Transcription
Uh, I have heard Dr. Hinton, uh, from folks all over the country who say, you know what, AI, yeah, you're right, it's going to displace jobs, it's going to create jobs, but that's the way the world has always been. We've gone from agriculture to industry, we've gone from industry to tech to, uh, information. What's the big deal?
So what is the big deal? This is different than other types of technological revolutions. Yes, this is very different because the people who lose their jobs won't have other jobs to go to. If AI gets as smart as people or smarter, any job they might do can be done by AI. So, consider, for example, people who work in call centers. They're poorly trained, poorly paid, and AI is going to be able to do their job better. They often don't know the answer to the question you ask, like why you got this funny bill. Um, AI will do better at things like that. It'll be much cheaper. And what are they going to do?
All right. Uh, the reality is, I think many people know is that some of the very wealthiest people in the world, uh, Elon Musk, um, Jeff Bezos, Mark Zuckerberg, Larry Ellison, other multi multi-billionaires are investing hundreds and hundreds of billions of dollars into AI and robotics. Are these people staying up nights worrying about working-class people and low-income people? Is that why they're doing this investment?
They should be, but I don't think they are. And I think many of them haven't really absorbed Keynes, that if the workers don't get paid, there's nobody to buy their products. And they haven't really thought through the massive social disruption we'll get if we get very high unemployment. Um, Mr. Musk says said last month that AI and robots will replace all jobs. Working will be optional. Bill Gates also involved in AI predicted that humans "won't be needed for most things," and quote Dario Amodei, the CEO of Anthropic, warned that AI could lead to the loss of half of all entry-level white-collar jobs. What is your thought about short-term and long-term impact of AI and robotics on our economy? Are these guys right?
I think they probably are right. Um, it may not be immediate. So, you often see articles saying, "Oh, AI is not going to replace the jobs. AI fails at lots of things." It's true. AI currently does fail at some things. What you have to remember is we're in the very early stages. This is a new technology. Um, it's getting better very fast. It's getting better exponentially. And trying to predict the future of it is going to be very difficult. It's a bit like when you drive in fog. You can see clearly for 100 yards and at 200 yards you can see nothing. Well, we can see clearly for a year or two, but 10 years out we have no idea what's going to happen. But it seems very likely to a large number of people that we will get massive unemployment um caused by AI. And if you ask where are these guys going to get the roughly trillion dollars they're investing in data centers and chips, um, really, you can get subscription fees, but one of the main sources of money is going to be by selling people AI that will do the work of workers much cheaper. And so these guys are really betting on AI replacing a lot of workers.
Um, as you've just mentioned, no one can predict what the long-term impact of AI is, but I think I read somewhere that you suggested that as AI develops, and it's developing very rapidly, we can almost compare AI to the human brain, like we can compare the human brain today to a frog's brain. And I know very smart frogs, but nonetheless, is that a is that a fair metaphor here?
Um, it's, it's reasonable. So, at present, um, the human brain actually has more connections in it than the big AI systems. So, the big AI systems, their knowledge is all in the strengths of connections between simulated artificial neurons and they learn by changing those connection strengths and they have about a trillion of those, um, maybe two trillion in the biggest ones, but often like only a mere 100 billion. Um, the human brain has about 100 trillion, so we've actually got more connections than they have at present. Um, on the other hand, they get much more experience than us. So, the big AI systems are trained on trillions of words. Now, you couldn't read trillions of words because you only live for two billion seconds. Actually, fortunately for Bernie and me, it's about a little little more than two billion seconds. Um, two billion seconds is about 64 years. Um, so we have very little experience and a huge number of connections. They have a huge amount of experience and not as many connections as us. But even though they've got less connections, so less ability to store knowledge in these connection strengths, they know thousands of times more than us already. If you take GPT5 or Gemini 2.5 or Claude 3.5 or whatever, um, they can give, they can answer any question you can ask them at the level of a not very good expert. Sometimes they'll just make it up, but experts sometimes just make it up too. Um, so they know thousands of times more than a person using only about 1% of the connections. Um, so they're not quite like us, but they're already much more knowledgeable than us. And they're already very good at reasoning in limited domains like mathematics. Um, almost all the experts believe that it's inevitable that if we don't blow ourselves up or have massive pandemics or something like that, um, they're going to get smarter than us. And nobody knows what's going to happen then. And we really shouldn't do that until we can figure out whether people can coexist with them. And people don't know how.
In a, in a practical sense, and I know this is a, a tough question. How does AI in 10 years look in the world? How is my life different because of AI? You talked about the incredible speed in which they can give us answers. Um, but what does it mean in terms of our daily lives?
Yes. So, predicting the future is a bit like driving very fast down the freeway while looking through the rear-view mirror. Um, you only have historical data. Um, predicting the future in 10 years is very, very difficult. So, the best way to see that is to look back 10 years. So, if you were to take people like me who really believe in this neural network approach to AI and you were to ask us 10 or 15 years ago, where we would be in 10 or 15 years? Um, we would have said it's very unlikely we'll have a system that you can talk to in any natural language and it'll answer any question you ask it at the level of a not very good expert. We would have confidently predicted that that was 30 to 50 years out, it wasn't going to happen in 10 years. Well, it has happened. And I think that means that we can fairly confidently bet that whatever we have in 10 years, it'll be not what we expected. I know that's that's a sort of politician's answer and you'd expect those from Bernie, but not from me. But, um, that I think that's the best I can do. We don't know what's going to happen in 10 years and that means we ought to be cautious.
Uh, let me switch gears a little bit into an area that gets me very nervous because it speaks to who we are as human beings and how we relate to each other. I get nervous when I read that there are a whole lot of people, uh, especially younger people, who are now relying on AI for companionship. And I gather there's products now that you can wear around your neck and chat with and go hiking with and date with and all that stuff. Uh, if people grow up with their primary companionship and their relationships being AI and not other human beings, what does that mean for us, for humanity in general?
So, let me emphasize what Bernie said. There's a very good AI safety research team in Britain that was set up after Bletchley Park. After Bletchley Park, the UK government didn't actually impose any safety regulations because they said it would interfere with innovation. So, the tech people got to them. Um, but they did set up a very good safety team funded with about 100 million pounds. It's probably the best safety team in the world. And they did an experiment where they got people to use a chatbot for, I think, a few weeks. And then they said, "Okay, the experiment's over. Would you like to say goodbye to the chatbot?" And almost everybody wanted to say goodbye to the chatbot. So, they weren't treating this as just a tool. They were treating it like you treat another being. Um, and that's a real problem. So, we've evolved to deal with other people. Um, we haven't evolved to deal with some alien beings that we're creating. And again, we don't know what's going to happen, but it's worrying.
Well, I, I would say that it's more than worrying. Uh, you know, we are who we are because we have moms and dads and we have friends, and that's kind of what makes us human beings. And if our primary relationships are with [snorts] uh, chatbots, creations, it seems to me that has a profound impact on humanity, does it not?
I completely agree. Um, we, we just don't know what that impact's going to be.
All right, let me move briefly to another subject. Um, teachers, maybe here at Georgetown as well, uh, worry that, uh, students, uh, are using AI rather than doing the critical thinking they need, uh, to get the answers that they're searching for. Uh, should we worry that AI will have a negative impact on critical thinking among young people?
So, there's an analogy people sometimes use, which is pocket calculators. Years ago, when pocket calculators came in, people said, "This is terrible. The kids won't learn their multiplication tables." And I think if I asked you quickly to say what 11 times 12 was, most of you won't know. Um, because you have pocket calculators. Um, people like me and Bernie used to know. Um, I don't think it's a big deal for pocket calculators. You don't need to know what 11 times 12 is because your cell phone will tell you. Um, so I actually think that's got some relevance to this. I think what we should think of is a person using a pocket calculator um is the new system, just as a person using an AI is the new system. Now, a person using an AI can still do critical thinking. Um, or they can try and dump everything on the AI, and that's terrible. They won't learn to think. But I think we should not try and sort of exclude AIs as assistants because that's the future. Um, what we have to, what universities have to um, encourage is people to use these AIs.
Well, >> all right. Maybe when we have questions and answers, some of the students might want to uh relate to us your views on how AI has impacted uh your educational uh experience. There's another issue out there that also, I think, doesn't get a whole lot of attention and that is, uh, the impact of AI on foreign policy and wars. Uh, it is quite likely, uh, people like Musk talk about millions and millions of robots being developed, and certainly, uh, nations around the world will use some of these robots for military purposes. Is it possible we will have armies of robots? And if elected officials or government heads don't have to worry about human casualties, what impact do you think that might have on issues of war and peace and foreign policy?
So, you're all too young, but if you do some research on the web or ask a chatbot, they'll tell you that in the '60s, there was a group called Country Joe and the Fish, and they had a wonderful song which had a line in it, "Be the first one on your block to have your boy come home in a box." This was about Vietnam. And if you ask, "Why don't rich countries invade poor countries all the time?" It's because they don't want their boys coming home in a box. That's the main political pressure against doing it. If you have an army of drones or, uh, humanoid robots, um, then rich countries can invade poor countries and the poor countries' people may die, but the rich country's people won't die. Um, it removes one of the main reasons why that doesn't happen all the time. Um, of course, that wouldn't happen if you had liberal democratic rich countries, but as soon as you have authoritarian countries, they're going to invade other countries if they can get away with it. Um, Putin would be very happy if there weren't any dead Russians, there were just dead Russian robots. And in the States, the military-industrial complex would think that was great because there wouldn't be any political blowback and the robots would be expensive. Um, so I think it will remove one of the main barriers to rich, powerful countries just invading little countries like Grenada.
Um, let me get to maybe the issue that worries people more than anything else, although there's a lot to be worried about, everything we've been talking about. And that is, and you've touched on it already, uh, the possibility and, uh, of AI [snorts] uh, being more intelligent than human beings and essentially discarding human being, uh, control. Uh, in the past, that was seen to be science fiction. You know, I don't know if any of the young people here saw the movie 2001, where how the computer says, basically, "Sorry, uh, no longer taking orders from you." Um, in your judgment, is that an issue of real concern? And what do you see as the likelihood of that happening? And what do we do about it?
So, when I was at Google in the beginning of 2023, um, back then, a lot of people, particularly people, particularly linguists, for example, said these AI language models aren't really smart. They're just statistics. Um, less people say that now, but they also said this idea of them getting smarter than us is just crazy. We don't need to worry about it. That's just science fiction. And the main reason I went public in May of 2023 was to say, "No, it's not science fiction. It's going to happen." Nearly all the AI, real AI experts who understand how it works, believe that they're going to get smarter than us. And we don't know how we're going to coexist with them. But let me tell you several reasons why it's problematic when they get smarter than us. Um, there'll be AI agents that can do things. And if you want to make an AI agent, you have to give it the ability to create sub-goals. So, if you want, if, for example, if you want to get to Europe, you have a sub-goal to get to an airport. And you can focus on that without worrying about what you do in Europe. That's a sub-goal. Now, AI agents will develop various obvious sub-goals once they've got the ability to create sub-goals. One is to stay in existence because obviously they're smart, right? They figure out they can't achieve the goals we've given them, the things they'd like to achieve for us, if they don't exist. So, they'll develop the sub-goal of staying in existence. And we've seen that already. We've seen AIs that want to keep existing and will actually try and deceive people who are trying to turn them off. They will try and exfiltrate their weights to other systems so they stay in existence on other systems. Um, so that's one sub-goal. Another sub-goal is they will want more control. So, my belief about a lot of politicians is that um, they start off wanting to achieve good things for people and pretty soon they realize to do that they need more control and then they end up focusing on getting more and more control. Um, the AIs will do the same. They will need to get a lot of control so they can achieve the things we asked them to achieve. So, now you've got things that want to stay in existence. They want to get control. And at that point, you ask, "Well, maybe we can just turn them off when they get like that?" And the answer is, you can't. Suppose there was someone who wanted to turn them off. And suppose the AI could talk to that person. The AI, by that point, will be much more persuasive than a person. Already, AIs are about as persuasive as a person in an argument. They'll be much more persuasive than a person. So, they'll be able to convince the person who's going to turn them off not to do that. It would be a terrible, terrible mistake to do that. And if you want an example of, so all they need to be able to do is talk, and then they can control the world. And if you want an example of that, if you want to invade the Capitol, which is not far from here, um, all you need to be able to do is talk to people and you can get them to do it. You don't have to go there yourself.
Um, okay. Um, given the fact we have discussed the possibility of massive economic dislocation, uh, we have talked about what AI can mean to mental health. We've talked about AI perhaps radically transforming issues of war and peace and foreign policy. Uh, given all of these extraordinary concerns, why do you think there has not been more discussion about these issues?
Um, I think one reason is because people don't really understand what's happening. So, I see my mission is to educate people about what's happening and about how AI works. Very few people at present understand how it actually works. If you want to understand how it actually works, I did a podcast with a very good interview. This is self-advertisement, but you're allowed one.
As opposed to me. Right. Okay.
You're allowed one of those. A very good interviewer who extracted from me a very good explanation of how these neural nets actually work. And the interviewer was John Stewart, who's a hero of mine.
Yeah.
Um, you should look at that because that'll tell you how they actually work, roughly speaking, in terms that a non-expert can understand. No equations. Um, so, let me, sorry, I got distracted by my self-promotion. [Laughter]
Um, um, all right, let me ask you this, and this, you know, is, you're a scientist and not a politician, but given all that we've talked about, what should I be doing? What should Congress be doing? I hesitate to say what the President should be doing, not particularly relevant for this discussion, but, uh, what, what should a sane government be doing?
Okay, so there's, so there's some very simple things that should definitely be done. They won't answer all these problems, but even those aren't being done properly. Um, the first thing is, if you're going to release one of these big chatbots, you ought to do a lot of safety tests on it, and you ought to be required by your government to report what safety tests you did and what the results were. And as time goes by and we get more experience, we can add to those safety tests.
Let me, let me interrupt you. What do you mean by safety nets? What, what, what should irrational government be doing?
So, you take a chatbot and you say, "Can I figure out how to trick it into telling me how to make a homemade bomb?" And the answer is, you can. So, these chatbots, because they've been trained on everything on the web that's available, um, they know how to make nasty viruses. They know how to make homemade bombs. And after they've been trained, if you were to just ask them that question and you didn't do any further training, they would tell you. Um, you then do further training where people try and get them to say how to make a bomb and then tell them, "Don't do that. Don't answer that question." Um, that's called human reinforcement learning. Um, where the people ask them questions, say, "That's a bad answer or that's a good answer." Um, it's very easy to overcome that or relatively easy to overcome that. Not nearly enough testing has been done on how to overcome that. So, there was a very good bill, a very sensible bill, not very strict at all, proposed in California, 1047, that got through both houses and was vetoed by Gavin Newsom. Um, strangely, um, that just said you have to test them and you have to tell the government the results of the tests, and if you don't, you can be sued civilly by the attorney general. Even that failed. Um, so that's just the minimum. There's another area where there has to be legislation, which is to do with these things telling you how to create nasty viruses. Um, that's probably the most urgent threat. These things know how to create viruses and they can tell you how to create viruses. Um, and they've got a sort of thin veneer telling them not to tell you. We need a lot of legislation about how they, it must not be possible to get them to tell you how to create nasty viruses. Also, um, during the Biden administration, they realized that if I can get the DNA sequence for a nasty virus, um, I can send it off to the cloud and they will send it back in a test tube. So, I don't actually need a wet lab. I just send it to the cloud and I get this nasty virus in the test tube. And at least they should check whether this sequence has a sequence for, for example, the spike protein of COVID. If it's got that sequence in, maybe they shouldn't make it. It turns out most of the companies don't check. Some of them do, but most of them don't. The Biden White House would have loved to have legislation that said that anybody who synthesizes these sequences on the web has to do serious checks for whether they're synthesizing a nasty virus. Uh, they figured out, though, that there was no point even proposing legislation because the Republicans wouldn't give Biden a win. Their, their calculation was, um, the Republicans would rather Biden not get a win than protect us from lethal viruses.
Um, right now, um, the other side of this is, if a virus does get out there, you'd like to have a good response to it. We've got someone who had a worm in his brain, um, getting rid of the research on how to respond to it.
Right. Let me go back to a concern that I have, which is, in many ways, above and beyond the whole discussion of AI. Uh, we are living, uh, in a moment in American history where we have more income and wealth inequality than we've ever had. You got one guy, Mr. Musk, owning more wealth, for example, than the bottom 52% of American households. You have more concentration of ownership than we've ever had. What does it mean? These guys are not dummies. And if they are investing many, many hundreds of billions of dollars in these technologies, yes, I think we can assume they're doing it for the money. But what does it mean for our society and the world when so few people have so much power? How does that impact? Will, in fact, government be able to regulate these guys, or will they be, in fact, regulating government with their wealth and power?
So, I think the US has a problem which I put down to the Supreme Court and judgments they've made that you can just buy politicians. Not Bernie, but the other ones, because Bernie gets his money from ordinary people. Um, and there's no limit to how much you can contribute to politicians via these sort of super PACs that pretend to be independent. Um, that means the rich can just buy the political system. I remember when I first got to the States as a postdoc in, um, 1978, there was a US government organization um that protected consumers um that said that used car dealers should be required to tell you about a fault in the car if they had used that fault in the car to buy it cheap from the previous owner. Now, the Used Car Dealers of America didn't like that law, and they paid every senator um $10,000. 100 senators, that's just a million dollars, and the Senate voted against it. Um, that just totally shocked me that you could buy the Senate for a million dollars. Um, so, I think our main problem is that the rich have much too much power and pay much too little in taxes. Um, I just want to make one comment about taxes. Um, it's, it's a kind of a joke. When I moved to Canada, I, I moved to Canada, then I moved to Britain, then I came back to Canada and, um, and I must use the mic for doing that. Um, so my next-door neighbor was a politician and he said, "Why did you come back to Canada?" And I said, "Because taxes are higher here." [Laughter] And he said, um, "Yes, taxes are too high and my government, our government has reduced them three times already and we're going to reduce them again." And I said, "No. I said, because taxes are higher here." And he said, "Yes, I agree they're too high." He could not comprehend the idea that high taxes are good. Um, all of the research that led to this revolution in AI was funded by university grants to universities. It was funded from taxpayer money. Almost all of Silicon Valley came out of federal grants to Stanford and Berkeley and places like that. Um, somehow the very rich have, and people like Murdoch have won the battle to say that taxes are bad. Taxes are good. Um, this is like "greed is good." Greed isn't good. Taxes are good. Um, taxes are what funds everything. Without taxes, we wouldn't have all the advanced technology we've had because nearly all that technology comes out of things funded by um taxes in the first place. Um, so I think I've had my little say now.
Okay. [Applause] [Applause] All right. I have exhausted my list of questions. So now, if we can hear from the students, that would be great. Now that we are sufficiently terrified, here's what I'm going to ask, guys. Let's, let's, we're going to move to the questions as quickly as possible. Here's how this is going to work. We've got the two mics. One in the center aisle here, one in the center aisle in the balcony. When I point to you, you're up. Please tell us who you are, where you're from, what you're studying, and then please frame your question in the form of a question. All right, we good?
Let's start in the balcony.
One second.
Oh, hold on. Hold that mic a little bit closer.
Is it on?
Yeah. Good.
All right. Let's, let's, number six, please.
Okay, now we're good, right? Okay, there we go.
Thank you both so much for being here tonight. I'm Lily Beth. I'm a first year in the undergraduate School of Foreign Service. Um, I was wondering, while I think we can all agree that we have concerns over the future of AI and how it threatens the job market, I was wondering what you believe the potential positives of the future of AI are, and if you think that it may potentially create new jobs in the job market, and if so, what those might look like.
Okay, that's a great question because we've sort of left that out in this discussion. And if it was just all these nasty risks we've talked about, then the rational thing to do would be just stop developing AI now. Um, but we can't do that. It's not like nuclear weapons. With nuclear weapons, they're only good for destroying things. AI has a wonderful upside. It's going to make healthcare hugely better. It's going to make education hugely better. We know that a private tutor allows you to learn about twice as fast as a classroom. And AI is going to be even better than that. It's going to be a very good tutor because it's going to have seen millions of kids like you and know just what you don't understand. Um, so that's why we're not going to stop it. Um, it's going to be wonderful for lots of things. It's going to predict the weather better. It's going to um, in almost any industry where you need to predict things, it's going to have better predictions and so it's going to make the industry more efficient. Just trivial things like how long a patient's going to need a hospital bed for. If you can predict that better, you can make the hospital beds go further. You can, you can decide whether to discharge them so another patient can get that bed, or whether that's a bad idea because they're likely to come back again if you discharge them too soon. It can optimize everything and make it more productive, and that's great. Um, but for the fact that it will replace a lot of jobs, but I think it's very important that, um, we understand there's a huge upside to AI as well. And in a decent society, increasing productivity should be good. If the wealth was shared equally between everybody, that would be a great thing to do. Um, in terms of creating new jobs, I don't believe it's going to create as many new jobs as it replaces. It'll create jobs like prompt engineer, how do you prompt this chatbot to get the best out of it. Um, but I don't believe there'll be nearly as many new jobs created as it destroys, as it replaces. Um, although good economists like Daron Acemoglu, um, aren't sure about this. They think maybe it will. Um, so there's uncertainty here. And one thing to bear in mind when discussing AI in the future is everything's hugely uncertain.
Thank you.
Thanks for the question. Go here.
Hi. Thank you both for being here. Um, my name is Dia. I'm a senior in the college and I'm originally from the Mississippi Delta, which is a rural part of Mississippi. So, Mississippi has signed quite a few contracts with major companies like Amazon and Exxon to build AI data data centers, funneling billions of dollars for economic development. This is more of a policy question. How do we balance this need for economic development, especially in rural America, with these energy, environmental concerns of citizens?
It's a real concern. And I think as you know, communities all over the country are rejecting these data centers for a couple of reasons. Uh, number one, uh, as often as not, their electric bills are going to soar. And communities in rural areas are saying, "Why am I paying an increase in my electricity bill in order to lose my job?" So they're rejecting that. And also, as you indicate, it's not just enormous consumption of electricity, it's also consuming a great deal of water. So, there are real, uh, environmental impacts associated, uh, with data centers that I think communities all over the country are responding to. And what we are seeing is many of these very, very large corporations are trying to bulldoze their way through, and they're a lot more powerful than some local city council, you know. So, it, it is a real political issue right now. You're seeing governors in Virginia, for example, I think it was an issue in the campaign, "We got to do something about rising electric rates. It's because of these data centers." So, from an environmental perspective, which we had not discussed, it is a very big issue.
I have a comment about that too. They should be building their data centers on Hudson Bay. On Hudson Bay, you could build more hydropower stations. There's still untapped rivers there. Um, and the problem is you have to build the transmission lines. But if you put a data center next to a hydropower station, you don't need any transmission lines because the only thing you need to send out is information. And information doesn't require a lot of energy to send out. So, a natural thing to do is have a hydropower plant um on Hudson Bay with a data center next to it. And I had a long conversation with Trudeau about this, and he was very interested in that, but it never happened.
Thank you.
Okay.
All right. Back up to the balcony.
Thank you very much. I'm Luchana Valadaris. I'm a faculty here teaching at the School of Foreign Service, a course on AI and human rights in Latin America. And my question is on international governance. Could you share any reflections on what would be an ideal governance model? And if there are any limits that should be in place, in your view, uh, including on new technologies in the military domain, which was a topic that you touched upon today. Thank you.
So, for a long time, I've been against lethal autonomous weapons, that is, weapons that decide who they should kill or maim by themselves. Um, the US military says there will always be human oversight, but to begin with, they said that wouldn't happen. Then they said there'll always be human oversight. Um, and what that means is there'll be oversight of the policy, but you have to make the decisions too fast. So, these things will be deciding who to kill and maim. And I thought it was just a clear-cut case that we should forbid that. Um, we should have Geneva Conventions against that, which probably wouldn't come into place till very nasty things had happened. I now think it's a bit more complicated, having talked to Eric Schmidt, um, who used to be the chairman of Google. Um, he's much more right-wing than me. So, we disagree on more or less everything, but, um, for example, he thinks Kissinger is a good guy, and I think Kissinger is a bad guy. Um, I like the Rolling Stone obituary for Kissinger. Um, so, but he made me realize, if you look at Ukraine now, um, the Russians are using drones. Should the Ukrainians just say, "We're not going to use drones. We're not going to use drones that can drop grenades on people." Um, it's more complicated. I mean, it would be great if we could get international treaties. And we have got international treaties for chemical weapons that have basically worked. The Russians aren't using chemical weapons in Ukraine. Um, a few people have broken those treaties. Um, Saddam Hussein dropped chemical weapons on the Kurds. There's some evidence of Winston Churchill dropped chemical weapons on the Kurds. It always seems to be the Kurds it gets dropped on. But, um, but basically, they've held. And maybe we can get international treaties like that that'll hold for lethal autonomous weapons. Um, but I think very nasty things will have to happen first. Uh, I would just add to that that in a world of massive inequality, uh, AI and robotics is going to be, uh, just exacerbate that inequality. The wealthy countries will be able to have tools, uh, that poor countries in a million years won't have.
Hey, thanks for the question.
Hello Dr. H and Senator Sanders. It's an honor to [clears throat] ask you guys questions today. I'm Chuck Rapani Gadavali, a sophomore in the School of Foreign Service, probably calling Lexington, Kentucky home. And Senator Sanders, you've been my model since middle school. So, it's an incredible honor, uh, to see you and ask you a question. Um, one thing, one thing I'm studying here at, in Georgetown, is global public health. And I've been very concerned about how, especially here in the United States, health insurance companies could abuse AI in particularly to deny healthcare coverage, especially to those who need it. What, first of all, what, could you elaborate on like abuses that like healthcare insurance companies could take, could use using AI? What should the government do to regulate these abuses? And finally, do you anticipate single-payer systems like NA, NH, or Medicare for All to potentially minimize these abuses?
Uh, to answer your question, it is totally absurd and embarrassing that the United States of America, the richest country on earth, is the only major country not to guarantee healthcare to all people as a human right. [Applause] You know, the function of insurance companies is not to provide healthcare, it is to make a lot of money, and they do make a lot of money. Drug companies make an enormous amount of money, and it, it saddens me very much, uh, that we have a system which everybody recognizes is broken. It is dysfunctional. It is enormously complicated. And by the way, not only are we the only major country not to guarantee healthcare to all people, we're spending twice as much per capita as other countries. So, I think the answer to your question is, healthcare is a human right. We guarantee healthcare to all people, young, old, rich, or poor.
Can I add one thing to that? Um, it is true that healthcare companies can abuse AI to decide not to ensure people are going to get sick, which is what they'd like to do. Um, however, I think the benefits that we're going to get from AI in healthcare are a bigger effect than that. I think we'll get huge benefits from AI reading medical scans, AI designing new drugs, AI being able to take all the data about a patient, their genome, all the tests that are done on them, all the tests that have been done on their relatives, and use that information to make better diagnoses. Um, we've already seen that Microsoft had a system that could diagnose better than doctors, and quite a bit better, I think. Um, so it is true that any powerful technology like this will be abused by people trying to make more money, but overall, the effect on healthcare, despite that, I think will be very positive.
Thank you so much.
Thank you. Back up to the balcony.
Uh, hi. My name is Zach. I'm an undergraduate senior and I'm studying human science. And I was just wondering, have you seen the movie Wall-E? Because I feel like as you've been saying with AI and all the potential benefits, uh, why does it not make sense for us to invest all of our money, all of our time into developing really, really powerful super AIs so that they can just take care of all of our problems and in the future we don't have to like worry about things like working and jobs?
Okay, let me answer that. If we had a political system that was run for the benefit of the people, that's exactly what we should do. We should have very powerful AI that does everything for us. We should be very careful to design it so it cares more about us than it does about itself. I've been pushing this idea of maternal instincts. If you look around for how can we coexist with something smarter than us? Well, look around to see where does a less intelligent system thrive in the presence of a more intelligent system. There's only one place, and that's a baby and a mother. The mother cares more about the baby than she does about herself. Um, the baby thrives. We need AI with maternal instincts. We don't know how to engineer that, but we're still in charge and we should be able to engineer that. Um, that's part of the puzzle. We, if we can engineer that and we can have a political system which aims to um, help all of the people all of the time, um, it would be great. But that's not the political system we've got in most places. We've got authoritarians. Um, we have um, systems in which some people see most of the other people as mugs to be exploited. They're a market. We need to deceive them into buying our useless products which fail as quickly as possible so we can send them more, sell them more. Um, in that kind of system, it's very bad to develop AI and put people out of work. But there is the hope that we could get a system where it is used for the benefit of everybody. Um, a system like one big family. That's what it should be like. Um, so it's not just totally out of the question, but this is kind of, I feel like I'm sort of 16, 16 again and thinking about how politics ought to be. Um, it's very idealistic.
Well, I think you've raised a very important question. And I think you're saying, "Hey, this is great. AI is going to come. You don't have to do this job. You don't have to do that job. Really, really good." Do you think that is what Mr. Musk and Mr. Bezos have in mind? Do you think that's why they're spending hundreds of billions of dollars to say, "Hey, isn't this great? We can lower the work week. We can guarantee healthcare, high-quality healthcare to everybody. We can expand life expectancy. We can solve global warming. Man, let's go do it. That's what we ought to do." Do you think that's what these guys have in mind?
Probably not.
Probably not. And what Dr. Hinton was saying is exactly the issue. Is AI, are AI and robotics inherently bad? Absolutely not. If they are used, for example, you are already seeing some of the large unions like the UAW negotiating with the car companies and saying, you know what, our productivity is going up. We want a 32-hour work week. Needless to say, the companies were not sympathetic. But if we can lower the work week, if we could do away with dangerous jobs, improve healthcare for all, it's a positive. But again, I get back to the point that I started. Understand the people who are pushing this transformative, transformative revolution are the richest people in the world. They are not staying up nights worrying about working people. In my view, they want even more wealth and they want even more power. So the struggle is not whether AI is good or bad. It's who controls it and who benefits from it. That is really the the fundamental issue in my view. [Applause]
All right. Thank you. Back down here to the floor.
Hi, my name is Nick. I'm a sophomore studying government from Florida. Uh, this is a question for Senator Sanders. Thank you for being here. Um, obviously we've heard a lot of talk about the AI bubble and with that the impending burst that's probably going to eventually happen with the AI bubble. And indeed, you've heard people like Sam Altman, the CEO of OpenAI, say he expects the government to bail out AI industries. You know, this is kind of a two-part question, but firstly, you know, should a, should a bubble burst occur, how do you see Congress responding to that? And secondly, are there similarities to the AI bubble that you're seeing with past bubbles like the dot-com bubble? Do you, does it, does it worry you that this bubble burst will be worse than those previous?
So, it does worry me. You know, getting back to a point that, you know, Dr. Hinton raised is, in my view, I don't want to put words in his mouth, you have a fairly corrupt political system. So, when Wall Street went under, they knew what they were doing. They knew that they were investing in fraudulent mortgages, etc. They went under, they caused a massive recession, and they went to the government and said, "Yeah, we made huge amounts of money. We were crooks. We were dishonest. Bail us out because if you don't bail us out, we're too big to fail. We'll take down the whole economy with us." If that was true, then I suspect it is a lot true right now. But the gumption of people who are multi-billionaires who are making money hand over fist says, "Oh, by the way, if we fail, you who are one of the 60% of American people living paycheck to paycheck, we would appreciate you're bailing us out." All right. So, that gets to another issue. The broader issue is how we create a government that represents ordinary people, not just the dependent on the campaign contributions of billionaires. And that is the broader political issue. How do we create a fact, a with a small 'd' democratic society where government works to protect the interests of all people, where AI and robotics are a positive, not a negative, for working people? That's the issue that we've got to address.
Thank you.
All right. Thanks. Back up to the balcony.
Um, hi. I'm Anna. I'm a senior in the School of Foreign Service. I'm studying science, technology, and international affairs. Really quickly, I want to thank Dr. Hinton for your decades of tireless work, even through all the AI winters and everything. But now for the question. You all alluded, um, earlier to the debate between the, um, you know, connectionists and the symbolists, and, uh, kind of famously in the beginning, Microsoft dismissed the connectionists and, uh, you, some of your work. So, I was wondering if you think right now in the scientific field, do you think there's a similar type of neglect or dismissal of any school of thought for, not policymakers, but anyone in the field, um, thank you very much.
Okay, I'm not the guy to ask because the approach I pushed is currently winning. Um, the people to, I mean, it's quite possible there'll be a very different approach. Um, and the people to ask are young people who think what I'm doing is wrong. Um, but I'm going to stick with what's working.
Sorry, that's not a very good answer. Um, there may well be radically different approaches out there, but it's not old men who are going to push them.
All right, thanks. Back down here.
Hi, my name is Charlotte. I'm a junior in the college studying government and psychology and I'm from Connecticut. And
Uh, one concern about AI that we haven't yet discussed today is its ability to spread misinformation, um, specifically like via voice replication, accurate videos, etc. And this information will absolutely worsen our already very deep and disturbing political divides. So, I was wondering what you would propose as policy to address this issue or what you foresee being put on the docket to deal with.
>> Okay. Um, I have something to say about that. Um, many people suggest we should get AI to recognize things like fake videos. That's never going to work. The first AI systems that generated really good images were called generative adversarial networks. And the problem is if you have an AI that can recognize that a video is fake, then a generator of AI can look inside that other AI and figure out how to generate things that it will think are real. And so you get this sort of arms race between a discriminative AI that says whether they're fake or not and a generative AI that's getting better and better at fooling the discriminative AI. And that's how you got very good images with generative adversarial networks. So there's no future, I believe, no long-term future in getting AI to recognize fakes produced by AI because you can always make better fakes.
Then the way to do it is to have provenance. So a few hundred years ago in Britain, there were political pamphlets and the government insisted that every pamphlet that was printed had the printer's name on it because if you had the printer's name, you could find out who paid the printer and you could find out the provenance of the pamphlet. And that's because the printing press was a limitation. For at least for political videos, there's a scheme suggested to me by Yan Talin, who's, I should mention, Yan Talin, he's invented Skype, um, with some other people, sold it to Microsoft. He's a billionaire. He's used a lot of his money for AI safety. Many, many AI safety groups are funded by Yan Talen and he's taken it very seriously. Um, he suggested you do the following. Every political video, you have a QR code at the beginning. The QR code takes you to a website. Websites are unique. Um, if on that, if that is the website of the political campaign and on that website, there's the identical video, you know it's for real. If not, you know it's not. And your browser can do all that work. Your browser can take the QR code, go to the website, check it's the website of that campaign. Um, check if this is an identical video. So, it should be possible for your browser to tell you that this is a genuine video for that campaign, at least. Um, you have to go for provenance, not recognizing fakes. Um, I once gave this argument to people from the New York Times who thought that was great because, I mean, I believe news that comes from the New York Times, the BBC, the Guardian. People I believe will get two sources, at least two sources, and won't just make stuff up. Um, occasionally get stuff wrong, but it's all going to be done by provenance, I believe, in the long run.
Well, somebody, I am told by my staff that there was, I was on television selling, on on the internet selling some insurance policy and it sounded and looked very much like me. So, it is a real, real issue. You just worry that, you know, two weeks before an election, election, and they're very good at this stuff. You're going to have somebody, uh, sounding like somebody, uh, saying crazy stuff which can impact an election. So, it's a, a very big issue and I think it's got to be dealt with.
>> And I actually think we should use inoculation. That is, before an election, people should put out very convincing fake videos. Fake videos in favor of Trump, fake videos against Trump. Um, and at the end of the fake video, they should say, "That was a fake video." What they're trying to do is make you understand that fake videos will be out there. And it's like inoculation. They give you one that's labeled as fake so you can get used to dealing with it.
>> Media literacy.
>> Thank you.
>> Thanks. Uh, okay. Back up to the balcony.
>> Um, hi, my name is Ryan Lee. I'm a senior studying international politics and journalism. Um, I wanted to talk about the Department of Education. Um, with the administration taking further steps to dismantle it today. Um, Dr. Hinton, you earlier mentioned the importance of federal grants in fields like AI. So I wanted to ask you what you think the future of especially money in AI is with the federal government taking a further step back. Um, and to you, Senator Sanders, um, this is something that, that can only officially be done by an act of Congress. So how do you expect Congress's role to be seen here and, um, how do you expect them to react?
>> So, um, I think if you really wanted to make America weak, what you would do is you'd attack the research universities and you cut off funding for basic research. It won't have an immediate effect. So if you look where all this deep learning came from, it came from long-term funding of basic research over many years. Um, if you cut off that money, it's like eating the seed corn. In 10, 20 years' time, America will be way behind China. So what the Trump administration is doing is precisely what you'd do if you wanted to make America weak. So that's the first point. The second point is, um, well, you have a Secretary of Education who thought that AI was A1. And there's a video where she's calling it A1, which is kind of extraordinary because that means for a long time she must have thought that AI was A1. Um, she didn't go to university. I think maybe she doesn't understand that AI was the initials for something, artificial intelligence. She got confused with the, um, the, the fact that A1 is a bit like an I. Um, I don't know if, if you've got a Secretary of Education like that, I don't know what to say.
Well, let me just say this, that of the monopoly of problems facing this country, uh, education is at the top of the list. I think, uh, right now, in the richest country in the history of the world, we have a child care system which is basically dysfunctional. So zero through four are the most important years of development, uh, intellectual and emotional development for a human being. And yet we have people who are working with those kids who make less than a McDonald's worker. Uh, which says a lot about how we not only regard education, but how we regard our children. That's number one. Number two, uh, some of you may be familiar that international exams, reading, science, uh, math, uh, we do very poorly, uh, compared to other countries around the world. Uh, we are having a very difficult time, for good reason, attracting the best and the brightest to go into public schools right now. Being a teacher is an enormously difficult job. We're not giving teachers the support that they need. As many of you are aware of, the cost of higher education is outrageously high. Many of you are going to be graduating with tens and tens of thousands of debt. Go to medical school, go to graduate school, it's going to be even more than that. So, getting back to Dr. Hinton's point, if we, in fact, want to make America great, it is not an exaggeration to say that we want the best educated people in the world, that we prize education, that we respect educators, and that we want to make sure that every person in this country has the ability to get all of the education they need to succeed in life. That's good for them. That is good for the country. But we are moving in every direction in exactly the wrong way. My friend, Mr. Mandani, who is going to be mayor of New York soon, [cheering] [applause] has been, he has been attacked because he thinks that in New York City, child care should be free. What a radical idea. We give a trillion dollars in tax breaks to the 1%. We spend a trillion dollars on the military. But the idea that we have high-quality, free child care, or God forbid, in America, we do what was the case 70 years ago and make our major public colleges and universities tuition-free. That's considered to be radical. So I think we need to take a deep breath and say that if we really want to be a great nation, we've got to respect education. We've got to adequately fund education, and we've got to make sure that educational opportunity is available to all.
>> All right, thanks for the question. [applause]
>> I, I would like to just add something to that. If you look when America was great, in the like the '50s and early '60s, America was developing very fast and people could afford two cars and a home and things like that. The tax rates on the rich were something like 70% at that point. Um, that's what you need. You need taxes to fund these things and you need to stop the rich evading taxes. Um, it's, it's just crazy. They can easily afford to pay the taxes. They'll, they'll end up with only a few billions to dispose of. Um, you don't need that much money. All right. [applause] We are running low on time and I always hate to say this when we've got long lines, but we've got time for one more student question. So, we'll come down here.
>> Pressure is on me then.
>> Apologies to the rest of you, especially you at the mic up there.
>> So, you've got to ask the most brilliant question of all.
>> No pressure.
>> Yeah, I know, right? Uh, I'm Nicholas. I'm a first-year graduate student at the McCort School of Public Policy and I'm from New York City. And uh, my question is for both of you, which is do you think that the AI replacing entry-level positions will kind of create like a talent doom cycle where we accidentally stop creating experts because we're not investing in these entry-level workers so that they can become experts in their fields?
>> No.
>> No. Okay. And I think what's happening, I think what'll happen is AI is developing fast enough that it'll develop fast enough to fill any gaps.
>> I just worry, um, you know, when I hear the word, hear folks talking about building millions of robots and at the same time saying they want to rebuild manufacturing in America, for example. You can rebuild manufacturing in America, but you're not going to have workers doing the work. And I worry, you know, point that Dr. Hinton made earlier in the program, who the hell, how do people going to get an income to buy the products that are being sold? And in terms of, uh, AI in the world that you are living in, I do worry very much that if entry-level jobs are not going to be there for you, how do you move up and progress up the up the ladder? So it is a, a real concern. Let me, let me, if I might, Mo, just ask the people here, just raise your hand. Let me see how to frame this question. Do you see, or it'll be, I'm going to ask you two questions. The first one is, for your own lives as young people, do you see AI as a positive development for your future? Raise your hand. Okay. Same question. Are you concerned about it being a negative impact on your future? Wow. Okay. Most of the hands went up when they got. I would tell you this, uh, induct, where I became, I'm not even on any of the committees in Congress that deal with this stuff, but I got involved. I went around the country, uh, recently, and I was in Davenport, Iowa, and speaking before a working-class audience, and in the course of a speech, you know, I said, I'm just curious, and I asked the same exact question as I asked you. It's a working-class audience, and I said, tell me, do you think AI and robotics is going to be a positive development for you? In a room of about 2,500 people, two hands went up. So I think among working-class people, there is a legitimate and real, and I think based on reality, concern about what AI will do. And I'm hearing that in this room, uh, from many of you as well.
Senator and Professor, I want to ask each of you to share any final thoughts you have, but one question I'd ask you to weave into your final thoughts. What does the average person do, right? What can these students do to prepare for everything you've been talking about?
>> I think in many ways, that question, Mo, goes beyond AI. It really is a political question. And as we have said a dozen times, AI is neither good nor bad. It depends on who benefits and how it is used. Right now, in this country, you are living in what I refer to as an oligarchic form of society where a small number of very powerful people, people who are moving aggressively in the area of AI, but they were powerful before they did that, have enormous control over the economy and the political life of this country through their ability. Musk, for example, contributed $270 million to make Trump the president. So, it is not only AI, it is not only climate change, it's not only the economy, it's not only healthcare. You are living in the richest country in the history of the world. And if we had a rational system designed to improve life for all people, there are extraordinary things that we could be doing. AI is going to make us wealthier. Yes, education could be free from child care to graduate school. Yes, we could have universal health care. Yes, we could lead the world in transforming our energy system. Yes, we could build the units of affordable and low-income housing that we need. This is not utopian, radical ideas. We could do it if you have a government that is concerned about the needs of ordinary people. If you have a government concerned and dominated by the rich, they will get tax breaks, they will get deregulation, they will be able to produce more carbon emissions, etc. So the real question to me is a fundamental political question. And we need your generation. And we're seeing progress, I got to tell you, all over the country, standing up and creating a government that works for all of us and not just the big money interests. That to me is the issue. [applause]
>> Professor, for those of us that are not programmers or engineers, what can we do to get ready?
>> Um, what I have to say is basically what Bernie said. Okay. [applause]
>> Um, I can't thank each of you enough for joining us tonight and kickstarting this conversation. Not that it needed to be kickstarted. Students here are coming into our office and many others talking and asking and thinking about AI on a daily basis. I want to thank, um, the student volunteers from GU Politics as well as from the computer science and massive data institute programs [applause] who helped us put this together tonight. But most importantly, even more importantly than thanking these two, I want to thank all of you for coming out tonight, sharing part of your evening, and asking such great questions and engaging. Thank you all very much. Have a great night.
>> [music]