Geoffrey Hinton: Why Digital Minds Might Beat Ours

A ten-minute clip, uploaded on 8 October 2026, cut from StarTalk’s February episode with Geoffrey Hinton. Despite the title, it stops just before the part about digital minds winning. What’s left is Hinton’s account of where neural networks came from, and it’s the better half.

In the 1950s, he says, AI split two ways. One camp took reasoning as the essence of intelligence: premises, rules for manipulating symbols, conclusions, much like algebra. The other started from the only intelligent things we know, brains, which are good at perception and analogy and bad at formal reasoning until about adolescence, and asked how networks of cells could learn. He puts von Neumann and Turing in the second camp and notes that both died young. His own start was a school friend in the mid-1960s who told him memories might be spread across many brain cells, like a hologram, rather than stored in one place.

As a graduate student he found the method he kept for life: if you have a theory of how the brain works, simulate it on a computer, and most theories fail when you do. He spent his career on how to change the strength of connections so that a network learns something hard, and admits he never found out how the brain gets the signal telling it which way to adjust. Computers do it, through the learning rule he helped popularise, and that is what frightened him in early 2023: the digital version might simply be better than ours. He explains the idea with gas laws. Pressure and heat come from countless molecules nobody sees, and a word is a large pattern of neural activity, with similar words making similar patterns, cat and dog sharing most of their features. Reasoning sits on top of that microscopic layer, and the action is underneath.