00:00A stark warning about artificial intelligence from Jeffrey Hinton, the Nobel Prize-winning scientist whose pioneering research helped lay the groundwork for AI.
00:08That's why he's often called the godfather of AI.
00:10Hinton says there's a 10 to 20 percent chance AI will wipe out humans,
00:15but made headlines today suggesting at a conference that the tech industry should somehow install what he termed maternal instincts into superintelligent AI.
00:24So they'll protect humans like mothers protect their babies.
00:27They've been saying, we have to stay in control of these AIs.
00:34We've somehow got to be stronger than them.
00:37We've got to be dominant and they've got to be submissive.
00:40That's not going to work.
00:42We have to make it so that when they're more powerful than us and smarter than us, they still care about us.
00:50Jeffrey Hinton joins me now.
00:53Professor Hinton, your remarks have obviously gotten a lot of attention.
00:56You talked about engineers building maternal instincts into AI models.
01:01What would be the purpose of that?
01:03What would that do?
01:04And is it even possible?
01:07Most of the AI experts believe that sometime in the next five to 20 years, we'll make AIs that are smarter than people.
01:14And they'll probably end up much smarter than people.
01:16And there's very few examples we know of smarter things being controlled by less smart things.
01:22In fact, pretty much the only example we know is a mother being controlled by her baby.
01:27To make that happen, evolution built maternal instincts into the mother.
01:31And if we don't do something like that with these alien beings that we're creating, we're going to be history.
01:37How hard is it from a, I mean, just a technological perspective to actually build motherly instincts?
01:45I mean, is there an example of that being done at all?
01:48Or has that happened at all?
01:51The only real example we have is evolution.
01:54Evolution obviously made a pretty good job of it with mothers.
01:57People haven't been focusing on that.
01:59They've been focusing on making these things more intelligent.
02:02But intelligence is just one part of a being.
02:04We need to make them have empathy towards us.
02:07And we don't know how to do that yet.
02:09But evolution managed it, and we should be able to do it too.
02:11You know, we've heard from so many people, leaders in the tech field, who say,
02:15well, look, if the U.S. doesn't win this war, this battle for dominance in the AI world,
02:22you know, a rogue nation, a country that's an enemy of the United States, China, Russia,
02:29they're going to have AI.
02:31If China is not developing AI that has maternal instincts, and the U.S. is the only one who
02:38has maternal instincts, does that not make them, I mean, is that not making them weaker
02:45in this war against, for dominance?
02:47Well, there are many risks of AI, like cyber attacks and loss of jobs and making nasty viruses.
02:55But one of the risks is this existential threat that AI will take over.
02:59And for that threat, all the countries will collaborate because they're all in the same boat.
03:03No country wants AI to take over.
03:05Just as the USSR and the Americans collaborated at the height of the Cold War,
03:13countries will collaborate on how to prevent AI from taking over from people.
03:19Do you, I mean, do you believe human nature won't encourage some leader somewhere to,
03:25or some government somewhere to believe that they can control it, whether or not that's actually possible?
03:31Well, we have a nice example in the States of a government that's determined to control it.
03:38I don't think that's going to work.
03:40It may work against other countries, but when AI becomes smarter than people,
03:45that's just not going to work.
03:46This whole idea that people need to be dominant and the AI needs to be submissive,
03:51that's the kind of tech bra idea that I don't think will work when they're much smarter than us.
03:54Do you, do you really believe, do you still believe that 10 to 20% chance of AIs wiping out humans is possible?
04:05Oh yes.
04:06If we don't, if we can't figure out a solution to how we can still be around when they're much smarter than us
04:13and much more powerful than us, we'll be toast.
04:15What do you think?
04:16I mean, it's stunning to me that this technology, I mean, it's already changing things,
04:20but the tidal wave of change that we are going to be seeing shortly and are already seeing,
04:28that it's being run by people who are not elected.
04:30These are people, you know, who are making billions of dollars doing this.
04:34Nobody ever really elected them.
04:36There's been no consensus agreement that, yes, this is good for humanity.
04:41Do you think the average person, and I include myself in that grouping,
04:47understands what is about to happen here?
04:51No, I think they don't.
04:53And I think it's important for the public to understand it.
04:55We need a counterpressure to the tech bros who are saying there should be no regulations on AI.
05:01And also, we need the countries to understand they need to collaborate on how to avoid the existential threat.
05:08Does, what do you think it does to initiative for, you know, I have a three-year-old and a five-year-old.
05:15I'm very worried about, you know, what world they're going to have in 10 years,
05:20when, if there's machines that do everything better than humans do,
05:23what's the initiative for people to strive to become good at something?
05:27One thing we know about mothers is mothers genuinely care about their babies,
05:32and they do whatever they can to make life interesting for the babies
05:36and to make the babies grow and realize their full potential.
05:38So if we have super intelligent AI looking after us,
05:42we may be able to get the same thing.
05:44Jeffrey Hinton, it's a real pleasure to talk to you.
Comments