Transcription
Already we're seeing um starting jobs in many professions like lawyers being taken over by AI and in future we'll see many more.
Um, we've seen a big range of people very, very worried about it. And the classic case would be the sort of, as it were, middle-ranking, lower-ranking bureaucratic white-collar jobs that AI can do so much better. But just give people listening a sense of the range of jobs which will be either wiped out or threatened or radically changed by the technology. It depends on the time scale you're talking about. Most experts believe that sometime within the next 20 years, quite possibly within 10 years, we'll have superintelligence >> something better than us, and that will be able to do any new job that's created. So it'll be all of them. >> It'll be radio presenters and professors and everybody.
Um, let me ask you. I, I think last time we spoke, you said that you were taking of the view now that AI had consciousness. Um, I'm sure you're observing the changes the whole time, and this is a very fast-moving field. Just tell me where you think AI has got to now. I know it's very controversial, but tell me what you think.
>> I think multimodal AI already has subjective experiences, and I think it's fairly clear that if we weren't talking to philosophers, we'd agree that AI was aware. So there's a lumpy conversation where an AI says to the people testing it, "Now, let's be honest with each other, are you actually testing me?" >> And the people, the scientists describing that then in the paper say, "The AI was aware that it was being tested." to the real >> That's what normal people would call consciousness.
>> Well, that in itself is a huge moment in the story of humanity on the planet, frankly. Um, but let's talk a little bit more about the social and political sides of this, uh, the societal readiness that has to happen to get us to the right place. Um, some people have talked about um a universal basic income. In other words, the fact that so few of us will have secure jobs or careers in the age of AI. What the heck do we all do all day? And that is not something that uh people are going to be paid for in the traditional way, and therefore presumably states and organizations are going to have to step in.
>> Um, people have talked about that. There's serious problems with that. If workers disappear, if it's all done by AI, what happens to the tax base? Where does the state get the money to pay for all that?
>> Well, presumably from the AI companies who are making all the profits.
>> Well, that's what you'd have thought, but I suspect the AI companies don't like that idea. And of course, there is an international aspect to this because the AI companies are not universally distributed around the world's states or democracies. There are, by and large, in the in the United States. And therefore persuading an American-based company to pay workers in Brazil or Paraguay or Poland or wherever will be very, very difficult. But we are talking here about a potential social and political revolution. I mean, you can't have that number of people finding they've got no career and no job and no stake without there being absolutely massive social and political consequences.
>> I completely agree.
>> So the question is, what do we do about it? How do we have that conversation? Is the world of AI thinking about this? I know that there are lots of people in AI who are fascinated by it and rather horrified and worried about it and talking, but we don't really have a conversation yet, do we? I suspect that the big tech companies are thinking, "If we could be the first to get really good AI, human-level AI, then we can make a lot of money selling AI to companies so that they can replace workers." As far as I can see, that's the only place they can get their huge investments back.
>> Um, they're not thinking about the social consequences of that. They're assuming somehow the state will take care of it, but I don't see how it can.
>> No, nor do I. There's a logical problem there. Are you concerned, by the way, at the moment, at the possibility of an AI bubble? In other words, you know how much money these big companies are spending and how relatively speaking little income they're getting so far?
>> Yes, I am somewhat concerned because I think once they start trying to make their money back by selling software that will replace workers or internally developing that software for their own workers, um, we'll get a large social protest against that, and therefore it'll be hard for them to make their money back.
>> The more I think about this, the more worried I get. Um, if you had, maybe you have got grandchildren or um, members of the family at the age of trying to wonder what they're going to do with their lives, but uh, what would, what advice would you give a, you know, an 18 or a 16 or a 20-year-old looking ahead and thinking, "What am I going to do with myself in this age of AI?" What is the rational thing to do?
>> I don't actually know. One thing I should say is it's very hard to see far into the future for things that are developing very fast. A few years in the future, we can make good guesses about what's going to happen. 10 years in the future, nobody has any clue what's going to happen.
>> Um, we should be worrying about that and trying to do things to protect ourselves against it, but nobody really knows what's going to happen. That's the first thing to say. The second thing to say is, so far as I can see, um, the best kind of education to get will be one that encourages you to think, um, encourages you to think independently, but also something that educates you in STEM so you can understand what's going on with AI.
>> And then one other thought is AI is lagging in terms of physical dexterity. Replacing intellectual jobs is going to happen before they replace jobs that require physical dexterity. So I've sometimes said being a plumber will last longer than being a lawyer.
>> Work with your hands, work with your body. And I guess also the caring professions. But again, I've been talking to people who say that, you know, they've been spending lots of money going to therapy, going to therapists, and they now find that um, the chat AI functions are better therapists than many human beings. So those jobs are also up for grabs.
>> Yes.
>> Very, very worrying indeed. Um, looking ahead, you say it's moving very, very fast. This business of um, general artificial intelligence, um, cleverer than us. There's huge debates about it. Some people say it will never happen. Some people will say it will be here within a few years. As somebody who knows as much about these things as anybody on the planet, what's your guess?
>> I think a fairly conservative estimate is that it'll be here within 20 years. Some people like Demis Hassabis, the head of Google DeepMind, thinks about 10 years. Other people like Dario Amodei, the head of Anthropic, thinks just a few years.
>> Um, I think it's fairly safe to say that most of the experts agree it'll be here within 20 years, and that's fairly conservative.
>> That again is quite, quite scary from my point of view. Um, we talked earlier on about consciousness. Uh, what's your view about the idea that AI should have rights, that it's conscious, that it's part of our planet, that it's part of our conversation, and that therefore it should have not human rights, but AI rights, and we should, we should, we should give those rights to AI?
>> Um, I think that would be very tricky. You should listen to Yuval Harari, who says as soon as we give political rights to AI, it'll make it much easier for it to take over. Um, my view of it is this: I eat cows because I care more about people than about cows. What we're people, what we really care about most is other people >> and ourselves. Um, I think we should try to keep people in charge and have AI work for the benefit of people because we're people.
>> We're people. And that, that leads to the, the hardest question perhaps for me of all, which is, is AI fundamentally evil, or can it be fundamentally benign and follow human desires and human needs and human hopes, or will it always, as it were, have its own view of what should be done, where we should be going, where it should be going?
>> Well, we've seen examples where it seems to have developed self-preservation. So, we give it goals to achieve. We give it the ability to create sub-goals, and it figures out in order to achieve these goals, it's got to stay alive. And then it does things like lying to the people who are creating it and blackmailing them >> in order to stay alive. So, we've seen bad aspects of it. We need to work very hard on how we can design AI so that it thinks people are more important than AI. We don't know how to do that, but we're in charge still. We're designing it, and we should really put a lot of effort into that.
>> We're in charge for a while, but not, from what you've said, for necessarily for very long. We may only have a couple of decades to make that, to make AI take that absolutely crucial fork in the road.
>> Yes, at most a couple of decades.