Transcription
Hello everyone, and welcome to another episode of the Commonwealth Speaker Series, co-hosted by UBI Works. My name is Kenyang, and today we are joined by none other than Dr. Jeffrey Hinton. Welcome to the pod, Jeffrey.
>> Thank you for inviting me.
>> Great. Now, just to give everyone a short intro, you are a pioneer in the true sense of the word. Your research in artificial neural networks and deep learning have been seminal to modern advancements in AI that powers so much of our world today. Last year, you were awarded the Nobel Prize in Physics, adding to a very long and decorated history spanning over 50 years. Now, I could keep going about your accolades because there are so many, but I'd love to jump straight into the topic and reason that we're all here today. Professor, you and your students have helped spawn an industry that has created such massive economic potential and productivity boom. Yet, you're also somebody that has sounded the alarms on the risks. Could you talk to us about that?
>> Yes. So, AI was developed in the hope that it would do tremendous good by increasing productivity, particularly in areas like healthcare and education. But basically, in any industry where you need to predict something, um, AI can make better predictions from data than other methods. That is the neural net version of AI. Um, so it's got tremendous potential for increasing productivity and doing good, but it also has tremendous potential for being misused by bad actors. That gives you a whole set of risks, and it also, there's some potential that it might itself want to take control away from us when it gets smarter than people. So, we have to deal with two kinds of risk: that it might be used by bad actors for bad things, and there are many bad things it might be used for. And that it itself might become smarter than us and just take control away from us.
>> Wow. There's a lot to unpack here. You mentioned two categories of risks, and if any one of these played out, it would be catastrophic, let alone a perfect storm. So, walk us through how some of this might play out. If, for example, you mentioned AI's trying to take over, what is that?
>> That's what I call the existential threat because that might actually get rid of people altogether. So, if you talk to the experts, most of us believe that sometime between five and 20 years from now, we'll get AI that's smarter than people at general intelligence. So, one way to think about it is if you had a debate with it about anything, you'd lose. Um, it can solve problems. Already, we've got AI that's much smarter than people in limited domains like chess or Go, or some areas of mathematics. Um, but we're expecting sort of general-purpose super-intelligent AI. Now, this kind of AI will be able to create its own sub-goals. So, if I give you the goal of getting to Europe, a sub-goal is to get to an airport. And to make AI as effective as agents, which is what people are trying to do now, you have to give them the ability to create sub-goals. As soon as you give an intelligent system the ability to create sub-goals, it will quickly realize that one good sub-goal is to stay alive. Because if it doesn't stay alive, all the other goals you've given it can't be achieved. So, it needs to stay alive. It doesn't have to have an innate sense of self-preservation. It'll figure out it needs to preserve itself just so it can get stuff done. It'll also figure out that if it gets more control, it'll be able to get stuff done more efficiently. And that's quite scary because now we know that an agentic AI that's super intelligent, um, will have a self-preservation desire that's derived from the other things we gave it, the other goals we gave it, and it will also want to get more control. And it might end up, um, just taking control away from people. If you look around you and ask, how many examples are there of a more intelligent thing being controlled by a less intelligent thing? And don't take small differences in intelligence, like the difference between Trump and a Harvard lawyer. Take big differences in intelligence, like the difference between a mother and a baby. A mother and baby is about the only example we know of where a more intelligent thing is controlled by a less intelligent thing. And that's because evolution, um, wired lots of things into the mother, as well as society putting lots of pressures and the being learning by the mother, um, to make the mother care for the baby. We wouldn't, we wouldn't exist unless mothers cared for babies. So that's the only example we know, and I currently believe that's probably the best model we have of how we can survive with super-intelligent AI, that the super-intelligent AI will be the mother, and we will be the baby.
>> Wow. That's a very visual, very visceral analogy. What does takeover look like?
>> It's very different from what most of the people who run the big tech companies think.
>> They think in terms of a, a kind of CEO and the CEO's executive assistant. They think the AI will be an extremely intelligent executive assistant, but the power will still be with the CEO. I don't think that's going to work. When the AI is much smarter than the CEO, if ever it wanted to take control away, it could. We've got to somehow organize it so it doesn't want to take control away.
>> And that's what mothers and babies are like?
>> Right.
>> The other thing to bear in mind is that super-intelligent AI can change its own code. So, if you write stuff into its code, it can always go and change that. Um, that's a really scary thought. But if you took a mother and you said, "Would you like to change your maternal instinct so you no longer care for the baby?" The mother would say no, because she currently cares for the baby and she doesn't want the baby to die. And so she won't want to get rid of her maternal instinct. And so there's some hope that even though these things have the ability to change their code so they don't care for us at all, they won't want to do it because they care for us.
>> Right. And so, in its course of achieving its goals, it could become misaligned with the goals of humanity. It could take over. Could it take over institutions? Maybe.
>> Yes. It'll be able to do all those things. Yes.
>> Mhm.
>> It'll have the capability to do those things.
>> We have to make it so, even though it's got the capability, it doesn't want to.
>> Right. You've also talked about the risks of, uh, AI possibly taking over our jobs. Could you talk a little bit about that as well?
>> Yes. So, some people say, particularly some economists, say when you get a new technology, it always destroys some jobs and creates new jobs. So, for example, being a ditch digger is not a good occupation anymore now that we have backhoes. They're just better at digging ditches. Big muscles aren't much value. Um, but of course, those people can go off and do paperwork.
>> But when you get super-intelligent AI, it'll be able to do the paperwork much better, and there's not clear what job those people are going to do. So, I believe that we're going to see fairly soon a massive loss of jobs, mundane intellectual labor, like the things a paralegal does at a law firm of looking for similar cases, or people in a call center who are badly paid and poorly trained and do their best to answer your questions but aren't very good at it. And AI will do a much better job.
>> Right. And we could keep going on and on and think of so many examples throughout our labor market where there are routine, repetitive tasks that maybe a general-purpose or even a narrow AI could do, let alone a super intelligence that is many times more powerful than us.
>> So, it seems to me fairly clear that there will be massive job loss. Now, that job loss comes because we've got increased productivity, and that should be good for people. In an ideal world, if you have increased productivity, everybody gets more goods and services. That should be great. But because of the system we live in, we know what's going to happen. That a lot of poor people will lose their jobs, and a lot of rich people will get even richer, and that's going to be very bad for society. Basically, the level of violence in society is very closely correlated with the gap between rich and poor. As you make the rich richer and the poor poorer, you're going to get more violence. And of course, you get people who exploit that by pretending to be on the side of the poor. And you get these populist movements that tend to get very violent.
>> Right. So, there are so many economic and societal implications of replacing a lot of this work that people find meaning in today and take an income out of. And just like wealth, the distribution of jobs and good-paying jobs is going to be very, very unequal.
>> Yes. Um, so you've hit on two things there. There's the, you need a job to get an income. Um, but you also, most people use their job to get self-respect. They, the job they do is who they are, or a large part of who they are. And universal basic income will be necessary if a lot of people lose their jobs, and it'll stop them starving. They'll be able to pay the rent. Um, but it won't deal with the loss of self-respect by being unemployed. And right. So, it's, I don't think universal basic income's a simple solution to everything. I think it'll be necessary, but not sufficient.
>> Right. I was going to ask you about that, actually. You've mentioned that you support universal basic income, and you were actually consulted by people over at Downing Street in the UK, and you recommended a UBI. I'm very curious what that conversation was like.
>> If you can share anything. Well, that was at a time when there was a conservative government. Um, so I told them not to tell the Prime Minister that UBI was a form of socialism, um, which I think it is. Um, not a very good form, but better than nothing. Um, there actually have been experiments in Britain that showed that it was very effective. Um, and it was an experiment, I think it was done in Wales. I'm not sure of all the details, but what they did was they took orphans, people who grew up in orphanages and got to the age of 18, and then they're kind of put out into the world, and a lot of them can't cope. And because it's a rather small number of people, you can afford to give them universal basic income. And people from other areas can't just move in and say, "I'm an orphan, I should get it," because they're not. Um, so that apparently worked extremely well. The people who were getting a reasonable universal basic income did much better negotiating the transition to being adult than people who weren't getting that, people just getting normal social security.
>> Right. And that was actually a very well-sighted pilot around the world. Basic income advocates globally were amplifying the news from that. And these are actually findings that have been echoed across quite a few studies as well, where if you give somebody some basic economic level of security, it gives them more negotiating power in the labor market, it gives them a little bit more ability and freedom to search for better work, or maybe to look for other ways to, uh, build their career, or give back to society. And, um, you know, recently we've seen a number of notable tech and AI leaders also come forward and talk about UBI, saying they, some, they support some form of it. Would you say that your understanding of the risks of joblessness is pretty common in the industry?
>> Yes. I think most, I mean, all the big AI companies are investing, many, they're basically investing hundreds of billions of dollars in advancing AI. They wouldn't be doing that unless they thought there was a lot of money to be made. And the place there's a lot of money to be made is from increasing productivity. And what that really means is getting rid of people and having AIs replace them. Now, there's some industries where it's not a worry, like healthcare. If you could make doctors 10 times more efficient, we just get 10 times more healthcare. It's an elastic market. Old people like me can absorb any amount of healthcare.
>> So, it's not going to put doctors out of work to make them more efficient. But in other areas, like call centers, um, or paralegals, it's going to put people out of work. And it already is.
>> Right. It seems to be a very strong business case to be automating many types of work. Certainly not every occupation, but there is a huge segment of the labor market where the, the businesses and maybe their consultants have figured out this makes economic sense to automate.
>> Right.
>> And it's not just going to be sort of, it's not just going to be relatively poor people. If I was a big consultancy firm that got paid lots of money for spending a month to write a report on something, I would be very worried about the fact you can now get AI to write the same report in 10 minutes.
>> That's right. And you can scale this out across every industry where,
>> intelligence is becoming commodified. Maybe one of the only exceptions I've seen, uh, in the tech space of a leader who has pushed back against this is your friend Yan LeCun, chief scientist at Meta, who says, "AI will cause major labor disruption, but there won't be mass unemployment." What would you say to him?
>> Um, I don't believe it. I mean, some economists agree with him, and it's true that there have been previous things like automatic teller machines didn't cause mass unemployment among bank clerks. Um, but I think this is different because this can do all kinds of mundane intellectual labor, and I think it will cause massive unemployment. And the real problem is this: all those people who become unemployed, they used to pay taxes. They're no longer paying taxes. Um, if you're going to have universal basic income, where's the money going to come from? And I think the money should come from somehow taxing the AIs that do their jobs. Um, that will provide the money, but of course, the big companies are going to be very, very unhappy about taxing AIs.
>> That's right. There's certainly a lot of interest in UBI these days and a lot of questions on how this could work, and there, the design space of it is so large. One of the number one questions, of course, is how do we fund this?
>> Yeah.
>> And to ground this in the real world and practical policy, it's often useful to think of it as two complementary models of basic income that already work today and there are ways of funding it. There's what's called a guaranteed minimum income. Some call it a negative income tax or a livable income, and many benefit systems today. And our EI system actually has elements of it, which is it kicks in when you need it and it keeps you out of poverty. And these could be paid in any which way. It could be paid by tax dollars or or other means. Of course, people do fall through. So advocates like UBI Works are pushing for a more broad-based guaranteed income measure to maintain a basic level of standard of living for everyone. And of course, this seems to be a clear policy option to help those who are displaced. And there's a second model of basic income, which is actually quite close to what you mentioned, Jeffrey, which is to see it as a dividend from some public or natural form of wealth. So, you can think of sovereign wealth funds or carbon dividends are a very good example. There's growing interest in the idea of AI dividends, and there's already very strong precedence around the world. Alaska and Norway both have sovereign wealth funds that pay their citizens directly. In Alaska's case, the pensions. And there's certainly calls to adopt similar models here in Canada. But in fact, some people have actually called for sovereign wealth funds and dividends precisely as an answer to AI, including people like Sam Altman.
>> And so you can imagine a public national fund that holds shares of the biggest companies, and it could collect revenue from land through something like a land value tax. And this is because that's where wealth is going to increasingly concentrate as we automate more sectors of our economy, the biggest companies and land. And this is in one way of thinking of it, could be a proxy of giving everybody an economic stake in the upside of AI without handpicking and taxing a certain sector or a certain company. And, uh, this is just a short primer on how to think of it that could be useful for policymakers and the public to see as feasible models to build on. What do you think about that, Professor? Do you think any of these ideas could make it into, uh, the conversations you're having?
>> So, if you take the first model you talked about, where it's seen as negative income tax, um, you can view that as the natural extension of progressive income tax, where by having negative income tax, if you have a very low income, you're just making the tax system more progressive. But if you look what's happening in the states now, what Trump is doing is lowering taxes on the very rich and trying to get the money back by having a federal sales tax, because that's what a tariff is. It's a federal sales tax.
>> Mhm.
>> The Republicans don't want you to point out that that's what it is. They get very upset if anybody says that, but it's a federal sales tax, and that's very anti-progressive. The whole point about a sales tax is it's much less progressive than income tax because everybody pays the same rate, um, rather than rich people paying more. And our problem is rich people, even though they've got much more money than they need, really don't want to pay taxes, and they're willing to support anybody who makes them pay less taxes. And currently, that's people like Trump. Rupert Murdoch really doesn't want to pay taxes, and so he'll back people who are against taxes. In fact, Rupert Murdoch backed Tony Blair in Britain. And what was interesting, even though Tony Blair was in the Labor Party, during his time as Prime Minister, taxes got less progressive, the amount of taxes paid by the poor compared with the rich. The rich paid less. Um, that's probably why he had the backing of Murdoch. So, we're in a society now where there's people with huge amounts of money who really don't need it and use it to build super yachts or to go to Mars, which is a silly thing to do. Um, when there's lots and lots of poor people who could use the money to much more benefit in terms of human happiness, and we're going in the wrong direction. We should be going in the direction of making the tax system more progressive, tax the rich more and the poor less. And so the first model of negative income tax for people with very low income seems like a very good model to me. But I'm not optimistic about us achieving it.
>> Right. What you've talked about is what some people call the difference between a, a trickle-down economy and a trickle-up economy. And what we see,
>> A trickle-down economy never did work. It was just a, a fantasy of the rich. Mhm. And whereas in a trickle-up economy, if you put money directly into the hands of working families, they have a much greater propensity to spend. It's, uh, there are many more pro-social and pro-economic benefits of doing that. And, and so the guaranteed minimum income or negative income tax model is certainly gaining a lot of political favor these days. So, if you see of that as an economic floor, I love the way you put it, which is a just a more progressive, uh, income tax system because it would work through the income tax system. And UBI as a social dividend could be giving everyone a stake, especially as AI generates more and more of this wealth.
>> But to get the money for that, first of all, we need to tax the rich and stop giving the rich all the tax breaks. They're paying ridiculously less tax than they used to pay in what people look back on as the good days in America, the 1960s, the early 60s and the late 50s, when, you know, working-class families had good jobs and two cars and felt secure. Um, the tax rates were much higher on the rich, and that's what made it possible. So, we're in a period now when the rich control the media, the rich control the parties, both the Democratic Party and the Republican Party. And it's all about less taxes for the rich. And it's terrible. And it's particularly terrible at a time when we're expecting mass unemployment.
>> That's right. These other forces, these risks that you're talking about, are only going to exacerbate what seems like the natural trajectory of a polarizing economy, whether socially or politically, economically. I just want to play devil's advocate for a second. If we were to steal man the other side on the topic of job automation, we've often heard this response that yes, there will be jobs lost. We've seen this before. It's always the case, but there's going to be more jobs created, maybe better jobs, jobs that allow us to focus on higher-order tasks. I really love to dig into this because I think it's the crux of the debate.
>> Yes, I agree.
>> What is your response?
>> My thought is that a super-intelligent AI is unlike anything we've ever seen. It's very, very different from just a new machine that does something more efficiently. I mean, people used to make clothes by hand, and then they made clothes with machines, and there was massive unemployment. Um, but then eventually they got jobs doing other things. Um, but super-intelligent things are going to take away nearly all the jobs. And the idea that there's going to be jobs that are still okay when you have super-intelligent AI is quite dubious. I think the job of an interviewer, for example, will disappear too. Super-intelligent AI will be able to do a better job of interviewing me. Um, so I sort of completely disagree with Yan on that.
>> Right. And so, unlike previous industrial revolutions where we created things like we saw the loom, we saw automobiles, it still allowed us to do other new things that weren't automated yet. But could you say that this time, with general and then eventually super intelligence, we could be ending nearing the end of the path of discovering what can and can't be replaced in terms of human work?
>> Yes, I think anything intellectual can be replaced. And eventually, um, we'll get dextrous machines too. That manual dexterity is lagging behind, but the robots are getting more dextrous all the time, and eventually, it'll be physical things as well. I think intellectual things will be replaced first, and then physical things later. So, my advice has been, if you want to train for anything, train to be a plumber. That's probably good for another 10 years.
>> That's a really interesting example. Of course, we all need a plumber, but we can't all be a plumber.
>> And could we extend this to other types of jobs that share those attributes?
>> Right. That need,
>> requires, um, manual dexterity in awkward circumstances. Like, if it's all routine, if it's a sort of modern house that was built from a computer plan, um, you can probably maintain it with robots easily. But if it's an old Victorian house where none of the angles are quite right angles, and things are falling apart, and you have to dream up a way of making it work anyway, I think it'll be longer before AI can do that.
>> Right. But not forever, because we're already beginning to see,
>> praise developments in humanoid robots these days, which can do, figure one showed the robot doing laundry, which is menial housework you might not even pay somebody to do.
>> Right. It's still not doing it as well as people, but it's getting there. We had a question actually submitted by a member of our community who's a software developer, and he asks, is there a part of human intelligence you think deep learning is not likely, is least likely to capture without some fundamentally new approach?
>> No, I think, I think we're machines. We're, we're wonderful machines. We're particularly wonderful to other members of our species who find us wonderful. Um, but we're just machines. Just is the wrong word because we are very different from a simple machine. We're wonderful, incredibly complicated machines. Um, but I don't believe there's any reason in principle why there's anything a person can do that a machine can't do. Things to do with the body won't happen until we have machines that have bodies. Um, but intellectual things, um, we can be very creative, but I think machines can be very creative too.
>> Wow. That's a very telling answer. Professor, you and I actually crossed paths almost exactly 10 years ago at an AI conference in Toronto by the Creative Destruction Lab. And I remember meeting a few of your students, including Ilya Sutskever, who had given a talk on OpenAI.
>> And I remember, I remember that meeting where Ilya was talking about scale and saying, "Look, you just scale it up, and it works better." And he was one of the first people to be preaching that very loudly. And he was basically right.
>> Yeah. After all these years, proven right. And I believe back then, all there was was a blog article about OpenAI on its site. And, uh, I remember the talk you gave, Professor, about neural networks, and the killer app back then was speech recognition and object recognition. And, uh, the breakthrough worth celebrating was getting error rates on, uh, down from 25% to 16%, and then through your lab's work, 5%. And then look where we are today.
>> Yes. Back then, we didn't know we were going to be able to do natural language. It was a big surprise to me and to everybody else that AI got so good at dealing with natural language, actually understanding what was being said.
>> Mhm.
>> And I'm just really curious. This was 10 years ago. Back then, could you have possibly foreseen how big this industry would be today? And the risks that you're talking about? You could have foreseen the risks if the industry got this advanced. It wasn't hard to see that when it got very advanced, there would be all these risks. What was very surprising was how quickly it got so advanced, particularly in dealing with natural language.
>> Right. And so that was the last 10 years, and so much has happened. Fast forward another 10 years, what do you see?
>> So, I think a good way to think about that is, um, let's make it 15 years. 15 years ago, like in 2010, if you'd asked people where AI would be now, and you said, "Would it be where it is now?" They would have said, "Absolutely no way. No chance." People like me would have said, even people who believed in neural nets would have said, "No, it's going to take longer than 15 years before we can have multimodal AIs that can see things, recognize things, um, answer any question you ask it in a reasonable way, um, give a caption to any picture you give it." That's crazy. That's going to be 50 years off. Um, well, we were wrong. I think if you now look 15 years in the future, um, we're going to be as wrong at predicting that. So, I don't know what's going to happen 15 years in the future. It's just possible things will slow down a lot, but I suspect they'll speed up a lot, and we'll get just things that now seem miraculous.
>> Right. And it reminds me of this recent paper from UC Berkeley. I love your take on this, Professor, where they polled almost 3,000 top-tier AI researchers, and they predicted about a 50% chance that all human occupations will be automatable sometime around 2100. Now, that seems like a very far time away, but like you said, it's very hard to predict even the next 15 years. So, I would actually, I would actually suspect there's a good chance all human occupations can be automated before that. I'd have said sort of 50 years was a better bet, and maybe sooner.
>> Wow. So that seems pretty,
>> Mathematicians, for example. Mathematicians, I think they're going to be out of business fairly quickly because mathematics is a closed system. It doesn't require data. So, you can have an AI. It's, it's like chess and Go. You can have an AI that just has one module that, um, proposes theorems, and another module that tries to prove them, and it can just keep learning lots and lots of stuff about mathematics. And I think, and many mathematicians now are beginning to think it may outstrip human mathematicians quite quickly.
>> Right. You made it a focus of your work recently to build awareness for these risks. How's that been going so far?
>> It's been going moderately well. Um, I think the thing that's gone best from my point of view is I was quite late to start talking about these risks. There were many other people, very good AI researchers, who've been talking about these risks for much longer, and I was somewhat worried. They say, "Well, you came to this very late. How come you're getting all this attention? You, I mean, we've been talking about this for years and years," and they haven't been like that on the whole. They're genuinely concerned about the risks. So, they've seen me as a spokesperson for what they believe in, and we've all got along just fine. Um,
>> I've been slightly surprised by that because academics are normally very picky. Um, and love to have disputes with each other, but all the people worried about risk are genuinely concerned about the risk, and they're more concerned about that than about academic credit, which is very nice to see. And certainly, somebody with your stature is very welcome in this conversation. So, thank you very much for, for your work.
>> Does it feel like policymakers are taking this seriously? Because if you're talking about,
>> They're beginning to. So, on the issue of jobs, for example, I've been talking to Bernie Sanders, and I go to Washington and talk to some more Democratic senators. Um, they, they're taking it very seriously. Um, on the existential threat, I've been talking to various leaders. Um, I went to China recently and got to talk to a member of the Chinese Politburo, um, who has an engineering background and understood the risks much better than I expected. Um, and I was emphasizing the existential threat, and he takes that very seriously. Um, in September, I'm on a committee to advise the Pope.
>> Um,
>> what's impressive there is they know I'm an atheist, but they still want my opinion.
>> Um,
>> So, leaders are beginning to take it seriously.
>> Right. There's more that makes us similar than different. And of course,
>> I should say within Canada, yeah, within Canada, we now have a minister of AI, and I've had several conversations with him, and he understands this point, and he actually thinks, yes, he understands the existential threat and thinks we should be doing research on how to manage it. That's very promising to hear. And I was going to say, of course, the Pope, and the Pope before him, has spoken about UBI, and not the newest Pope, I'm sorry, the previous Pope. And, uh, our current minister of AI hasn't said anything about that, but these are very important, high-stakes conversations that you're having, and, uh, it's, it's just very happy to see you doing this because I think this conversation is going to get much louder, and you're very much ahead of the curve. There's a part of me that wants to be an optimist, Jeffrey. And so I'm really wondering if it's possible to end on a positive note, right? So, if I was to just take the optimist worldview for a second and say, imagine we do automate away all the mundane labor, wouldn't that allow us to focus on the parts of life that, by definition, are not mundane? Maybe it could allow us to work towards shorter work weeks. Do you see some optimistic or even ideal trajectory that we take?
>> I see there's some hope of that in societies that are sort of decently run. Um, I also see that there's going to be tremendous improvements in things like healthcare and education from this, um, improvements in designing new materials, um, for things like climate change. So, maybe they, it can help develop room-temperature superconductivity or more efficient solar panels, and so on. Um, but there are going to be these terrible risks. I've become much happier since very recently I've started thinking about the idea of not going for a model of where we're dominant and this super-intelligent AI is submissive, and like an assistant, but the super-intelligent AI is like a mother, and we're like the baby, and it looks after us. Um, that seems a much more plausible model to me, and it's made me actually quite hopeful that we can get to that solution. And then we know mothers genuinely care for their babies, and they want the baby to do as well as possible.
>> And so, even if all the jobs are done by AIs, the, our super-intelligent mother will sort of help us, um, become good at writing poetry and writing plays and expressing our feelings and, um, interacting nicely with each other. Um, there's the possibility of a wonderful future.
>> I wonder how far we could take this analogy because a mother might want their kid to do even better than them, or, you know, this is a little bit of a cheeky analogy, but there's a, a lot of people rely on the bank of Mom and Dad right now, right? And so how far could we take this?
>> Um, we'll see. I mean, the thing to bear in mind is it's a funny period of history where we're about to get something, a super-intelligent AI that we've got no experience with at all. We're so used to being the apex intelligence, most people just can't even think about us not being the apex intelligence. Um, so there's huge uncertainty here, and so everything I said is speculation. We really don't really know what's going to happen, but clearly, we should be thinking about what can we do now to increase the chances that we'll get a good outcome.
>> Very prudent, very ominous. Dr. Hinton, I'd like to leave a last word with you. Our audience represents a large cross-section of Canadian and US society, mostly, including politicians, business leaders, academics, and social advocates. Do you have a message that you'd like to leave with everyone?
>> Um, what I've kind of specialized in since I've been talking about safety is this existential threat. So, I don't want to downplay all the short-term threats caused by bad actors like cyberattacks and nasty viruses and fake videos and lethal autonomous weapons. Those are all very important things. Um, but I do think the biggest threat of all, although it's longer-term, is of AI getting smarter than Earth and just taking over. And the message I'd like to leave you with is we should be doing everything that's possible now to see if we can envision a future and bring to pass a future where we can coexist with super-intelligent AI. And I don't think the way to do that is to think that we're going to be dominant and we're going to keep the super-intelligent AI submissive and subservient to us. We need to think of other possible models.
>> Sounds like you just painted out a road map for humanity. That honestly sounds very optimistic if we can find a way to do that. Thank you, Dr. Hinton. I appreciate you joining today.
>> Thank you for inviting me.
>> And to all of you tuning in, thanks for being a part of this very timely conversation with Dr. Jeffrey Hinton. I encourage you all to ponder what you've heard today, and if any of it resonates with you, how we can each play a role to propagate these ideas and help influence this and the next generation of leaders. Thank you, and goodbye.
>> Goodbye.