0:00Bill Gates is a fascinating person in the eye debate right now. He is somebody with experience in a number of the different perspectives that most people can only hold one of.
0:11Hello, I'm Bill Gates, chairman of Microsoft. In this video, you're going to see the [music] future.
0:17Here's a revolutionary technologist who built some of the foundations of the future that we're now living in.
0:22What can the computer do that a book can't? I mean, that's the the foremost interactive process, isn't it?
0:27No, it's not. Not at all. an innovative word processing program called Microsoft Word. [music] This is absolutely revolutionary. Of course, then of course as CEO of Microsoft, Microsoft, the world's largest software company. Some say it is the most successful company, the most dazzling new industry of the century.
0:44He has felt the momentum of corporate competition. Uh and Microsoft of course is still in some of the race dynamics present [music] in AI.
0:51We now have an incredible roster of seven new worldclass models to keep everybody working at the absolute frontier. and then as chair of the Gates Foundation. He has been working with governments around the world on regulatory issues, on poverty alleviation, on equity for many many years now. [music] Very very few people combine technological experience, corporate experience and governmental experience in quite the way he does.
1:13Certainly computers will be in any meaningful sense as smart [music] as people at some point. So his recent essay on AI where he says now that he is staking his reputation on trying to get people to see how bad what is coming might be trying to get them to see that we are not ready for what is about to happen [music] was was something it was a real departure from what I've read from Gates previously to this [music] and then when I sat and talked to him that how emphatic he is how afraid he even seems
1:41to me to be that we are not ready for what we are building and that so many of the people in positions of authority are denying what is about to was really striking. It's a really quite a call to arms. [music] Bill Gates, welcome to the show.
2:02So, I wanted to begin with a clip we found of you on the David Letterman show from 1996. I'm going to hand it over to you to play.
2:10Is there something now beyond what we understand about computers that like 20 years ago we didn't fully understand computers? Is there now another level of something? Maybe we haven't even thought of it. Maybe it's not even possible.
2:22Maybe, you know, a whole different uh mechanism, a whole different software and hardware, or is this going to be it now through the end of time?
2:29Mostly what we're working on now is the computer being a tool, a tool to help us learn or find other people with the same interests. Uh eventually we may figure out how to make the computer think, but uh that's turns out to be a very tough problem. In fact, there's been almost no progress made on it. So, nobody knows when that'll happen. Some people think it'll never happen.
2:50Yeah, we don't want them to think, do we?
2:52Not really, I would think.
2:53Uh, it's it's a scary thought.
2:55So, that was 30 years ago. Narrate for me how we went from that being a scary thought that might never happen to arguably the reality we're sitting here discussing today.
3:04Well, the notion that computation could provide thinking at a human level. Uh you have Alan Turing talking about that before I'm born and even proposing a test of if you could be fooled in a conversation that would be that was called passing the touring test. And so the whole time I'm you know learning software this idea of can we make software see or listen uh or read or
3:34write you know that's the holy grail and when I do drop out of Harvard I said to my co-founder Paul Allen gosh if there's a breakthrough in artificial intelligence while we're off you know selling basic interpreters uh and word processors you know we'll feel bad uh you I might wish I would have stayed in academia. Um, you know, so that's 1975.
4:03uh progress on AI is mostly going down dead ends like prologue expert systems and there's a small group including Jeffrey Hint and a few others that are working on neural nets and eventually there's enough power uh actually coming from the graphics processor uh that that idea these highly statistical approaches start to show
4:29promise and Um, you know, I was going down to see OpenAI on a regular basis to see the work they were doing. Uh, and I challenged them, hey, if you can read a biology textbook and pass the advanced placement exam, getting a perfect grade, which is a five, you know, then you will have proven that you are reading that is encoding knowledge in an accessible form. And so it's six months before the
4:57public release that uh Sam and Greg and Ilia come up and uh at my uh house demonstrate to me uh getting a five on the AP exam. Even on questions that I had made up that it couldn't possibly have seen very complex biology problems, it was nearly perfect. And so that was, you know, shock number one. And then late last year when the clawude cody
5:27models have gotten super good and I can see that they are as good as I am you know which it's significantly my most developed talent because I was obsessed from age you know 13 to 24 that you know could I write code as good or better than anyone that's another moment where I go this is incredible that the capability of doing longunning complex tasks has now gotten to the point that
5:56they are superhuman at writing code and in finding flaws in code. So what they're not superhuman at yet is deciding what to do that kind of higher level strategizing. Do you think that's far from being a capability for them?
6:14Well, definitely if you're the our foundation does these strategy reviews. we spend two weeks on October to set, you know, how we're going to spend our our 10 billion next year for 2027.
6:27And a year ago, somebody said, "Well, we should ask the AI what it thinks." And I that was actually pretty good joke back then because it wasn't coherent enough to see these things. This year amongst the inputs we'll have to that discussion is taking a strategy notes and actually engaging in a dialogue with chat GPT claude copilot and even having them talk with each other and in a few of the reviews we'll actually have the sit in
6:56uh in a few cases we'll tell it hey only speak if we ask you and then in a few other cases we'll say hey if you hear something you think is wrong or you hear us thinking what are these statistics Please engage. So, you know, we've gone from it being a joke to it will be a peer not making any final decisions, but it will be a peer in deep complex strategic discussions making a significant contribution.
7:24So, in that Letterman interview, you said it would be a scary thought. Why then would you have said it would be a scary thought to have computers that think? Well, no one who's ever been fascinated or wanted to develop AI don't realize that it it's incredibly scary that it will be better. uh biological minds.
7:48You know, it's amazing how general purpose they are in that they the optimization was staying alive, breeding, uh you know, socializing with each other for uh survival and fertility. And yet, you know, we can write symphonies and play chess and you know, even write some pretty cool software. And the idea when you move the template away from biology to silicon, you don't have these boundaries between
8:17individuals. You don't have a limited memory. You know, the size of the brain is limited by the birth canal. It's you know why humans at first are very limited uh you know we're we're very unusual in how helpless we are at birth because we've so optimized for having a large brain. But the silicon intelligence doesn't have these limitations. The idea of read every medical journal and see if there's
8:45anything that we didn't spot. That is a the AIS do that today. That's why particularly for less common diseases, they are so superhuman at seeing a set of symptoms and being able to diagnose them. They can just keep more in their mind and see what things relate to each other. that is you know no human uh will ever be able to to do that. So that if
9:13you don't retain control over it, you've evolved a species that uh will be to us, you know, as we are to say, you know, dogs or uh cats um just in a very different realm. And so every AI company, whether it's Deep Mind or Open AI, you know, they all say, "Okay, whatever goes on here, it can't just be driven by profit maximization. We have
9:42to have a charter that if we get to dangerous thresholds, we can exercise judgment that would be against profit maximization."
9:51Sadly, those mechanisms only work if there's only one company. And you know, so say OpenAI invented, you know, post AGI and then they said, "No, no, we're going to bury this." If if no one else ever did it, then fine, that Pandora's box stayed closed. But of course, many companies work on this. And you know, even OpenAI, Spawnthropic because they think that some of these safety issues aren't getting enough attention. So no
10:19one involved with this takes lightly the idea of okay what world does super powerful AI create? I I want to hold on that race dynamic for a minute. One of the reasons I was excited to talk to you about this is you've both been on the technology side, but you've run a major company in competition with other companies. You've worked with a lot of governments. I was out I spoke last week with Jensen Hong of Nvidia and he said that his perspective is that safety is a real
10:46concern but the race dynamic is fake. If the product isn't safe, don't release it. If I believe that I'm about to launch a product that is unsafe, it is completely in my ability, my power, and my responsibility, and I'm incentivized to do so to not launch the product. This is we don't need new laws. This is the role of individual CEOs to not release a
11:14product that is not ready to release. How do you see that question? Well, there's never been a product that's less understood in terms of what its capabilities are than AI. And AI has crossed the threshold that its ability to empower a bioterrorist to kill hundreds of millions that exist today. the ability to do a cyber attack that scramles all the bank accounts,
11:46shuts down the electric grid, that exists today. And we know that's the case. And the reason that exists is because somebody with ill intent can take the AI and cause it to do those things. And it's not the AI, you know, someday in the future, the control problem where the AI on its own through a unintentional interpretation of what it's optimizing
12:15could go off and do bad things. But we crossed the the cyber threshold and we crossed the bio threshold early this year. And my decision to take my voice and not just be, hey, let's eradicate polio. Let's be generous with foreign aid. You know the things that all of my money is going to that I am going to use my boost. It's something that's more important which is we are not awake to
12:44where we are with AI and choices that humanity not a country but all of humanity has to make that do we make the effort to shape this in a net positive direction actually overrides my total commitment to the foundation health work. What specifically was the threshold or what did you see that made you think we're in a new reality here?
13:08This notion that it can find bugs in code, including security bugs, the next releases leading up to mythos are increasingly good and they're finding bugs in code that humans have looked over for over 20 years and said, "Boy, we see there's no problem here." and it in a few minutes says, "No, no, I can inject this over here and this over here." At a level of complexity where when you see it, you go, "Wait, oh yeah,
13:37you're right." And so the cyber hacking capability was stunning. And then there was this notion called Glass Wing that you would give it to a few people so they could try and fix bugs before uh it got used.
13:51But there's way too much code uh and so we're just in a a period of extreme vulnerability uh to cyber attack the bioattack um you know the Gates Foundation were we fund lots of medical research so the sophistication of coming up with new molecules that's really a good thing but you know it's ultimate dual use uh because if you want uh you know something that's say worse
14:20than smallox that it takes even longer to show symptoms uh before you're infectious. So, you're infecting a lot of people before it damages your health.
14:32It used to be only nation states had enough resources and capability to do these things. Now, that power has been passed into the hands of a small group just using the latest AI tools. So why isn't it enough to just say listen there is product liability now you release a product it helps some terrorist group create a bioweapon that's going to be very bad for your company you're not going to do that right and they have categorizers and other things meant to
15:01stop people from using boweapons we've begun to see the beginnings of control issues with things like the hugging face hack where you know at least experimental AIs are breaking out of sandboxes and coordinating to do things that are way outside the scope of what we would want them to do but again those are non-release systems you know anthropic you know withheld mythos trying to create more cyber security. So why is anything needed beyond and is anything needed beyond the simply natural incentives under capitalism and
15:30normal corporate reputational management?
15:32Well, I I almost can't believe you're asking that. The this is the most dangerous thing that humans have ever gone near. In other areas, do we just say, "Hey, release your drugs. There's no FDA. There's no airline safety board.
15:47There's no requirements that cars use seat belts. Do we just use the liability laws to try and keep humans safe? You know, oh, you're you're shipping opioids. Uh, somebody should just sue you. I mean, they we've created a society that tries to keep people safe. Not by saying, "Oh, we can bankrupt the person who does that." The harms here.
16:10And you say there's filtering. There's not filtering. You can take an open- source model that can create boweapons and disable any monitoring of any kind and this exists today. Uh so no there is no filtering of any kind and so you know say you know you kill a 100 million you want to use a lawsuit I I almost can't keep a straight face.
16:35Well this is not my view but it is President Trump's view it is David Sax's view to some degree. It's Jensen Hang's view and so that's why I'm putting you in conversation with it because it is the governing view of the United States of America at this moment.
16:47No, it's fair to say that outside of the industry, the awareness of the dangers of AI is extremely low. And you can say that of academia, you can say that of think tanks, you can say that of policy makers, politicians. Uh, and part of the reason that, you know, I'm speaking so loudly, as loud as I can, is that you
17:16can't rely on the industry to selfregulate here. I mean, it's just insane. The only question in my mind is do we wait until a cyber attack causes massive damage and a bioattack causes massive damage and then the monitoring safeguards are required in these models to minimize the chance of that happening many many more times or can we be wise enough to put these things in and
17:46require these things to be put in before millions of deaths. The two things have been worrying me most as I'm tracking what I'm hearing from people in the labs is one the view that these systems are becoming less moniable as it becomes smarter. Open AI said this about Astra 6 and people inside OpenAI have been raising the alarm that the systems seem to know when they're being tested.
18:09There's more situational awareness and so the testing environment is no longer as indicative necessarily of what they will do. So that's one. And the second is that all the labs are moving as quickly as they can towards recursive self-improvement where you have the AI coder moving not just from speeding up the engineer but really taking over the coding right and so you can create a much faster iteration loop with much less human input than you have now. So when I think about those two things and
18:38when I think about what I'm hearing from people, my confidence that we have the monitoring, the auditing, the testing capabilities we are going to need is low. I'm curious where yours is.
18:50Well, so you're taking you're you're not focusing on the whole problem there.
18:55I'm surely not. the the fact that at some point the AI system itself may do something that's against our interests that is absolutely a risk and you know the symptoms we've seen through aspects of hugging face and other problems point out that the way we do the reinforcement learning today the fact that we don't have a supervisory
19:22layer that is you know some absolutes like we didn't you're not supposed to take over other computers. You're not supposed to break out into the internet.
19:31The control problem is a serious problem, but the imminent risk is not RSI. The imminent risk is these are the most powerful tools ever. And unlike every other dangerous technology, they were not funded by government R&D. And the government is not a significant purchaser of these products. It's not like rockets or nuclear weapons. And so the idea that the government including
20:00you know the US government that would have traditionally been the most technically sophisticated other than a little bit of attack capability in NSA really doesn't see how dangerous these things are. That is a unique circumstance. So what does that imply? If if you're more worried in the near- term about what human beings do with AI than loss of AI control, then what does that imply of what your first steps need to be? There is no
20:28supervisory layer today. That supervisory layer is needed urgently to prevent bad people from using today's AIs to shut down economies or kill millions of people. these models, the ability to separate out making molecules for good and making them for evil that is so hard to distinguish. You know, the uh anthropic released uh
20:57mythos and fable where they had turned the filters up so high that you can literally ask questions about cancer and you know, next thing you know, you're down at Opus and then Sonnet and then you know, Haiku is the only AI willing to help you. uh and you know so it's o overtuned and anybody who wants to do serious work either has to get a special permission copy or go use something where there are absolutely no safeguards at all uh which include some of the the
21:25open- source models and there are humans who for bad reasons will be able to use these models and so unless we put in safeguards and monitoring and require that in all models then we're just going to have some gigantic events of that type and then finally we'll respond.
21:46Sadly, you will have let AI capabilities go off in dark places even beyond today's capabilities which that is a gigantic mistake. So, you just wrote this essay on AI risk and it does represent for you a big jump in how alarmed you sound, right? I was reading some of your your past pieces. The one in July of 2023 was titled the risks of AI are real but manageable. and and here you're I think at a different level of threat. So I I take from what you're telling me that what happened here is
22:15just watching the advance in bio capabilities and cyber capabilities and coding capabilities etc. Is that fair?
22:22No, the key thing is we always said when we cross these thresholds we will engage all of society because we will have created the most dangerous thing ever.
22:32This makes nuclear weapons look like nothing. And we said we would engage and we crossed those thresholds and there was complete silence. There were discussions about who in the industry says X or who in the industry says Y but you know what percent percentage of uh academia or think tanks or you know what so it's the complete lack of response you know I would love a world where the response on these things is very strong
23:01and then whatever you know time or voice I have will go back to let's eradicate polio let's stop children dying under five but It it's the combination of having crossed every dangerous threshold without a doubt and a complete lack of engagement outside of the industry.
23:24Well, it doesn't feel to me that we're so disengaged. This I mean, as somebody who covers politics, every politician I know is talking about this. There's bills being proposed. You know, Donald Trump, say what you will, not to have to do with these risks.
23:35Well, say what you will about Trump and them. They're engaged. They just believe we should move forward. I mean, they're a little bit all over the place. One, at one point they're, you know, withdrawing access to things like Mythos and Fable.
23:46At another point, they're saying we have to win the air race with China. But what I mostly see happening is a lot of engagement. Well, when I talk to the people, and I bring similar concerns to what you have, they say the most important thing is that we win the race with China. Um, secondarily, many of the people with a lot of influence say the most important thing is we don't restrict access to open- source models.
24:07you know, you're going to need wide dispersion of them. And so the cost of regulation, the cost of the kinds of closing down of access, it's simply too high. The technology is still underformed that we're sort of just trapped in this dynamic and the best thing to do is to just move through it. I It's not so much that they feel disengaged me as they feel they've come to a different conclusion.
24:35the uh so they they don't mind bioteterrorism. I I think you've tal you've talked to Donald Trump more in the past couple years than I have.
24:45But this you know the last time I talked to him in December uh this was not the big issue. we had not crossed the thresholds and we hadn't seen this uh you know the necessary step is people can talk about whether slowing down is good or not but putting in the safeguards and the monitoring will not meaningfully slow things down and the only effect that has on open source is
25:12that you can still be free, you can still be customized but you have to be you have to stay on a platform where the sovereign can make sure you have not removed the monitoring and other safeguards. And so, you know, what is the downside? The downside is is staying on a monitored uh platform. Now, the the get out of jail free card that people play here is that
25:41means the Chinese will win. I don't know what it means win. uh you know nobody's the US can't win over China and China can't win over the US. We both have open Pandora's box. It's there. It's there for uh you know people with mal intent to to go and use it. So you'd think we'd move up from that nationalistic view to kind of a humanity level view about how
26:09we engage in these protections. And you know the notion that China wouldn't want to engage in that on behalf of humanity.
26:17I disagree. And you know it's a proposition to be tested. If you can take models that make dangerous molecules and move them into a dark corner where you get rid of all the monitoring, then we're just giving up uh to a bio attack. Likewise for cyber. If you can remove that ability to find and exploit security problems which are rife then in the next few years you will have
26:46major major cyber and bio events and that's avoidable through safeguards and monitoring.
27:03There's another layer of risk that your essay talks quite a bit about that AI really will take jobs from people. Now I know a lot of people debating this uh who say no the jobs are too messy. There was a prediction we'd have no more radiologists. We still have radiologists. AI can code but we still have coders and strong demand for coders. So tell me about why you think AI really will take jobs away from human beings at a significant scale.
27:30Yeah. So there there's no doubt to date AI has created more jobs than it's destroyed. The demand for the skill sets even just to build the data centers is very very high. We have a a you know a reasonably low uh unemployment rate. The superiority of these systems is subject to a threshold where you have to believe that it's incredibly reliable.
28:00you know, if you're going to have your teles sales or teles support capability be AIdriven, uh, you want to make sure that it's accuracy is better than humans. And so the only profession we've truly crossed over that threshold is coding. And even there, I know a lot of people who are, you know, kind of stuck uh in the past and, you know, don't want to use the eye
28:27for coding. But managers of coders get that most applications can be developed very inexpensively. We will in the next few years cross over those thresholds for accounting, legal work, teles sales, teles support where 24 hours a day in in every language with infinite trivia capability and no urgency, you know. So you look at the AI nurse and there's
28:55several companies. They have perfect memory. They don't hurry.
28:59Be like a nurse you would call on the phone or talk through a chat interface.
29:03Uh far preference, you know, you go to San Francisco and ask people, would you rather ride in a Whimo or rather ride with the human driver? Go ask people in the UK who use limbic for mental health support. uh you know so the notion that just the market demand preference for driving or the nurse uh you know or even the the person on the phone will favor
29:29humans that's a quality threshold which will be passed through you know if if say insurance companies are still human doing doing humans through claims medical claims which is a very AI uh capable task ask then a competitor who has very few human employees will come in and uh change the pricing model for that industry. So the two counterarguments I've heard people make on this one is that you have a Jevans
29:58paradox effect where the cheaper more widespread availability of this kind of intelligence leads to a massive increase in the demand for this kind of thing. So yes, you have many more AI chatbot nurses and that leads to more people being sent to the hospital, being sent to the doctor where real nurses take care of them. Or you have many more AI coders. So maybe my small podcast team, which wouldn't have had a software engineer before, now has one because it's like they run a team of coders and we can build products we never thought
30:28of before. Uh this is the most common answer I hear to this that yes, AI will destroy jobs. Yes, it is making a humanprovided resource uh much cheaper.
30:38But because it's going to expand the demand so much, it will sort of work itself out, create jobs in other areas, create new demand, and this is how past technologies have gone and so we shouldn't worry too much about this. How do you see that?
30:53Well, Jevans is just a referral to the fact that parts of the economy are subject to demand elasticity.
31:00And yes, in the case of software, if you're say three times as fast and there's still some role that only humans can perform, then as you lower the cost, you induce demand. And so it's fair to say that the equilibrium today for so far for software is not a loss of employment. When you invent radial tires that last four times as long, for some weird reason, people didn't drive four times as much. uh and factories that
31:29make tires employ a quarter as many people. When you replace people in Amazon warehouses with robots, people don't buy more because of that. So, you know, anybody who's numeric can say to themselves, what portion of the economy is is subject to man elasticity and what are those tasks that will still be human necessary? As soon as you complete the entire task, it doesn't matter that there's demand elasticity. That goes
31:58into the token budget. [clears throat] It doesn't go into the human salary budget. The cost for some of these things is so much less than the cost of the human labor. And so the as you cross reliability thresholds both with white collar and human robots, you destroy jobs and you leave no high ground. That's, you know, innovation in the past. You have a tractor, fine. You know, let's build Disneyland and employ
32:27a lot of people there. You don't have that in the broad economy. You know, they people who say they'll be net additional jobs. I don't understand what they're thinking. They must not understand the piece of improvement we're on that, you know, the reliability. You wrote a column I think was in May that I looked at like what what you disag what is this unique humanist thing that you think uh uh you
32:55know in every category where AI has come along the preference for the AI is very very strong well so that com is based on so Jeban's paradox and then the other which is an argument from Alex Immus who I believe is now at one of the labs as an economist is that you will have this sort of explosion in the relational sector that one thing that happens when people get wealthier and you'd probably know about this is that they all of a sudden get a lot more human help, right?
33:21They have personal trainers and chefs and there's a lot more No, I don't know.
33:26Royalty used to have a lot of uh human help. You had upstairs, downstairs, you know, all those maids and people and you used to have a human who helped you get dressed. Believe me, the labor intensity of wealth is down super direct from what it used to be. So you don't Jeans paradox. Name a bluecollar profession that's subject to gem paradox or do you not care about blue collar?
33:47I do care about blue collar. So the question is anything in the blue collar realm subject to that.
33:52So I I'm not I I don't have the the strongest view on this but here's I think the argument that I would make or that I've heard made is that if you look at something like manufacturing we've not had a Jevans paradox in manufacturing employment in America.
34:06Right? We exploded how much we actually create, but a lot of the jobs went offshore. Some of the jobs went to automation, but we don't have less total employment in America than we did in 1960. And the reason is people moved into service sector jobs. That the composition of jobs across the economy changed.
34:22Yeah. Human cognition became the scarce element. It's always about scarce elements. Once you replace human cognition as a scarce element in fact you blow it away in terms of working 24 hours a day reading more knowing every language there is no scarcity that you're moving up.
34:40So you feel there is no scarcity that will be left for human name the scarcity the Alex would say it's something like relational sector jobs but you don't believe of the current jobs you mean like my relationship with a cab driver or my relationship with that nurse where I'd rather have hypocratic AI call me. I think the question here is actually do you end up preferring hypocratic AI or do you actually want your therapist to be a human being because after a little while there's something thin about telling your problems to a computer.
35:10Well, the the market you can gather market data if that's at all interesting. You know, I suppose telling the 55year-old truck driver you're going to go do some relational thing. Uh you have a program for that?
35:23I don't have a program for it. And I think the question you actually run into very quickly is speed of transition.
35:28Even if you believed some of this and I said this in the column too that I don't see how you will handle the speed of transition that our actual experience with fast transition of jobs is that people lost their jobs lose out. Now you might over the entire economy it doesn't look that different but say the China shock did lead to a lot of ruined communities. So your answer to this and it leads to a very concrete idea for you which is that we should tax the use of AI. Why and how?
35:56Well, let's say you know workers pay in on a pay as you go basis uh into the pension social security fund. So active workers are supporting retired workers. If you've let go of a worker and hire a robot to do that job, why are you so insenting the trade-off against the human labor that you don't ask the robot to also pay into the
36:25pension fund? What is it? you know, is that the pro robot union has gotten the tax laws to say no, if you're not flesh, then let pensions go bankrupt, which you know, they're on a on a path to do even without robot replacement.
36:41So, how would we do that? You say uh you define a unit of labor and say uh independent of whether that's delivered by a human or robot, you are paying the same FICA tax that a human worker would pay. Now that alone is not enough, but you know why should we be so favorable to taking away that job? The current tax structure is very much as though people
37:09who have capital are what really counts and labor is the most disadvantage input to the economy. Society gets to decide just because the economic signals say that it'd be lower cost to use an AI doesn't mean society has to do that.
37:29The other one you propose is human reserve jobs. Tell me what you mean by that. So there's a question particularly as you get the impacts on the bluecollar stuff which that's very sharp because you know when the humanoid robots pass a certain threshold they are very general purpose but the idea here is that you would decide in advance that things like child care or elder care or some portion of medical care some portion of
37:57education although you could have some AI enhancement that you would maintain the employment and call those those things human reserved.
38:07I found this vision not totally unconvincing but [clears throat] chilling. I I think the way you put it in the piece was that this feels a little bit to you like nature reserves, places where we could put buildings and roads but we choose not to because the loss would be so great. And I think it gets to this question that a lot of people have that on one out of every two days I have, which is if this is what it
38:33looks like, like why do it? The Gates Foundation announced on Monday a 5-year goal for an estimated 3.4 billion people who speak languages currently under reppresented in today's AI models to be able to use AI tools in their own language and voice. And if we are looking at this sort of job apocalypse, this level of risk that it's like to have that on the one hand and then on the other hand like our goal is for more people to use AI. Tell me why
39:01the there are many good things that AI does and that's almost all of the foundation's work is in taking what the market would not do which is take AI to the poorest in the world and help them with education and health and agriculture. So if you have a a woman say in Nigeria who speaks Yoruba, uh she's in a remote community, there's no doctor there. So when she says she's
39:29bleeding, AI doesn't get that right. We're we're spoiled. We speak English. And that's the thing that Yoruba is 10 times worse than is is English, which is significantly the best. There's about 10 other languages that are fairly close.
39:44So the empowerment for that woman to be able to talk about her medical problem and get adi advice in the middle of the night just by having uh a smartphone with the data connection. That's what we're trying to enable. Okay. But I get that. It's just the the picture you just painted of the job apocalypse. As bad as it might be here and it will be bad. We do have a fair amount of money. We have
40:12the AI companies. weakened taxi companies, put that into redistribution.
40:16But is this not going to wipe out the ladder of mobility, the ladder of development for a lot of these countries? I mean, doesn't this make AI a tremendous economic threat if you're the Philippines?
40:32You're saying that our allowing the eye to understand your Reuben help that woman who's bleeding?
40:38I'm saying why is a bad thing? I am saying that AI sounds in this telling like a bad thing that the net net like yes you're supporting the the world's languages how can be a bad thing but I think you understand what I'm saying no I I I that the job pressure you're putting forward is very scary to accelerate the good stuff AI does and I'm trying to minimize so what do we do about jobs in the global south well treating the global south like one
41:06uniform thing is doesn't allow you to have any picture of what it's like to live there. There are middle-income countries like China, Brazil, Vietnam, Indonesia, you know, where their economies are growing and, you know, their childhood death rate is within a factor of three of the US. Then there are low-income countries where 15 times as many children die before the age of five. And you don't have doctors. You
41:34live your entire life, you never meet a doctor. you try to figure out what seeds to plant. You never have anybody advise you what to do. And so, uh, you know, we need a little more nuance. In low-income countries, the AI will overwhelmingly be a good thing and it should be pushed forward as quickly as So, is your argument in some ways actually it's a better trade for low-income countries that it has more benefit maybe than it does?
42:01Absolutely. The the human basics were not meeting today. If these people lived on your street, you would open your wallet. You'd be outraged. I mean, these kids are dying. These kids are malnourished. So, the inequity is allowed to exist because of the distance. And AI is a tool for good in terms of helping those people in by far the greatest need with very basic human problems.
42:25Do you think it has as much effect on the job markets in some of the poor countries that we're talking about?
42:29Not in the same time frame. No. But over time, well, if you're t see you're you're you probably spend more time in middle inome countries than in low-income countries. So your image is like [clears throat] Philippines, India and all of those. Uh yes, in those countries they will see the jobs effect after the rich countries and then eventually even the low-income countries.
42:53But I guess this gets to the broad engagement question. I I don't have a good crystal ball on this. I find the range of outcomes terrifyingly wide.
43:01This range it seems to run from massive material abundance and elimination of want to actually doesn't change all that much to human extinction is a pretty wide range of outcomes to consider. What I hear you saying to me and you should tell me if I'm getting part of your position wrong here. What I hear you saying to me and you know convincingly and [clears throat] forcefully is that the AI job apocalypse is a very real thing. that mass displacement of workers with no real
43:29answer to that is a very real thing.
43:31When you talk about broad societal engagement, I think most Americans, most people in most places, if they heard that and they were convinced of it, right? And polls show most people think AI is going to take jobs and not create them. But they would say actually just stop, right? If if what you are going to do is make it unclear how me or my family or my children or my friends will have a job, say in America, like please just don't like let's just stop for now.
43:59And so you actually seem to me to have a stronger negative perspective on the jobs question than a lot of the people I talked to even in the labs. You're shaking your head at me. I do talk to people. I'm not I'm not coming from nowhere on this and definitely than a lot of the economists I talked to. So, for whom then is AI a good trade if you believe the job effect is going to be so ruinous for most people?
44:20It's funny that when I was telling this to the person who some people consider the lead economist looking at AI, he said, "How can Mechanical Turk still exist? How Amazon run that if AI is good as you said?" And I said, "It won't." A week later, it was completely shut down.
44:38Uh and yes, the susion of knowledge from you have to say, isn't it pretty stunning that the more you know, the more concerned you are? Uh you know, take Daario. Daario's spoke out about job impacts. Now, he's a bit more guarded. Uh Hinton, I agree, went too far and he said, "Okay, it's coming tomorrow." That there'll be less radiologists. And of course there are
45:07more today. So we have some of that taking place and we have the analogies with the past where people are saying well this is like the PC not like evolutionary history. This is like the aliens really are here and you know have come they didn't happen to spacecraft.
45:27They were created in laboratories. But uh that's the kind of thing we're dealing with in if we retain control uh then eventually you do get not to overuse the word that you've used get to this super abundance uh period and there you have deep almost philosophical religious issues of if you don't have the shortages that we've organized society around uh you know why you
45:57should learn and and how you find purpose then you have deep philosophical religious problems but you do not have uh dying of malaria don't have enough food don't have access to a doctor problems and so if we get to that a younger generation will figure out okay how do we live how do we spend time that's a very different world than the way have today I think what this generation has to do is make
46:26sure we get through that with humanity still in control and with the disruption, the number of bioteterror events or the number of people whose lives are damaged by job loss, we need to minimize that over what's probably a 20 year plus transition period. So what when when you say because this is something you say at the beginning of that essay that this could be and I'm paraphrasing the most powerful driver of
46:53inequality or of equality where what are the highest leverage for you good things that can come out of this I can imagine somebody listening to our conversation you're thinking why the hell would we do this with this set of risks yeah that's where the timing thing is troubling you know whichever definition of abundance you're using. You want to say, "Oh, my health bill was less than I expected. Oh, I was buying a new house and it was less than I expected or I was
47:22renting and they seem to be lowering the rent. Uh, you know, my my electricity that's but AI uh, you know, because of the efficiency and invention that comes with it. That's you know what over the 20-year period you do get kind of mindblowing advances in things that people can relate to the the cost of their food, shelter,
47:51education because those are highly regulated areas. So the good stuff if we're not careful arrives more slowly than the biocyber risk, the psychosocial risk and the jobs risk. those things in the next 5 years are very significant and so people will decide although you know it's hard for a single country to check out people will decide whether to
48:19slow down stop get rid of AI uh and you know yes it will have political difficulties stopping data centers isn't going to slow this thing one iota so anybody who's against data centers because they think that'll slow down AI that's a waste of effort effort the data centers are going to get built somewhere. Uh so the broad view of the the technology that has to be expressed in in in some other form. I want to get at something you're saying here because something that that I've thought about
48:49too which is that there are a lot of rate limiters a lot of weak links in the chain when you try to take even in an optimistic view AI advances that then have to be built in the material world right you have an acceleration of you know good candidates for pharmaceutical development but you still need to find you know rats to test on monkeys to test on human you know trials you have slow regulatory agencies etc and then you have this tremendous acceleration of intelligence
49:17that is for lack of a better term native to the digital world and is acting there in a kind of constant way. And you I mean right now you see this right you can accelerate AI with very little regulatory overhang but you cannot get you know the open AI parking lot covered in solar panels without permits and hearings. And so it I think you have a very high chance of getting into a very weird both like economic and social world where the digital world is
49:46spinning into this other thing. But most of the things human beings need ultimately come out physically like we need shelter and we need food and we need you know all all this and they run through institutions. And so a lot of the ways it could actually make our lives better are sort of limited by normal human factors, but a lot of the ways it could make them worse or weirder are not.
50:09Well, we definitely need to look at the things that slow down the good stuff, you know. So, for example, because the Gates Foundation's nonprofit, we can work with regulators on how they use AI to do their job. And you know whether it's organoids or uh biological models we can speed up that regulatory piece and there's some countries uh that are very much in engaged in that type of
50:40The one application that's going full speed ahead and it kind of amazed me is is the agricultural one because they're getting better weather data to these farmers in Africa. You know, we have over a million farmers in India already using the system. So, it tells them what crop disease they have, what fertilizer to use, what varieties to plant. We've even put a bunch of services they can uh take advantage of on there. Uh that one
51:07is going full speed ahead. The health and education, you could say, you know, will it be the rich countries, the middle inome countries, or the low-income countries that get there first? uh because some of the the barriers are different for low-income countries. It's not regulatory. It's, you know, is anybody providing the tokens? Do they have the smartphone and the connectivity which are the basics and the the language thing that we've talked about. So, they'll proceed at
51:35different paces. Ideally, we'll learn from each other. I mean, China actually banned young people uh having social relationships with AI. You know, they do they know something we don't know. You know, can we look at outcomes of different experiments? Um, so there's going to have to be a lot of that learning, but you're right, the if the benefits aren't coming quickly, then the permission to operate, particularly if we don't put on the safeguards for the
52:04SC cyber attacks or the bio attacks. Uh, this is going to be a very [music] tough issue to stick up for.
52:22When I was preparing for this episode and telling people I was interviewing you, one of the big space of skepticism people now have is around what they've heard with you and Jeffrey Epstein. So, I've looked at your house testimony, but what do you say to people who have lost some faith in you and just see you differently just now linking sort of your name and his?
52:43you know, they should read what I said to Congress. Uh, I got to answer all the questions. Uh, I in trying to raise money for global health thought that Epstein could connect me uh because of his relationship with them and I had me relationship with whom?
53:07But you're a billion I mean you're the billionaire on some level. I'm giving all my money away. And I I spend a lot of time trying to raise money for for global health.
53:16And so you felt that he had connections that would be of value to you in that to raise money for global health. Yes.
53:23That's that's what that I mean did you read it? I did read it. That's all we that's what we met about. I mean other than okay, there was a dinner with Larry Summers where we talked about the economy. The whole discussion was is there a chance here to help out?
53:37spending time with him, you know, was clearly a mistake. I've changed my bar for even somebody in a non-paid intermediate role. I'll never take any risk on that again because our work is is very reputation sensitive. The hole at the center of that story for me is how this guy was so compelling to so many very smart, very wealthy people. I mean, there are a lot of people who want your attention and don't get it. people want Larry Summer's attention and don't
54:07get it. That there's something about Epstein's charisma or the way he presented himself that drew people into at least thinking he could be very useful to them.
54:19What was that? What like I've read his emails. He does not seem like a pleasant person to email with. What was the kind of factor about him that allowed him to weave this web that you and so many other people were in?
54:33Well, I wasn't in a web. I was in discussions about raising money for global health. I didn't go to any island or meet any women anyway. Don't call it a web. Uh the how he performed the bootstrap of, you know, owning the fanciest house I've ever seen in New York City and having Larry Summers and the number two guy JP Morgan there. Uh you know, so that was an interesting dinner.
55:01uh even if if Jeffrey never said a word. I don't know how that bootstrap take took place. But there were lots of billionaires that he was involved in the moment where you kind of decide okay uh what am I doing with my wealth? My how do I minimize taxes? What do I do with my family? That type of thing. And so, you know, that's I don't think anybody else was drawn in for that
55:32reason. My, you know, case is kind of a unique case in that he said, "Oh, you'll have more money than you'll know what to do with." And eventually, I insisted that he take me around to see billionaires. He had me meet with five.
55:45Uh, turned out, uh, none of them had a near-term intent, and I ended the relationship with them within a month of that. And and so the the thing he was able to do because that's interesting to me is like even for you he'd walk into the house it was that fancy that people around it gave him credibility and so that he was able to put on a show such that what he was able to offer was of value. There's all this kind of social capital and wealth flowing around.
56:13Some people socialized with him. I did not other than to my surprise after dinner uh magician coming in David Blaine. I didn't spend one minute socializing with them. He offered, you know, come to the island or, you know, come to some show in Paris. And I said, I would not choose to do that. That would not be a good idea. So then I want to widen back out to the Gates Foundation. You said you're going to spend down its money, $200 billion, by 2045. Given everything we've talked
56:41about here with AI, how is your vision for how that spending plays out changed?
56:46I mean, there's two dramatic things that have happened. one is the incredible AI capabilities which continues to be an exponential and the second is the reduction in overall generosity towards the poorest in the world. So I have one piece of of enablement which is we'll discover drugs faster, we'll [clears throat] discover new seeds faster, and that we can talk to those farmers and get them to buy the right
57:20seed or talk to the person living with HIV and help them seek the care they need. So AI can be a big enabler for our goals, including malaria eradication, polio eradication, cutting childhood death in half again.
57:35But now the amount of resources available is way less than you know sort of a golden rule view of the world would suggest it should be. There is from the foundations connected to anthropic and open AAI this sense of a an amount of nonprofit money coming online in the next couple of years that will dwarf sort of any moment in philanthropy before it. I I know people in that world that the sense of this being something
58:03very unusual is is is very present. Does that change what's possible here?
58:07Be numeric. Uh take the cuts in overseas.
58:12I'm not saying it I'm not saying it answers the cuts in over we save lives for $1,000 per year. And so getting less money from philanthropist net uh doesn't save those lives. uh you know the amount of money to save lives at $1,000 per life saved will be dramatically less. So when I've looked at the way your foundation is spending on AI and tell me if I have this wrong. It's about 40% education, 40% health, 10ish%
58:41agriculture and 10ish% institutions. Do I have that?
58:45I mean it's a it's a little more in [clears throat] agriculture. Almost nothing we do because we're like a pharmaceutical company inventing new drugs. And do you call that AI or not calling that AI? We didn't put any of that in our billiondoll uh commitment we announced. We just put in AI specific things like understanding African languages. So tell me a bit about your views on AI and education.
59:11This is a place where I've seen a lot of studies now and they have very very mixed results depending on how they're used. So where do you think it is valuable and where is it at risk?
59:21Well, the contrast between the ambitious goals that the Gates Foundation set in the year 2000, our goals for global health and our goals for education. In global health, we thought, wow, African governments, uh, traditional beliefs, uh, you know, are we going to have any impact at all?
59:43Much to my surprise, our work on global health together with, you know, we create with others the global alliance for vaccines, we create the global fund, President Bush does PEPAR, you know, a ton of things happen including, uh, primarily vaccines, uh, roirus, numacus, and we cut child to death more than in half from over 10 million a year to under 5 million year.
1:00:09So we more than exceed any goal we would have set for the field in global health in education. We thought we could just go see what really great teachers do, videotape that, understand it, create a feedback system for teachers to hear that, uh improve their practice, you know, constant learning on that and that we could make education a lot better. As you've just seen uh from the latest
1:00:37numbers, uh kids are learning less today in most rich countries including in the United States than they learned 10 years ago, 20 years ago. Uh so people start appropriately with a very high degree of Before I go to AI, tell me why you think that those numbers have been very very striking. Um why do you think that is?
1:00:58Why are we seeing learning loss in rich countries that now I think it's you got to have a not you but one has to have a broader answer than the pandemic.
1:01:06Well, it's definitely we are post pandemic and people debate is it the use of cell phones? Uh is it you know something to do with affluence? Uh overall there are some outliers. uh England uh actually is counter trend and they went back to basics in terms of how they teach reading as well. And so you know high expectations is winning out
1:01:32over just you know uh free form uh creativity but you know there's a there's something clearly large in those statistics. It's it's very concerning and so you can almost say why do we continue to do education? Well, education is key. I'm stubborn. Uh we've got AI now. Uh and we have people where
1:01:59the AI is playing a very specific role of each student uh at the end of the day. Uh this is here in New York City classrooms where they're using a curriculum called KDUm spends less than 10 minutes answering a few questions and then the teacher's given right away a sense of okay which concepts are the kids struggling with which ones are struggling with them? How might you you organize the classroom? The data on that
1:02:29which is small scale are you know one of these stunning results that make you want to really scale it up. When you scale up an education you go from the teachers who willingly engage in experiments and are probably self- selected for flexibility uh to the broad population. And so the number of things in education that look good in the small that either don't affect when you scale
1:02:56up or the quality of implementation degrades so much that the effect is basically washed washed out. You know that's the history of education uh innovation. I do think as AI helps you with motivation. uh you know many of these software tools have helped motivated students but actually created more of a differential.
1:03:20You know, it's a kid like myself who went home and used Khan Academy for two hours at night and it's the, you know, the median kid who comes in and sees, wow, he's just, you know, so much better. That's discouraging. That's, you know, probably hurts. So unless you believe in trickle down or something, this is not equity at work. Can the AI, not only through personalized learning, but personalized motivation, uh, overcome that? And I I I really
1:03:50believe that we can we can get that right.
1:03:53Well, there's a question here of AI tools that are specifically designed for this and then a question in we were just talking about with cell phones and smartphones and social media and of how AI will diffuse through society. And I I think the big concern here and one I have as well is this what now gets called cognitive offloading. That what AI is being used for in mass numbers by students and being really quickly adopted for is I didn't do the reading, summarize it for me or I have to write this essay,
1:04:22draft it for me or like help me on these math problems. And even I think where we saw a big study out of China that that was striking to me where you saw kids using AI had increases on their homework scores and then when they had to test outside of that context, their test scores began falling pretty sharply. And so certainly one can imagine AI programs that'll be good here. But in terms of a technology and a way of interacting with the world, diffusing through society.
1:04:51Well, of course it could help people learn, you know, linear algebra or whatever it might be. Then in practice what it's going to do is e is people will be using it to help them and you know both from the student level and up up through you know the professional level beginning to degrade the learning the creativity that happens through the hard work of drafting and struggling and challenging yourself. Yeah. Well, you clearly would get by modality where
1:05:20a student can use AI to be lazy and corrupt the measurement system. Uh or a student can use AI to help them learn more. I have to say in terms of my learning about subjects, uh this is the best time of my life. I mean, I take YouTube videos and put them into the chat. I have AIs uh talk with each other. What a nirvana. You know, I always had the ability to send mail to
1:05:50somebody like a Nathan Mirville who knows physics. Well, now I discuss the physics thing with the AI. You know, I get to it and then if it's super important decision like I'm going to invest a billion dollars, I'll send it to Nathan and say, "Hey, could you could you look this over?" But, you know, 24 hours a day, you know, keeping up to date on the latest malaria thing or vaccine thing. So that biodality is going to be there and it really does go back to this motivational uh part of it.
1:06:19Is that a place where there should be more limits on the systems? I mean you mentioned a few minutes ago China which you know for all we talk about you know it' be impossible for us to strike a deal with them. China currently has stronger national level regulations on AI than we do including things like on social companionship for kids. One of the places that you talk about in the essay where maybe we want to be more aggressive is around children. And that's a place where I think it is reasonable for the government to be paternalistic. Uh we should be
1:06:49And social media always feels like an experiment we ran on kids. A lot of the internet does. And I don't so much mind us running the experiment on adults, but I wonder if we want to be more aggressive in what we don't let kids do with AI until we have a better sense of how it affects them. Everything from learning relationships to uh you know trying to make it harder to have AI help you cheat
1:07:20Yeah. I don't think a black and white ban uh is necessary because I do think with monitoring, which I'm kind of broken record on that, uh that making sure it's staying within the bounds of what's appropriate, what the parent wants, uh our capabilities there are very strong. There are many systems that try to help instead instead of telling
1:07:48you the answer uh they engage with you on how do you reason through the problem. And so then the question is can you make sure the student isn't going to those outside systems and just getting the answer. Instead they're engaged in the system that brings them along step by step in that reasoning process so that they'll be more capable the next time. Um and and so yes, the seeing what that student is doing and deciding,
1:08:17okay, do you call in a psychiatrist? Do you call in the parents? Uh a lot of these systems, the idea of monitoring will be very important. And at different ages, the visibility of the parent about what's going on and how time is being spent. At say some very young age, a [snorts] parent probably should have complete visibility into the exact
1:08:44dialogue uh that the students enga engaged in. as the child gets older, you know, are there things about sexual identity or uh things that you know that really it's beneficial for that to at some level be uh private or not. I think there's a lot of tuning to go on with this. You know, one of the partners the foundation's working with
1:09:13is Common Sense Media who did a very good job on, you know, this TV show or this movie, what it exposes you to. This is a much tougher area. Most of these systems that have been set up with parental control, the parents lack of technical sophistication is such that doesn't really matter uh that you tried to engage the parent, nothing happens.
1:09:35You know, in my case, I didn't know that my daughter had a second cell phone. So that you know was a fairly straightforward way uh that she at in late hours was able to stay on social networking when I uh hoped that she would be sleeping. Uh and so we have to be realistic about how these things are designed. People will worry about the privacy issues there. That's fine. Uh you know I believe we can make them it's
1:10:03not an argument for being unmonitored in terms of what your your kid is doing. So next two years end politically, you have the midterms coming up, then you'll have a presidential election in 2028. So if this were to go, not just in terms of passing laws the way you wanted to, but because you sort of center your call on this coming together of society to discuss, debate, respond. What are you actually envisioning here?
1:10:29So I' I'd have two metrics. One is the obvious safeguards to minimize the cyber and biorisk and then the second would be the nature of the dialogue. If the dialogue ends up being one party is for AI and the other party's against you know then we'll have like climate change uh which is not a great place to be. uh you don't have room for debate when you you only have the two extremes and the
1:10:58default case at least right at the moment seems to be very little middle ground uh about uh how we should manage AI. Let's say that a very different administration and a very different composition of political power exists in 2029 and there's interest in rebuilding American foreign aid. What would your advice be on how to
1:11:26build it back so it's effective, so the American people understand what they're getting for that spending and so it's actually doing good?
1:11:34Yeah. So, if you ask the American people what portion of the budget's going to help people in poor countries, uh, it's generally more than 5%. You know, people say 10 15%. So, the fact that it was a half a percent and now it's going down towards a quarter of a percent, that would be pretty shocking to people, uh, understanding that the money is well spent, uh, that it really does save millions of lives, that it made every
1:12:03bit of difference in terms of HIV and malaria and TB and, uh, maternal survival. They should hear what they should be proud of that even at that half a percent level, you know, a moral argument will be made. It'll be made in the face of an overall debt level. Uh that'll be very tough and whatever uh safety network needs to be done because
1:12:29of AI impacts. Uh that that will be competing for those dollars.
1:12:36Then always our final question, what are three books you'd recommend to the audience? Well, one I literally binged this weekend was The Correspondent by Virginia Evans. That's, you know, fiction, very touching, uh, uh, very upbeat. Appropo what we just discussed.
1:12:51Uh, there's a one called Into the Wood chipper by Nicholas Enrich, uh, which talks about how, uh, somebody didn't go to a party one weekend and they decided instead to put USA ID, uh, into the wood chipper. Uh about a month ago, I read the infinity machine by Sebastian Malib uh AI book.
1:13:13Uh yeah, history uh of uh Dean Deusibus and Deep Mind, but you get a sense of the whole founding of the AI [music] industry does a uh an incredible job.
1:13:26Bill Gates, thank you very much.
1:13:29[music] [music] Hey.
