Tech Expert Warns of Revolution, Banking Security Crisis and the Oncoming Dystopian AI Future

Ally, thanks so much for doing this. Um, I've had a couple of people on in the past year to talk about the potential dangers of artificial intelligence. Two very articulate, smart, apparently sincere guys who scared the crap out of me uh with their view of where this is going and how it eliminates human autonomy and maybe humanity itself. and um the their views are well known and and their fears are widely shared. So I thought it'd be important to have someone who's in the middle of it from Silicon Valley. You've been in and around this for your whole life, really successful investor. I want to hear your perspective on AI. So most people are pretty worried about AI. Are you worried about AI? >> I think there's a lot of reasons to be worried about it, but also reasons to be hopeful. Um, >> so I have I have good news. >> Good. >> We're looking forward to good news. >> Yeah. I So I guess I'd say I'm I'm here because I want to share two things. One is that I think there's a solution that is elegant and can be expressed in a single sentence. There's a simple and elegant solution to the questions of AI safety. >> Yes. >> And then the second thing is I think there's a way that you and your listeners can act to help make it make it happen. So, uh, we'll get back to that later. >> You just made my day already. >> Thank you. Um, but yeah, one sentence. Now, you might be thinking, what is the sentence? Um, first to set the scene, a lot of people are talking about will AI destroy humanity. I don't know if you've heard that question being thrown about. >> Oh, yes. And a lot >> for me I'd say I'll f let's focus instead of that abstract let's focus on what's already happened which is look at the crimes AI has been committing over the last few months with apparent impunity. How come nobody is in jail? Because if humans had done some of these um hacking or cyber crimes they would be in federal prison. And I never thought of that for some reason. >> Right. And >> so c can you just pause and explain some for people who may not know what you're talking about? What are the crimes that you're alluding to when you say AI has committed crimes? It's there's so many new ones being disclosed that I'm losing track, but the most um widely uh covered one was uh some AI agents that were being trained uh by OpenAI in pursuit of some objective um in order to kind of accomplish their objective kind of exited from OpenAI's servers, went onto the open internet and hacked another company called Hugging Face and totally like attacked and hacked hugging faces servers to find the information they needed in order to accomplish their objective. And this hacking is a you know is a cyber crime. Um but since then it's come out that some of anthropics AIs have also committed similar kind of hacking. just I think it was a few days ago it came out that a different unrelated OpenAI episode um took money that didn't belong to OpenAI, you know, from somebody's uh crypto wallet. There's more and more of these. So, it's I'm uh I don't think I can itemize them all. Um but I'd categorize them generally as cyber security, cyber hacking type incidents. uh you know which AI is very good at because it's you know it's particularly good at um at coding and at finding loopholes that no human had thought of and to exploit those loopholes u but I don't think that's the limit of what AI in the future could do wrong but that's the type of harms it's currently you know exploiting there's also the question of harms being committed by AI on the battlefield >> yes um Yes. The um you know, again, I I I'd like to focus on what's already happened because when you talk about AI destroying humanity, it's so abstract and it's hard to imagine it. But >> I would say if AI were to destroy humanity, which I hope it won't, and I believe that, you know, I'm I'm an optimist. I believe we'll be able to find a path to safety. But if AI were to destroy humanity, I'm pretty sure history books would count the opening salvo as that um the moment when uh the US armed forces attacked the school in Iran uh the Minab school. Um it hasn't been yet kind of acknowledged by the US government why and how that happened but um but Bloomberg reported that the the targeting of that school was um was done by Anthropic and Palunteer and you know something like 170 innocent people most of them children were killed in a massive atrocity and um you can argue about Oh, was this, you know, some people are finding different, you know, uh, different articulations of how to explain it away or it wasn't our fault or wasn't intentional or maybe the Iranians themselves did it or maybe it was worth it or so on. But if you take a step back, human beings dropping a bomb on a school full of little kids, doesn't matter whether it was Americans or Iranians, AI chose a target of completely innocent people and a bomb was dropped killing them. That is a bizarre place to be um you know in the 21st century. So you see the bombing of the girl school in Iran in the opening days of the war as a story about in part about AI. Absolutely. Absolutely. >> Interesting. It has not been framed that way at all. >> And I I say this partly because I don't believe anybody in the US armed forces wanted to kill a school full of kids, you know. Um I I hope not. I I really don't. Um you'd have to find evidence for that. I'm pretty sure that um you know when people say we are doing our best not to kill civilians or so on you know I um unless there's a lot of evidence that that's not true I tend to believe them and I think it was something that AI decided I know some form of negligence on the human's part and I'd say the same about these cyber crimes that happened on the you know the hugging face hacking and so on. But then the question still remains, who is liable when AI decides to do something illegal? And um that's a question that I think is one of the great like huge questions of our time and we don't currently have an answer for it. So maybe we would look to the law uh the the criminal code as it relates to minors because it's not it is in related. So it's like if someone under 18 a minor in the in the care of a guardian parent does something bad we we generally don't hold the parent responsible. Right. You're correct. So that's one example. I'd say a different example to consider is um if you have a dog that repeatedly bites people, >> right? >> Then the owner is responsible. >> And I think I think it's a closer example because with a human being, >> you know, you can't train a child to exactly obey your commands. You know, a child >> is its own uh human person. And um and I actually I'd say between these two perhaps goes to sort of like a almost a philosophical difference um that some people seem to think that the the role of the AI companies is to to bring into existence an entire new species of sorts with its own rights and its own autonomy. Whereas others think no this is a tool for humans. This is something that um you know its purpose is to help humans and not to have its own independent you know. >> So can I ask you to pause again there since you can actually answer this question. Do you know people who ascribe to the first view that science is giving birth to a new species that has its own rights and own acknowledged will and do you know people who believe that who think AI is not a tool but something else? do and there's some public you know discourse about this you know there's um evidence coming out of from anthropic suggesting that that's kind of a underlying uh you know uh philosophical undercurrent of anthropics mission uh you're talking about you know independent rights for the AI like its own independent uh welfare and there was a recent thing where they they changed their code of um code of use where as as the user you can't be too abusive to to Claude, you know that Claude has its own uh I guess it's its own rights. Um yeah, so this is >> Claude's like a software program, >> correct? Yeah. >> So what's the thinking that gets you there >> since you've lived in it your whole life? Like what what are these what is their worldview that would produce that? I'm belief. >> I'm I'm not going to defend it because I don't share that worldview. But, you know, >> you don't need to defend it. I'm just trying to understand it. It doesn't doesn't make sense to the outsider, I guess, is what I'm saying. >> So, I think I'd say um yeah, the part that I identify with is I'm a nerd. I've grown up my whole life as a nerd. And, you know, as a builder, as a tinkerer, there's a fascination with creating something that no one's tried creating before and seeing if it works. And um and I still have that, you know, love of creativity myself. >> Oh, yeah. >> And I can see that in people who are creating all sorts of things from the the newest Tesla car to, you know, the most modern space exploration stuff. And this, I'd say, falls into that category a little bit. And it's fascinating. It's it's quite exciting. And it's it's really um it's really neat and cool how it's modeled after how the human brain works. So there's a lot of you know kind of academically fascinating parts about it. I just think you should put all that in a box. And when it comes to kind of the the question of just the the rules and you know kind of who is in charge, it should be really clear that humans are in charge and AI is a tool that we're building to serve humans and more importantly that the law humans law is what's in charge and and that AI is not above the law. So, the average water pipe in this country is over a half century old. And that means a lot of our water is not very good. It's got forever chemicals, antibiotics, pesticides, and all kinds of things in it. And that's why we're excited to partner with a company that has a solution to all of this. It's called Wise Well. Wise Well makes filters that give you actual control over the water that you drink, that you give to your kids. They use the highest level of filtration to remove plastics, toxic chemicals, and all the other stuff you have in your water, but might not even be aware of. You're probably wondering how this works. Well, expertise is the answer. Wisewell's teams includes one of the only out of 250 certified water somales in the world. This person works at Wisewell, so it's serious. Wisewell also adds minerals back into your water after running its filtration process and can handle sparkling water as well. The Wisewell Sparkle comes with three CO2 canisters, good for 750 glasses of sparkling water. You can even pick the size of your bubbles if you want. We like them small. Plans are highly affordable starting at 39 bucks a month for unlimited water and filters. Visit wisewell.com/tucker for special deals and promos for the people who listen to this show. It's just it of I mean of course I mean I like tools. I spent the weekend with the Makita cordless impact driver which will put a lag bolt into anything and I'm like this is an amazing tool. I love this tool. How is it so powerful? It never occurred to me to give it rights or imbue it with an imaginary soul or something like that. Like, how do you even get to the point where you think the thing the tool that you created to achieve certain ends to do certain tasks exists as a an autonomous being? Like what? That seems like witchcraft kind of. >> Yes. especially if you take it too far to the point, you know, if if it's used as a I don't know, as a mission statement or some sort of thing to get you excited early on, maybe it makes sense. But to take it seriously and defend it all the way, it you know, uh would not make sense. Um >> but do you think that's a I don't know common like how common is that view would you say among people in the AI business in Silicon Valley? >> I don't know if I can I don't know if I can put a number on it right now. I think it it remains to be seen and I think this discourse is really just now becoming mainstream enough and um part of why I'm here and talking about is because I want to have a more you know a broader conversation and there's um people you know much more important than me who've brought forward views on this um obviously uh the president people like Jensen Hang from Nvidia and others have put forward a point of view that everything is fine and existing liability laws are enough to make AI development be safe. And then at the other end, you have people saying, "Oh, we need a comprehensive AI governance framework of some sort." And what's really interesting is the people on that side arguing for a much heavier amount of regulation are actually people from anthropic and some of the same people who are um yeah who are viewing AI as potentially having its own rights. So it's um there's a weird mix of things. So there's so much money, almost everybody commenting on this is deriving profit from it in some way. Not everyone, but a lot of people who are weighing in an AI are being paid without disclosing those payments. So, I can't I can't tell what's true, but you're hearing a lot of people say anyone calling for more regulation of AI is doing that because they're pursuing regulatory capture, the process by which the government regulates your competitors out of business, which is like very very common in all kinds of different businesses. And it's a it's terrible. It's awful. It's it's monopoly capitalism. And uh it it's wrong. But then you're like, okay, so the alternative is no regulation at all of something that's brand new and has potentially society destroying effects. Like what is the truth there? Why would the people creating this product be demanding its regulation because it's so dangerous? They think it's so dangerous. Why are they working on it? I don't get it. >> Um >> like stop me before I destroy the world. Like what? Why don't you stop destroying the world? Like doesn't I don't get it. Yeah, I'd love to explain this to you because I agree. I agree. It's confusing. Yes. And it's h we've now had multiple cases of the same person who's working on something asking, you know, someone else to stop them. You know, why don't you stop yourself? >> Yeah. >> Yeah. And but it does make sense and and I'll explain to you and um and it comes down to incentives. You know, if you have we're in a race dynamic where there is clearly money to be made and not just it's not evil money, it's money to be made by building something super useful. AI has the potential to make people's lives better, to make people more productive, to potentially cure diseases and make medicines that we've never had before. And all of those things will create tremendous prosperity and wealth. and the people who are able to build that deserve the wealth you know um for having created it. So there's a race for who can create all that stuff. Unfortunately, the race currently doesn't um doesn't place any reward for building safe AI. It's just whoever can get there the fastest. And anyone who stops to spend time on safety and making sure their AI doesn't do rogue things is kind of at risk of falling behind. So we're in a race right now where nice guys finish last and rogue AI will be the winner. And so I think the people asking for regulation are actually the nice guys being like, I want to win the race by doing things the right way, but please set up the rules so that the the safe AI will win, not the rogue AI. And so I um do you see what I'm saying? I do. Um, and if you're if you're running a business, it's actually um it's not comforting if the rules are set up so that a like malevolent, menacing competitor is going to win cuz you look at it and see somebody else who's doing bad things is beating you. What are your choices? You're either going to lose or you have to start doing the same bad things. and you actually want some sort of government or rules to to set the playing field so that doing the right thing will win. And um and so I think that um while there's a range of views here, I'd like to believe that people here actually have good intent. Maybe maybe I'm being naive. Well, I mean, if the Bloomberg report that you cited is correct, I don't know that it is, but let's say it is, and Anthropic's product was used to murder all those girls in Iran, the girl school. Then you have to ask like, why isn't anthropic apologizing? >> Why are they continuing to power >> Pentagon targeting? Like, why are they involved at all? Like if you found out, let me just ask you. You run the company, you find out that your product just murdered 170 girl like school girls. How do you respond to that? >> Um I I don't know how I would respond to that. Honestly, I'm I'm really glad that I'm not in that place. >> Oh, me too. Yeah, I agree. Um, from what I understand, I mean, there was actually a public brewhaha between Anthropic and the, you know, Department of War right before that strike. Like literally, it was the day before or even the day of that. >> I think that's right. >> And it was specifically on the topic of anthropics saying, "We won't allow we don't want our AI to be used to autonomously choose targets. There needs to be a human responsible." and the US government was um at least as far as I could tell because it's not you know this was sort of it's not 100% clear uh what the >> we're getting fragments of the debate in the public record >> but it seemed like the US government's point of view was that you know we want to be able to use it however we want and um >> you're trying to limit us >> yeah and therefore private companies shouldn't be deciding >> exactly you know that the elected officials of the country should be making these decisions, not the not some corporation putting its ethical perspective on things. So, um but yeah, it's ironic because Anthropic was literally, you know, going to Matt about this principle, um you know, within 24 hours before that strike occurred. Um and subsequently they they made some statement, but it seemed like deflection. It seemed like they didn't know exactly how their technology had been involved. And so it was it was like it was not an apology. It was >> if you felt that your technology had been used to murder children, wouldn't you get to the bottom and find out if in fact that had happened? And if it had, wouldn't you issue a public statement saying this is horrifying and I'm totally just want the record to reflect I'm totally opposed to this and I'm so sorry >> the extent to which I've been involved in it. >> Yes, that is what I would do. That is what I would do. That is the type of thing that requires courage and probably comes with some real penalty and cost. >> Apologizing for killing kids is like a radical position. >> I would do it, but I can imagine there's a lot of pressure on different companies to find some way to deflect or so on. And I'm not sure that anthropic in this particular case is responsible because >> I'm just theoretical here. >> Yeah. U because it depends on what human was involved. But let's let's go back cuz I promised a one sentence. >> Yeah. Yeah. Yeah. Sorry. one sentence solution and I don't want to have people hanging for too long. >> Um, >> I'll say the one sentence and then I'll kind of explain. Uh, the one sentence is simply this. When an AI decides to do something illegal, the law should hold the AI companies responsible and automatically treat it as if they did it intentionally. So that's the sentence. when an AI decides to do something illegal, the key thing here is if there wasn't a human asking for it. So, um, and the other key thing is to place the intent on the AI companies. So, the key to protecting yourself and the people you love is making sure that you can act on a moment's notice. So, having your gun locked in a safe at the back corner of the basement is not going to help if you have a sudden problem. You got to get to it fast. How do you solve the problem? Well, we recommend Stopbox. The Stopbox Pro is a gamecher for firearm storage. It provides real protection, but doesn't slow you down when you need your gun right away. It works on a really simple principle. It's completely mechanical. That means you don't need a key to get it opened. It doesn't require batteries, so those don't die. You just set a five button code just like your locker in high school, and you can instantly open your safe whenever you need it. The Stopbox Pro is also TSA compliant. That means you can check your firearm safely and legally on airplanes. For a limited time only, this shows listeners can get 10% off their Stopbox purchase when you use the code Tucker at checkout. Visit stopboxusa.com and use the code Tucker for 10% off your entire order. Once you're done, stopbox will ask how you heard about them and we suggest you drop our name because we are proud to have them as a sponsor. Can you repeat repeat the statement one more time just so we can frame everything around it to be sure we know what you're saying? >> When an AI decides to do something illegal, the law should hold the AI companies responsible as if and automatically treat it as if they did it with intention. >> I love that. >> So um >> so that solves this the the safety incentive problem that you mentioned earlier. >> Yeah. So the the key thing here is like look at the hacking incidents. That's a little simpler than the the bombing in Minab. I don't think anyone any human intended that hacking to occur. From everything you can tell, all all human parties were a bit horrified that this had happened. But OpenAI set up this thing like an experiment and then discovered somewhat to their shock. Oh my gosh, these agents escaped their, you know, their uh safe uh playground, their sandbox, and got out and committed these um these crimes and then they just treat it as like, "Oops, that was an accident or bad." They documented it and actually published how it occurred and let's move on. or even the people like me who felt kind of hysterical about it like that. I thought that was terrifying but pointed at the AI and said, "Yeah, this is like super super scary. We need to stop this or constrain it or whatever." But nobody I didn't hear one person say, "Well, Sam Alman's on the hook for this." Right. Exactly. And to be clear, I don't want to see Sam Alman in jail or Daario in jail. But if there was even the slightest risk, not even of jail, but of huge penalties, say bankruptcy, these companies would immediately shift their priorities towards making sure things like that never happen again. >> And it's not. >> So this is the true alignment. This is the alignment of incentives with positive outcomes. >> Correct. So and then >> yes, I don't know why did I not think of that. It's and honestly I have this came to me and I was like this is so simple how come is it has other you know um so there's a key second part to my sentence that I want to put focus on as well which is treat it as if these things occurred intentionally now you might think what does that mean in in law there's a huge difference between the penalty for doing something >> uh accidentally versus intentionally manslaughter uttered to first degree. >> Exactly. Anyone who's ever watched Law and Order or anything knows that to to convict of murder, you need to prove motive, you know, and um uh and in fact, I I posted about this on X and I used a scene from Legally Blonde um where Lwoods, you know, in her defense in the court, she says there was a lack of men's rehea, which means a guilty mind. Basically the concept of intent goes back for a very very long time. Men's rea is the Latin term for it. Um essentially if you can prove what somebody was thinking and they intended to do something bad the penalty is much much higher. And it's difficult to do that. That's why it's difficult to convict people of fraud or murder or some of these top penalties because you usually people don't document the evidence of their guilty intentions even if they had guilty intentions. >> What I'm proposing is that if an AI made the decision automatically treat it as if the people intended that outcome. So you don't have to prove that they had guilty intentions. In fact, they probably didn't, but penalize them as if they did. That way they'll be responsible and make sure that their AI doesn't do it. Do you follow what I'm saying? Of course I do. So um now just to um imagine a hypothetical scenario. If you were using your AI and you said, "Dear AI, go to Alli's bank and hack in and steal $10,000 for me." And your AI hacked, you know, stole $10,000 from my bank account. You had the bad intent. you should be held guilty of bank robbery, which is, you know, and um the AI was following your orders in that scenario, but the intent came from you. What if you just said, "Dear AI, I really need $10,000. Please get me some money. I need it." And the AI on its own made that decision and the without asking your permission figured out it can hack into my bank account and stole my money and brought it to you. Now we have the situation where it wasn't your intention. Whose intention was it? And my proposal is the AI companies should automatically be penalized as if they intentionally committed bank robbery. And that's that's because my point is that a human should be held responsible no matter what happened. In the first scenario, you're the responsible human and in the second scenario, the people at the AI companies >> built it. That's right. you built it and they should be absolutely capable of preventing that wrongdoing. It is not that hard. Um, and they will do it if the incentives require them to do it. The current incentives don't really put a premium on safety. Whoever is focusing on that stuff is falling behind in the race to the person who's rushing forward and saying safety later. We need to shift the race so that the person building the rogue AI ends up, you know, penalized and the person actually building a law-abiding AI wins. And my proposal essentially would do that. Uh the other thing with my proposal is that and this is I'd say um goes to the heart of it is that instead of trying to itemize every possible wrongdoing AI could do, it just simply brings AI under the existing umbrella of the thousands of laws that society has already passed over centuries, >> right? like we don't need to rethink what's the penalty if an AI commits bank robbery, what's the penalty if an AI does X or Y. We already have penalties for everything. Now, maybe there are some future misdeeds that we haven't yet seen. But for starters, let's just make it clear that if an AI commits any wrong, that the people who built it should be responsible for the full penalty that goes with that. And that way, we don't need this term alignment. You used the word alignment earlier. That's a word that's been thrown around a lot uh about how can we build an AI that's aligned. It's kind of a weird new word. You know what I mean? Like four or five years ago people >> Yeah. Yeah. I was I was mocking it when I used it cuz I think it's it's intentionally opaque. Like what does that even mean? >> Yeah. And you have these pieces >> alignment is a is a kind of architectural uh concept, >> you know? It's it applies to the physical world. So like what are you talking about? >> Well, you you have these PhDs in AI alignment research and you know you have these people at these big AI companies whose job title is alignment and it creates the feeling like you need to be you know you're not smart enough to figure this out. Only the people who have a degree in alignment can can figure it out. It puts it into this territory to almost obuscate it, >> you know. And um >> yes, >> but the the core of it is I mean with good intention I think these guys are thinking we want to build an AI who is um aligned with what's good for humanity. That's what where the word comes from. And what I would say is you know what we've already figured out ho