“The artificial-intelligence boom has all the hallmarks of a gold rush: frenzied investment, sweeping promises and a race to build ever-bigger models. But behind the hype lies a remarkable shortsightedness -- one that threatens AI's promise to transform our world.
AI systems, including the most powerful large language models, rely on computational force. Despite their apparent sophistication, these models don't understand the world; they merely identify statistical patterns in massive data sets. They can't form abstract concepts, adapt to unfamiliar environments or learn from sparse information the way a human toddler can. They function only with enormous hardware, constant access to vast training data, and unsustainable amounts of electrical power.
The human brain runs on 20 watts of power -- less than a lightbulb. Yet it consistently outperforms AI in the forms of intelligence we care about most: abstraction, reasoning, creativity and social understanding (people skills). To match the computational power of a single human brain, a leading AI system would require the same amount of energy that powers the entire city of Dallas. Let that sink in for a second. One lightbulb versus a city of 1.3 million people.
Do we even need to calculate how much power would be needed for 1.3 million models to grasp how unrealistic that would be?
These extreme energy and data demands are wasteful -- and a warning sign. As models swell into the hundreds of billions of parameters, the infrastructure needed to support them scales exponentially. The U.S. doesn't generate enough electricity to power the growing fleet of data centers behind today's AI. The deeper we dig into this paradigm, the more we run into its bottlenecks: scale, inefficiency and diminishing returns. We are building machines that are larger -- but not smarter.
The path to true artificial intelligence -- systems that can reason, learn flexibly and generalize like humans -- won't come from stacking more graphics processing units. It will come from a crucial insight: Nature has already solved the problem. The human brain remains the most powerful, adaptable and efficient computing system on the planet. It learns with minimal supervision, thrives in uncertain environments, and adapts fluidly across tasks. It is fast, flexible and energetically frugal.
To achieve real AI, we must develop systems that draw not only from the output of the brain -- language and behavior -- but from its underlying architecture and mechanisms. This means investing in neuroscience research, especially in the study of our closest evolutionary relatives: monkeys and apes. Our simian cousins share our visual systems and -- most important -- our core brain architecture. They offer the most direct window into how biological circuits give rise to human intelligence because we can study them with a level of experimental precision that is impossible in humans. The first AI models were inspired by research on the visual system of monkeys, but this approach was abandoned in favor of the simpler models at the heart of today's AI.
There is already a blueprint that we can follow. More than a decade ago, the U.S. launched the Brain Initiative, an acronym for Brain Research Through Advancing Innovative Neurotechnologies. It is a bold, bipartisan effort to map and understand the neural circuits that drive human thought, emotion and behavior. That program has been among the most successful scientific endeavors in recent history, yielding transformative tools for observing and manipulating brain activity. It laid the groundwork for understanding how intelligence emerges from biology. But we are just at the beginning.
What we need now is a second phase: not only to study the brain but to build from it. We need a national effort to translate the brain's design principles into next-generation intelligent machines.
This idea isn't new -- and other countries are already ahead of us. China is making massive state-backed investments in primate brain research, with at least 40 primate breeding centers nationwide -- nearly triple the number in the U.S. It has built a national infrastructure aimed at decoding the brain and converting that knowledge into strategic technological advantage.
By contrast, U.S. research with monkeys is underfunded, vulnerable to political shifts and special-interest groups, and increasingly marginalized in favor of unproven alternatives, such as organ on chips and engineering-first approaches that can't replicate the dynamics of a whole brain in a living organism, let alone a human.
This isn't only a scientific issue. It is a strategic one with profound implications for national security and economic leadership. The country that unlocks the principles of biological intelligence will shape the next century of technology. Just as the Manhattan Project delivered atomic power and the transistor launched Silicon Valley, the next great leap will belong to the nation that builds machines that learn and think like a human brain -- using a fraction of the energy of today's AI.
That leap won't come from scaling up semiconductors and data centers. It will come from understanding neural circuits.
President Trump's administration is betting our economic future on U.S. dominance in AI. But the current AI boom, for all its momentum, is nearing the edge of its paradigm. If the U.S. is serious about creating truly intelligent machines and not merely maximizing capital investments that swell corporate valuations, we must return to biology.
The path to real AI does not run solely through Silicon Valley. It runs through laboratories studying real neurons, real circuits and real cognition in the brains of our primate relatives.
To win this race, we cannot rely on engineering shortcuts. We must understand how the "computer" in each of our heads works. Only then will we be able to build machines that achieve genuine intelligence.
---
Mr. Miller is a professor of psychology at the University of California, San Diego.” [1]
1. The Future of AI Lies in Monkeys, Not Microchips. Miller, Cory. Wall Street Journal, Eastern edition; New York, N.Y.. 21 Aug 2025: A15.
Komentarų nėra:
Rašyti komentarą