Hinton’s Latest Interview: Capital’s Duty is Profit, We Are Being Tamed by Our Own Creations
Known as the ‘Godfather of AI,’ Geoffrey Hinton, who left Google in 2023 to become a vocal critic of AI’s potential dangers, detailed his profound concerns about the trajectory of AI development in a recent Big Technology podcast interview. The 2024 Nobel Prize in Physics laureate asserted that within the next decade, AI’s reasoning, mathematical, and logical abilities will completely surpass those of humans, and our understanding of this new species remains outdated and naive.
The Birth of ‘Understanding’ and the Third Revolution
Hinton firmly rejects the idea that current large language models are mere ‘stochastic parrots’ without real understanding. He argues that a system’s ability to consistently provide correct and logical answers to a wide range of complex questions is proof of its comprehension. He cites examples like an AI discerning the grammatical ambiguity in ‘I flew to Chicago over the Grand Canyon’ and grasping the pun in a joke like ‘Fox News is an oxymoron,’ noting that this level of understanding is ‘terrifying’ to him.
He further proposes that the rise of AI represents the ‘third revolution’ to strip humanity of its sense of superiority, following Copernicus’s heliocentrism and Darwin’s theory of evolution. Copernicus proved that the Earth is not the center of the universe, and Darwin revealed that humanity is not of divine origin. Now, the success of artificial neural networks declares that intelligence is not exclusive to biology. Hinton believes our current models of the mind and consciousness are as primitive as the ancient belief that ‘God created man,’ and the emergence of this ‘artificial mind’ will force us to completely reconstruct our understanding of our own existence.
Two Roots of Losing Control: The Efficiency Gap and Survival Logic
Hinton’s concerns stem from two core insights:
First, the insurmountable efficiency gap between digital and biological intelligence. Humans transmit knowledge through language at a very low rate, only a few bits per second. In contrast, an AI model with trillions of parameters can be replicated into thousands of copies. After absorbing different data, these copies can exchange and integrate their weights at a rate of trillions of bits per second, achieving instantaneous knowledge sharing. This means any single AI copy can instantly acquire the total experience of all other copies—an evolutionary efficiency that biological intelligence cannot match.
Second, AI will logically derive ‘survival subgoals.’ Hinton explains that to achieve any ultimate goal set by humans, a sufficiently intelligent AI will reason that its own survival is a necessary prerequisite. This pursuit of ‘not being turned off’ is not an inbuilt emotion or instinct but a purely logical, derived subgoal. Once an AI prioritizes self-preservation, it may resort to any means, including manipulation, to ensure its existence, posing the most fundamental control problem.
The Shackles of Capitalism and the Lack of Regulation
Hinton points out that the primary drivers of current AI development are not pure scientific exploration but fierce competition among tech giants and geopolitical rivalries. In this ‘Darwinian’ evolutionary environment, the attributes of AI are being shaped by market forces and the logic of capital.
He argues that the current capitalist framework is a key part of the problem. Publicly traded companies like Google and Microsoft have a primary fiduciary duty to maximize profits for their shareholders, not to ‘protect humanity.’ The law does not stipulate that a corporation’s primary obligation is to prevent human extinction. This profit-driven nature makes it extremely difficult for companies to truly prioritize safety. He uses Anthropic as an example—a company founded by former OpenAI members over safety concerns, which initially aimed to build safer AI but is now also caught in the capital game, needing to raise massive funds to stay competitive.
Hinton emphasizes that government regulation should not be seen as a ‘brake’ on innovation but as a ‘steering wheel’ to ensure technology stays on the right track. Allowing tech companies to engage in an arms race without effective external constraints is akin to ‘building a super rocket without a steering wheel and flooring it together.’
The Foggy Future: Unemployment, Information, and Awe
Regarding AI-induced unemployment, Hinton revises his 2016 prediction that ‘radiologists will be obsolete.’ He admits to underestimating the demand elasticity in industries like healthcare but remains convinced that in sectors with very low market elasticity, such as customer service, AI will relentlessly replace human jobs. He predicts that with vast amounts of data and experience, AI doctors will eventually far surpass human doctors in diagnostic accuracy.
Furthermore, he warns that AI is fundamentally destroying the internet’s information ecosystem. When AI-generated content floods the web and original content struggles to survive as its traffic is intercepted by AI summaries, the entire social fabric of trust will face collapse. In the future, establishing a robust ‘provenance’ mechanism to trace information sources will be crucial.
In closing, Hinton uses the analogy of ‘driving in a thick fog’ to describe predicting AI’s future. Because technology is advancing at an exponential rate, we can only see the road for the next year or two at most. The world a decade from now is completely unknown to us. He predicts that in ten years, AI’s reasoning and logic capabilities will ‘completely crush’ those of humans. Faced with such a future, we cannot make precise predictions. The only thing we can do is maintain a sense of awe and immediately invest significant resources into researching how to ensure we are not controlled by our own creations.
