A profile of and interview with deep learning pioneer Geoffrey Hinton, who shares why he now thinks neural networks represent a “better form of intelligence”
“I have suddenly switched my views on whether these things are going to be more intelligent than us.”
MIT Technology ReviewWill Douglas Heaven
Context & Ripple Effects
Hinton’s assessment marks a sharp departure from his earlier public posture: in his 2016 discussion of why people should not fear AI, he emphasized what systems could take on next rather than the prospect of them exceeding human intelligence.
The interview also sits beside an active debate over deep learning’s limits and whether neurosymbolic approaches could be needed for AGI. Hinton’s change of view gives that debate added significance because his work helped establish the neural-network methods now at its center.
First-order effects
Hinton’s public endorsement of neural networks as a potentially superior intelligence strengthens the credibility of claims that current deep-learning approaches may have broader capabilities than many researchers previously assumed.
It raises the salience of safety questions around neural-network progress, particularly given his separate warning in the corpus that companies are under-investing in safety research.
Second-order effects
AI labs and funders face greater pressure to show that capability work is matched by rigorous safety research, rather than treating safety as a secondary concern.
The reversal sharpens competition between research agendas: advocates of scaling neural networks gain a prominent voice, while researchers focused on their limits have a clearer case for testing where the approach fails.
Third-order effects
If influential researchers increasingly treat neural networks as a route to intelligence beyond human performance, AI governance will be shaped less by whether such systems are plausible and more by how their development and deployment should be constrained.
The field may become more bifurcated between scaling-led development and efforts to build systems with stronger reasoning or safety properties; the corpus does not establish which path will prevail.
The trend: This is one data point in AI’s shift from debating whether deep learning can generalize into a broad intelligence to debating the risks, safeguards, and alternatives around that possibility.
If only people were half as scared of the climate crisis as AI world domination, then maybe we'd have half a chance of solving it. https://www.technologyreview.com/ ...
An excellent, nuanced interview with Geoffrey Hinton about his fears of AI by @strwbilly - presumably rushed to print after the prof quit Google yesterday and warned of dangers ahead. It seems Netflix's Don't Look Up might have played into his thinking. https://www.technologyrevi…
Good piece on AI. I believe the US already uses AI for strategy. It explains their clinical destruction of free press while at the same time ‘glorifying it’. Let's face it, the USG is rubbish at most things, but lately they are ‘winning’. Why is that? https://www.technologyreview…
We have a new, in-depth interview with Geoffrey Hinton up today, detailing his growing concerns with AI. It's thought provoking and frankly a little unnerving. Worth a read. https://www.technologyreview.com/ ...
Fantastic piece by @strwbilly dissecting Hinton's fears about AI. If you read only one piece about him, make it this one. https://www.technologyreview.com/ ...
“When Hinton saw me out, the spring day had turned gray and wet. ‘Enjoy yourself, because you may not have long left,’ he said. He chuckled and shut the door.” https://www.technologyreview.com/ ...
A timely warning from @geoffreyhinton on the dangers of #AI surpassing human intelligence - and control - with worrying implications for everything from disinformation and deep fakes, to human resources & even autonomous weapons systems. https://www.bbc.com/...
There's a weird sort of hubristic humblebrag in all this - the thing we created is going to be so smart it will kill us all, but still, it's at least a thing *we created* https://www.bbc.com/...
“These things are totally different from us,” he says. “Sometimes I think it's as if aliens had landed and people haven't realized because they speak very good English.” https://www.technologyreview.com/ ...
Hinton now thinks there are two types of intelligence in the world: animal brains and neural networks. Great in-depth piece from @strwbilly after his exclusive conversation with Hinton last week. https://www.technologyreview.com/ ...
Quick observation: when was the last time someone had to leave a university to speak freely about their work? Research at universities is by far a superior model. https://twitter.com/...