driven sectors, the AI “gold rush” must translate into tangible, long-term value for both investors and society.

The Future of AI and Human Thought-Hours
The trajectory of AI’s intellectual capacity will shape the future in ways we are only beginning to comprehend. As AI thought-hours scale and possibly surpass human cognitive work, we may find ourselves at a critical juncture—one where artificial intelligence becomes not just a tool for assisting human labor but a leading force in cognitive labor itself. What this means for human society, work, and identity is still unfolding, but the scale of transformation could rival that of the industrial and information revolutions.
For now, one thing is clear: AI is not just a tool for tomorrow but a transformative force shaping the future today. Understanding when and how AI will outthink humans allows us to plan for the rapid, world-changing shifts ahead, but also prompts us to ask deeper questions about the future of intelligence, both human and artificial.