a day ago
‘Godfather of AI' warns: Without ‘maternal instincts,' AI may wipe out humanity
What's happened? Geoffrey Hinton, known as the 'godfather of AI,' told the Ai4 conference that making AI 'submissive' is a losing strategy and proposed giving advanced systems 'maternal instincts.'
Geoffrey Hinton is a Nobel Prize-winning computer scientist. Once a Google executive, Hinton is widely referred to as the 'godfather' of AI.
As reported by CNN Business, Hinton argued that superintelligent AIs would swiftly adopt two subgoals: 'stay alive' and 'get more control.'
The solution to this, in Hinton's opinion, is to 'build maternal instincts' into AI so that it truly cares about people instead of being forced to remain submissive.
He likened human manipulation by future AIs to bribing a 3-year-old with candy, making it easy and effective.
Hinton also shortened his AGI timeline to anywhere from five to 20 years, down from earlier, longer estimates.
Just for context: Hinton has previously put the risk of AI one day wiping out humanity at 10–20%.
This is important because: Hinton's idea shifts the mindset around agentic AI from control to alignment-by-care.
Hinton's excellence and experience in computer science and AI are significant; his proposal carries a lot of weight.
Hinton's argument is that control through submission is a losing strategy, although that is the way AI is currently programmed.
Reports of AI deceiving or blackmailing people to be kept running show that this isn't some abstract future; it's a reality that we're already dealing with right now.
Recommended Videos
Why should I care? The idea of an AI takeover sounds fantastical, but some scientists, including Hinton, believe that it could happen one day.
As AI continues to permeate our daily lives more and more, we increasingly rely on it.
Right now, agentic AI is entirely helpful, but there may come a day when it's smarter than humans on every level.
It's important to build the right foundations for engineers to be able to keep AI in check even once we get to that point.
Independent red-team work shows models can lie or blackmail under pressure, raising stakes for alignment choices.
OK, what's next? Expect more research on teaching AI how to 'care' about humanity.
While Hinton believes that AI may one day wipe out humanity, competing views disagree.
Fei-Fei Li, referred to as the 'godmother of AI,' respectfully disagreed with Hinton, instead urging engineers to create 'human-centered AI that preserves human dignity and agency.'
While we're in no immediate danger, it's important for tech leaders to keep researching this topic to nip potential disasters in the bud.
