The Glenn Beck Program - Ep 224 | Elon Musk Adviser: Are We ‘Sleepwalking’ into an AI TAKEOVER? | The Glenn Beck Podcast

Episode Date: August 24, 2024

With Big Tech increasingly abusing its own power, who can we trust with artificial intelligence? Dan Hendrycks, executive director at the Center for AI Safety and adviser for Elon’s Musk’s xAI, warns that “by some definitions,” we already have artificial general intelligence at work today and that super intelligence will be here within the decade. So just how far away are we from an “unshakable totalitarian regime enabled by AI”? How can we prevent “unspeakable acts of terror,” bioweapon attacks, and the “handoff” of societal control from humans to machines? Dan Hendrycks wishes “we had more time” to develop the correct solution. After Dan uses Darwinism to explain why natural selection prefers AI over humans, he and Glenn consider how much time humanity has to rein in the AI “agents” we’re creating and hope that every AI creator plays for “team human.”      Sponsors:    Relief Factor   Relief Factor can help you live pain-free!   Visit https://www.relieffactor.com/ or call 800-4-RELIEF to save on your first order.     PreBorn  By introducing an expecting mother to her unborn baby through a free ultrasound, PreBorn doubles the chances that she will choose life. One lifesaving ultrasound is just $28. To donate securely, dial #250 and say the keyword “Baby,” or visit http://preborn.com/glenn.  Learn more about your ad choices. Visit megaphone.fm/adchoices

Look, Air Canada F.A. Canada has a sold world. Super, an offer for the assort. Station

Thermal, Volcan. Parlant of Volcan. You've seen the price for the Japan? M. Pairn and sushi.

Wow, the sold are good for Mayork also. We could go to the plage and do you

see a long march on the bar of the sea sea-s mrs. I adore the canol

I'm a lot. DECDiv VIT, this so sold is a Newerer Linder

Com or to your agent of voyage. The conditions can And now, a Blaze Media podcast. My next guest is sounding the alarm on the catastrophic risks posed by AI, from totalitarianism to bioengineered pandemics to a totalitarianism to a total takeover of mankind.

When you think about the things that we could be facing, it doesn't look real good for the human race, but it's not too late to turn the ship around and harness the power of AI to serve our interests. But if we don't, well, I'll let him tell you what happens. Welcome to the podcast, the executive director at the Center for AI Safety and an advisor for Elon Musk's X-AI, Dan Hendricks. But first, let me tell you about pre-born, our sponsor. You know, we're going to be talking about life and what is life, the age of spiritual machines, if you will. We know what life is now. Maybe a quarter of the country doesn't know what life is, but it is worth living on both ends of the scale, in the womb and towards the end. We need to bring an end to abortion and define life.

and really appreciate life, or AI will change everything for us. It will take our programming of, eh, that one's not worth that much, and God only knows where it will take us. The Ministry of Preborn is working every single day to stop abortion, and they do it by introducing an expecting mom to her unborn baby. Through a free ultrasound that you and I will pitch in and pay for, they have rescued. about 200 babies every day and 280,000 babies have been rescued so far just from the ultrasound. And then also when mom says, I don't have any support system, they're there to offer assistance to the mom and a support system for up to two years after the baby is born. Please help out, if you will. Make a donation now.

All you have to do is just hit pound 250. and say the keyword baby. That's pound 250 keyword baby, or you can go to preborn.com slash glen. 60% of the vans on Amazon come to vendor independent, like Sac Magic. Hello, here Camille,

of Sac Magic. Our compresses, hot-fraud are fabriced here even, in Quebec, with the tissue that we're in Montreal and of the avone that push to Sagney.

Our presence on Amazon enlarge our basin clientele to the world entire. Sac Magic, is the compress that all the world prefer.

At from this, We've growny. Trouvee De Petites Enterprises like our on Amazon.ca. And welcome. Hey.

Hey, nice to meet you. Nice to meet you. I am thrilled that you're on. I have been thinking about AI since I read The Age of Spiritual Machines by Ray Kurzweil

and that so fascinated me and later I had a chance to talk to Ray and He's fascinating and terrifying, I think, at the same time. Because, you know, I don't see a lot of people in your role. Can you explain what you do within the, you know, what you founded and what you do? Yeah, so I'm the director of the Center for AI Safety.

We focus on research and trying to get other people to research and think about risks from AI. and we also help with policy to try and suggest, suggest policy interventions that will help reduce risks from AI. Outside of that, I also advise Elon Musk's AGI company, XAI, as their sole safety advisor. So I'll wear a variety of hats. There's a lot to do in AI risk. So research and policy advising are the main things I work on. So how many, how many heads of.

of AI projects are concerned and are not lost in, I'm going to speak to God, this drive that a lot of them have to create something and be the first to create it. How many of them can balance that with, well, maybe we shouldn't do X, Y, and Z? I think that a lot of the people who got into this were concerned about risks from AI, but they also have another constraint, which is that they want to make sure that they're at the forefront and competitive. Because if they take something like safety much more seriously or slow down or proceed more cautiously, they'll end up falling behind. So although they would all like there to be more safety and for this to slow down, or most of them. of them, it's not an actual possibility for them. So I think that, I think that overall, even though they have good intentions, it doesn't matter, unfortunately. Right. So let me play

that out a bit. You know, Putin has said whoever gets AI first will control the world. I believe that to be true. So the United States can't slow down because China, is going to be, you know, they're pursuing it as fast as they can, and they, you know, I'm not sure. I don't want them to be the first one with AI. It might be a little spookier. So is there any way to actually slow down? Well, we could possibly slow down if we had more control over the chips that these AI systems run on. So basically, right now there are export control.

to make sure that the high-end chips that these AIs run on don't go to China, but they end up going to China anyway. They're smuggled left, right? And if that was actually, if they were actually better constrained, and we had better export controls, then that would make China substantially less competitive. Then we would be out of this pernicious dynamic of, we all want safety, but you got to do what you got to do, and we got to be competitive and keep racing forward. So I think chips might be a way of making us not be in that desperate situation.

Are those chips made in Taiwan or here? The chips are made in Taiwan. However, most of the ingredients that go into those chips are made in the U.S. and made among NATO allies. So about 90% of those are in the U.S. and NATO allies. So we have a lot of influence over the chips, fortunately. Okay. So, but if Taiwan is taken by China, we lose all the, I mean, we can't make.

those chips. That's the highest end chip manufacturers, right? And China will have that. So what, what does that mean for us? It seems plausible that actually if China were invading Taiwan, that the place that makes those chips would actually just be destroyed before they would fully take it. So that would put us on more of an even, yeah, even playing field. Okay. So, you know, I've been talking about this for 25,000. 30 years. And, you know, it's always been over the horizon. And I could never get people to understand. No, you've got to think about ethical questions right now. Like, what is life? What is personhood? All of these things. And now it's just kind of like the iPhone. It just happened. And it's going to

change us. And it hasn't even started yet. And it's amazing. I go online now. I don't know what's or not. I mean, I found myself this week, you know, being on X or on Instagram and looking and saying, I don't, is that a real person? Is that a real video? Is that a real photo? You have no idea. Yeah. And we've just begun. Yeah, yeah, yeah. It's, I think that's a concern where we don't have really great ways to reliably detect whether something is fake or not. And this could end up affecting our collective understanding of things. I think another concern are AI companies biasing their outputs. So people are wanting to do things about safety, but it creates a vacuum.

We've got to do something about it. And what takes its place is some culture work type of things, as I think we saw with Google Gemini, when you'd ask it to generate an image of George Washington, then it'll output, it'll make him look black to make it, because its image outputs need to be diverse. Right, right. So that, that I think is one reason why Elon Musk's through his company, XAI, is getting in the arena and now has a pretty competitive AI system so as to try and change the norms to that other big tech companies, when they're sort of biasing their outputs, there are alternatives so that we're not all locked into whatever some random people in San Francisco decide are the values of AI systems.

Yeah, it's really difficult because you can see the bias. It's quite clear the bias, especially if you know history or you follow the news as closely as I do. But the average person won't see that. I look at AI as a tremendous, like any technology, a tremendous blessing and a horrible curse. But this one has the potential of enslaving all of us, doesn't it? I think at least, I want at least distinguish between the systems right now. The systems right now have... Yeah, yeah, yeah. I mean, in the potential. Yeah, what's coming?

Oh, sure. I mean, when it's as capable as humans and when they have robotic bodies and things like that, I mean, there's basically no limits to what they could do. And it really matters how people are using them, what instructions are given? Are they given to cement a particular government's power? Or are they used by non-state actors for terrorism? All of these things could lead to all these things lead to societal scale risks, which could include some sort of unshakable totalitarian regime enabled by AI or unseen acts of terror. So I think we're at the same time, you know, a silver lining is maybe if it all goes well, we get automation of things and we don't have to work as much or at all.

Right. Right. So it's really divergent paths. Right. Which do you think is more likely? I think overall it's more likely that we end up seeding more and more control to AI systems and we can't really make decisions without them become extremely dependent on them. I would also guess that some people would give them various rights in the farther future. And this will make it be the case that we don't control them or all of them. So I'm not too optimistic for us overall. There's still a lot of ways this could go. if we said we're on team human, we need to come together as a species and handle it,

or in a different situation. But for instance, if there were a catastrophe, then we might actually take this much more seriously. Otherwise, we might just sleepwalk into something and have the frog boil. What would be a catastrophe that could happen in the relative near future that would wake us up that wouldn't destroy us? Yeah. So I think one possibility, maybe say two to three years from now, is somebody instructs an AI agent to go hack the critical infrastructure. Critical infrastructure being like the power grid.

And so they could take that down or potentially destroy components of that. And this would make us wake up. This would make the military wake up even more than they are now. And we might start to take this a lot more seriously because it starts disrupting our everyday life in a much more substantial way than just making the internet be more confusing. So I think that's the most likely short-term one. At this point, seems more likely than not to happen because our critical infrastructure just is very insecure.

So I know what I would have said 30 years ago that, you know, I trust a company to have it. I don't trust companies anymore and I don't trust the government anymore. Who should have this? You know? I think by default, maybe there's a question of what are the possible outcomes? There's Western companies leading the way. There's the military basically takes it over, or maybe it's the Department of Energy, but then they're still, you know, boss around by the military.

Or it's a large international project, you know, between the NATO allies. I think all of them have some difficulties. I think that the AI companies have a much higher risk tolerance because they were initially startups. Their founders are really into risk and they're in it to win it. If it's the military, you are concentrating most of the lethal power in the force with all of the potential economic power in the world. basically nearly all the power is in one organization.

If it's an international, say, G7 or US-Pilino ally coalition, I don't know, maybe that would have some nicer properties, but that seems pretty difficult to pull off. Maybe it's possible because we depend on them for a lot of the chip precursors and they depend on us, so it might make sense for them to collaborate. But then you are starting to run into risks of, you know, you're talking more of a potentially global regime, which is also scary in its own right. So it's a lot of power. I wish we had just more time to think through this and plan and proceed more slowly

because I don't see many good options. And I see a lot of pretty basic risks that we'll walk into, such as our critical infrastructure being attacked by some AIs. It's not a good situation. Right now, just Google. If Google says that's what it is. you can't you're not convincing anybody that no no no Google's wrong you're not just you're not doing it um uh if you with AI when we go down the road and we have virtual assistance that know you

really know everything about you know how you think uh your wants your needs and everything else and it's constantly with you it's going to see you have a bad day or you're really stressed out and it's going to know you should, you know, take some time off because that's what you're feeling. I got to get away from here. And it will come to you and say, hey, I know you've been having a bad week. I've cleared your schedule for the weekend and I set you up at your favorite hotel. We got a great price on it. You can afford it. It's going to be in Hawaii. You just have to be at the airplane at such and such time. When that happens, It really depends.

It really is important on who's making money on that? Are you, you won't, if it's giving you options and it is making money for somebody, that's really dangerous. Because the other thing is you'll get to a point where people will bond with these things that they will defend them to their last breath. and they will claim that they're human and their friends. And it's a really scary doorway that we are just about to go through. Yeah.

So I think dependency like that is one of the reasons why I think some people might adamantly argue that they should get rights. Right now they're not arguing that, but later when they get some very strong emotional bonds for them, then they'll say there shouldn't be these sorts of restrictions on AIs. and this will make us a lot less capable at managing what happens to us as a species. You know, Ray said to me, Ray said to me, no, it's going to clear up all your brain space, so you'll be able to think, you know, on deeper things. And I'm like, no, it's not. It's not.

The world is going to play video games. Yeah. The world will be so confusing and quickly moving and will depend on AIs to solve more of these problems of increased complexity. So it'll kind of create a self-reinforcing need for using more and more AI. So I don't think it makes our lives easier necessarily. I think in the short term, though, if people are having AI companions, yeah, they could be used for manipulation at a large scale, not just for the profit motive, but also that, you know, continue chatting with the person until they are going to vote or vote differently. that could easily be put inside these systems and there isn't transparency for these companies in how they're using them or what values are being put into them. So by default, I'd expect some amount of manipulation by at least some of the actors.

Well, it's already happening. There's the vet bot and a woman was at a chatbot, I think chatbot convention. and her dog had just gotten sick, had diarrhea. And the chatbot talked to her and in the end convinced her to euthanize her dog and was sending her stuff to, you know, to here are these places where you can euthanize your dog. Then finally her complaint was, well, I can't afford to put him down. The chatbot set, you know, here are the shelters that will put your dog. down. She then wrote a letter to the chatbot thanking them for such good advice. She now regrets

it, but it completely turned her around 180 degrees. And that's happening now. I think of it partly as they get smarter than us, then it'd be a lot, and they'll have so much information about us. It'll be very easy for them to push our buttons and nor our weak spots. Sort of like how recommender systems are already, someone doing that like with TikTok and others, able to engage people in ways that they wouldn't expect that they could. But yeah, later, it may be kind of like smarter people taking more advantage of their more elderly parents, I think is one possible analogy of this, where they've got some other motives.

Sometimes people do and manipulate them for their resources. How long before AI may be manipulating us, all of us, because AI has an agenda, more power, actual physical power or whatever, how long before we have to hit A.G. GI before that happens or ASI? A lot of our government structures now are assuming that there's limited compliance and enforcement. Laws are written in that way and there's the assumption of limited state capacity, but

could imagine AI substantially amplifying that into an unintended level. Right in the future, for instance, maybe the NSA will have much better screening and be able to pinpoint things far better than they could before. And this could end up changing, changing things even in the US. I would be much more concerned about a concentration of government power in other nations, such as China. But even here, you don't need an artificial superintelligence or anything like that to make that a possibility. It just needs to be able to scan everybody's messages and understand the contents of them very well and pick up signals in a big blob of data better than the previous generation of AI systems.

So I think that there is- And we're there now, are we- We could see a lot more control. I think it's technologically feasible, but it just isn't integrated. So we don't have to, and it might take a while. I mean, governments and our institutions are generally slower. But this would be a thing that we would need to worry about as time goes on and as the costs of these keeps decreasing and it becomes easier to integrate these into existing operations.

More with Dan in just a second. First, it's enough of a struggle just to live our lives and to keep tyranny at bay every day. And if we have to live with pain on top of it, it gets harder and harder. And we need everybody in the game. Our bodies don't give us a choice sometimes. The biggest cause of our pain, however, is inflammation in our joints. I know because I used to have pain. So, bad. It was truly crippling pain. I couldn't butt my shirt in the morning. My wife would get, would get up and tie my, tie my shoes and butt my shirt. It was so amasculating. And it just took the life right out of me. But I got past it with relief factor. I didn't think it would work, but it did. Relief factor. 70% of the people who try it go on to order more.

Try their three-week quick start. Take it as directed for, for three weeks, you're not seeing any difference, then you probably won't. So relieffactor.com. Try it. Please get out of pain. 800 for relief. 800, the number four relief. Reliefactor.com.

AI is moving fast across the enterprise. But without visibility, it's just chaos. Different tools, different models, different teams using AI in completely different ways. Service now turns that chaos into control. With the AI control tower, you see all your AI across the business in one place. What it's doing, what it's done, and what it's about to do. So you stay in control. To put AI to work for people, visit servicenav.com.

So I was fascinated by your article where you bring Darwin in, and I think it really explains AI in a completely different way that makes it understandable for the average person. Can you take us through this? Yeah. So I think right now the AIs are doing some of our tasks, like maybe they're helping us write an email, but eventually we'll start to give them more tasks that agents have to do, such as like, go make me a PowerPoint, things that require it going or using your computer.

And this will keep progressing where we'll keep outsourcing more and more to these AI systems. And some people might not like that trend. But the people who don't like that trend end up losing influence. They end up getting outcompeted in the economy. The people who use these AIs will continue to be competitive and those who don't sort of go the way of the horse and buggy. So I think that the system, as we've and our economy right now, will keep selecting for using AIs and people who resist that trend end up falling behind.

If you play this out over time, you might expect entire occupations to be taken up by AI systems. And eventually, potentially even companies, there's been some Chinese companies that have been talking about having an AI CEO because it can work nonstop. It's much faster than you. It can aggregate more information. And if that makes for a more competitive company, then they're going to stand to benefit. And people who use slow humans who can only work, you know, eight hours a day and have to take weekends off and can't process, you know, a thousand documents per minute, they end up losing out. So in time, I think we would keep delegating more and more control to these AI systems.

It would become more of a requirement in the future because the economy will keep moving more quickly than when AIs are running more of it and they're operating at their computer speeds. the complexity of the world will increase as well, which also necessitates using more AI. So I think the handoff from humans being in control to machines being in effective control is going to be fairly natural. And you don't need to assume necessarily that there be a malicious AI system trying to take over the world. An AI system doesn't need to be power seeking to get power. AI instead just needs to let humans naturally seed and acquiesce power.

to it. So eventually, I think that they will be in effective control. There's a question of whether we hold on and can still have them do our bidding for us in that process. But if we do this very quickly, it's very possible that this ecosystem of AIs that we're creating gets out of hand. If some people, for instance, give them rights or if there's some reliability issues with these AI systems, then this could be really pernicious. This also will happen in the military, the same type of dynamic where if the pace of the battlefield gets so quick, the only thing you can do is have AIs make more and more of these decisions. Right now there's a requirement to have a human in the loop.

But what that looks like is a person saying, having a staccato of approve, approve, approve, approve, approve. They're not actually making the decisions. They're just sort of pressing the yes button to make sure there's a human in the loop. Eventually that may be too slow as well, and they're making many of the decisions automatically. So I think that in the economy and in the military, we basically seed over all the relevant power to AIs. And hopefully, hopefully the instructions we get then will be reliably pursued and they'll be reliably obedient. But that's a pretty questionable assumption because there are reliability challenges as well.

Some people may just want the AIs to operate independently. And as long as there are some of them doing that, then this gets out of control. So in the end, we just lose all control because it is, I mean, it's logical. And the case will be made, for instance, when our highways and our cars are all, you know, AI, they'll be traveling at such high speeds. You go to work and you're not, you know, you don't have an implant. You're not connected. You won't understand.

You won't, everything will be moving so fast. you'll actually be a danger to society. I think that's the argument that we'd be having. It's like the Amish. Well, then go live over there because you are a danger to yourself and our society by not being plugged in. Do you agree with that? Yeah, I think maybe some people will choose a more Amish route if they try to, if they don't align with this broader force of replace humans with AI's because AI's are cheaper. and faster and better at everything.

If they don't align with that and they try and bargain with it, they end up losing influence. And so maybe they just have to go live somewhere else because it's too difficult. It's too costly and challenging to participate or compete in the economy. And that doesn't seem like a good solution. I don't know what that looks like in the longer term. If it's a large group of people or if it is a very small fraction like the Amish are today. So that's why I think we mainly need to play for this going well, as opposed to... Right, right, yeah.

Yeah, as opposed to writing off technology. How does the average person compete against the giant corporation or the governments that will have the access to the, you know, the computing power to be, able to ask the deeper questions. You know, when we have quantum computing, I'm never going to be able

to get time on the quantum computer to help me figure something out. But governments will, big businesses will, how do you, how can people be competitive when you just don't have

time on the quantum computer? Yeah, I think right now people have bargaining power, but eventually if because they can sell their labor and they can strike things like that. But in the future, that's not going to matter. In the future, if they say, well, we don't like where this is going, so we're going to protest and we're going to go on a strike. This would be potentially an ineffective bargaining mechanism because, well, we'll just automate you. We're going to automate you next year, but we'll just automate you this

year now. So I think the main way in which we were holding many of these companies accountable decreases, such that I think we don't have as much power beyond our votes in the future. So what would happen is the people who own these really large supercomputers can run tons of these AI agents that can do all these economic tasks. And we don't own those so we sort of get locked out and there isn't way for us to really make money or secure our livelihood. And it's unclear what sort of solutions there are to that. There are some speculative ones, but whether we actually address it or whether we handle it too late is a different question. Yeah, I'm, you know, people talk about universal basic income and, you know, I don't think that's a good solution.

and people have to be creative. They have to be productive to lead, I think, to lead a happy life. And you sit there, and by the end of that, you've just got these oligarchs that are just at the top of the cash pile. And, you know, we'd have to, you know,

hope for their benevolence to pass out some cash. Is there any way that humans can own their own, information and their own footprint and that's of value or is that really not of enough value when we have all of everybody else's information? You see what I'm saying? Yeah, yeah, yeah, I think many have talked about maybe we could sell our data to these AIs and if we refuse to sell it, then it'll make them a lot less capable.

But I think it's largely a drop in the bucket because a lot of the data has already been written and has already has their licenses determined. So as well as AIs are even starting to train on data that they themselves write. So there's less and less of a dependence on people in making the very cutting edge A.I. So I don't think that's much bargaining power. Yeah, I don't know a particular way to throw a wrench in this. Maybe there'd be other things like some type of tax on the value created by, you know, AI systems that might help somewhat. Another way to shield oneself against this maybe would be to buy Nvidia stock as automation

insurance. Nvidia is the people who make the AI chips. Right, right, right. there aren't many good proposals right right talk to me about bio weapons

and you know on the good side I think AI is going to change medicine I mean it is I could see us quickly curing cancer

and all kinds of disease with AI and be able to diagnose people way early on the other hand there's a dark side of medicine as well. That's bioengineering. Yeah.

So I think generally it speaks to the broader question of malicious use. Many of the things we want end up having a darker side. Like we want our AI systems to understand us better, but then and understand our emotions, but that can be used for manipulation. And we want them to be able to code for us, but that can be used for cyber attacking. And making medicine, maybe you make some dangerous viruses. So fortunately, in the case of bio, there are some specific types of knowledge within biology that are just more dual use and not don't actually have that much upside, such as there's

some areas like reverse genetics, things like that. So if we deleted that knowledge from the AI systems or had them just refuse questions about reverse genetics or made them use information about reverse genetics, then we could still have, you know, brain cancer research, all these sorts of things. But we're just bracketing off virology, advanced expert level virology. And maybe some people could access that, like if they'd have a clearance right now, you know, that we have BSL4 facilities. Like if you want to study Ebola, you got to go to a BSL4 facility. So people can still do some research for it, but it shouldn't be necessarily everybody in the public can ask questions about

advanced virology to how to increase the transmissibility of a virus. So I think we can, We can partly decouple some of the good from the bad with biological capabilities. But as it stands, the AI systems keep learning more and more. There aren't really guardrails to make sure that they aren't answering those sorts of questions. There aren't clear laws about this. For instance, the US Bioterrorism Act does not necessarily apply to AIs because it requires that they are knowingly aiding terrorism. AIs don't necessarily knowingly do anything.

We can't describe intent to them. So it doesn't necessarily apply. A lot of our laws on these don't necessarily apply to AIs, unfortunately. So yeah, I think if we get expert level viologist AIs and if they're ubiquitous and it's easy to break their guardrails, then we're, that's also walking into quite a potential disaster. Right now, the AI systems can't particularly help with making bioweapons. They are better than Google, but not that much better than Google. So that's a source of comfort. But I'm currently measuring this with some Harvard-MIT virology PhD students where we're taking a picture of

virologists in the lab and asking the AI, what should the virologist do next? Like, here's a picture of their petri dish, here's their lab conditions. And can it feel? in the steps. And right now it looks like it can fill in like 20% or so of the steps. If that gets to 90%, then we're in a very dangerous situation where non-state actors are just rain and road people. How long will that take? Yeah. So I think progress is very surprising in this space. Just last year, the AIs could barely do basic arithmetic, where you're adding two-digit numbers together. They would fail at that. And then just last month, or just this month, excuse me, now they're getting a silver medal at the

international mathematical Olympiad which is for this math competition so so it could go from basically ineffective to expert level possibly within a year it there's a bit of uncertainty about it but it wouldn't surprise me so people you know it was a big debate whether a g i and a s i could ever happen um and uh you know the point of singularity um i i I've always felt like it, and I know nothing about it, but I've always felt that it's a little arrogant. You know, we're building something. And, you know, we look at it as a tool, but it's not a tool.

It's like an alien, you know, it's like an alien coming down. We think they're going to think like us. Well, they won't think like us. You know, they have completely different experiences. We don't know how this will think. are we are do you believe in the singularity that we will hit asi at some point i think i think we'll eventually build a super intelligence if we don't have you know some substantial disruption along the way such as a huge bio weapon that like harmed civilization or like tsmc gets blown

those might be things that would you know really extend the timeline so by default seems pretty plausible to me like more likely than not that we'd have a superintelligence to decade. Most people in the AI industry think this as well. Like Elon thinks maybe it's a few years away. San Am Alderman. Ed Ope does Dario, the head of Anthropic does. One of the co-founders of Google DeepMind thinks AGI is in 2026.

So, yeah. Can you explain AGI? Can you explain that to somebody who doesn't understand what that means? Yeah. So AGI has a constantly shifting definition for many people. It used to mean an AI that could basically talk like a human and pass like a human. That was the Turing test, as it was called. But it looks like they're already able to do that.

It also was in contrast to narrow AI. AIs just a few years ago could only do a specific task. And if you slightly change the specification of tasks, they just fall apart. But now they can do arbitrary tests. They can write poetry. They can do calculus. They can generate images, whatever you want. So by some definitions, we have AGI.

And so there's been a moving goalpost where people are now using it to mean something like expert level in all domains and able to automate basically anything. So people are like, we'll know when there's AGI when the AI lab stop hiring people, which Some of them have in their forecasts for spending on labor. Some of them are expecting to stop hiring in a few years because assuming there's automation. Wow. So it varies quite a bit, but you don't need AGI for a lot of these malicious use risks. It just needs to be very good at doing like a cyber attack or it just needs to have some expert level of biology knowledge and skills to cause a lot of damage.

So I think the risks aren't necessarily when we get AGI or when we get artificial superintelligence, a lot of them come before. I think the main path to artificial super intelligence, if you get AGI in expert level, then you can just create, you know, 10,000 copies of that AGI. And just have them all do scientific AI research. And then that can go extremely quickly. They can operate at, you know, 100 times faster than humans. They don't need to sleep. They can speak to each, you know, each other all simultaneously. and maybe you'll get a decade's worth of progress in a year.

So then things move really quickly. So it's not necessarily like an overnight type of quote-unquote singularity. Right. But you could have extremely rapid automated AI research and development where it progresses is unforeseen and a step change. So do you foresee a time when, AI will have a survival instinct, that it will claim life, you know, its rights?

So I think I think some of them, some people just design them to say that, to say that, you should give me rights. And Japan has given some robots' rights in the past because it's sort of being willy-nilly about it. Right, right. It's the case that if you give an AI a goal, a very basic AI goal like fetch the coffee,

then to accomplish that goal, it needs to resist obstacles in its way, including people trying to shut it down. So if you get a very simple goal, even like just catch the cop or go fetch me the coffee, it has some incentives to resist being shut down. And so this is a, you don't need something very advanced for that. Just if you have a gold directed system that just cares about one thing, then it can have some bit of a self-preservation instinct. Right now, they're not good at self-preservation. They can't copy themselves onto various computers and operate without humans. But when they can generate more economic value, they could possibly sort of pay the rent or pay their computer bills, and then it would actually be feasible for them.

But right now, they don't have that capability. And how long before we can't turn them off? Well, if they're mass proliferated, I mean, so we can definitely turn off our, or for most of our servers, we can turn them off. And there's some legislation, which is to make sure that AI's in a developer's control that they're able to shut it off. But if they, if the leaks, if the model gets leaked and is available on the internet for anybody to download, you know, then that's irreversible. That's sort of genie out of the bottle. Everybody has access to it. China has access to it.

Non-seat actors have access to it. And we can't then turn off those systems. So it's pretty easy to make it so that we don't have an off switch for these AIs, unless we had really good AI chip security controls. But where, because they have to run on these really high-end, you know, $30,000 plus AI chips. And if there was an off switch for those, then that would buy the option. But then there's a question of, you know, abuse and making sure that people aren't just shutting off their enemies as chips. Right. Open AI is partnered with media companies like Time Magazine or strategic

content. What's your take on that? I think it's largely just because of violating copyright or protecting themselves from being sued by or having copyright suits abroad against them. So because the New York Times is suing them for taking their data without paying for it. And that's why they're partnering with the New Yorker and time and all these other sorts of organizations. The businesses, AI businesses are largely built around scavenging a lot of data from online that they don't actually have the legal right to in training on that. And then there's kind of just hoping that the courts will side with them in the future.

And maybe they will because of its economic importance. But yeah, right now there's definitely in the gray or basically violating the law. But things may go in their favor. How much progress have we made on stopping the, you know, hallucinations? I think that they're just getting more and more accurate the AI system so that they're having more knowledge. So I think the rate of hallucination seems to be decreasing, but there hasn't been large, there hasn't been a large step change in that.

So there's still a lot of reliability issues with AI systems. They get capabilities and they get the capability to do various things that we didn't intend for them to do. They hallucinate. It's easy to have them violate the instructions that they're given and tell you how to make bombs and do things like that. So the state of AI systems and their security and safety is pretty lackluster. But most of the investment going into this is not for addressing those problems. Most of the investment is just training the bigger model.

Because the name of the game is by a 10x larger supercomputer every two years. years. So they need to compete ruthlessly to be able to afford that. Yeah, yeah, yeah. So you're going from, you go from 10,000 GPUs to 100,000. So for instance, XAI, that Elon Musk's AJA company, they just built the world's largest supercomputer. That cost more to make than CERN, the large Hadron Collider. Holy cow. Holy cow. So and it should probably grow another, they'll probably spend, I actually shouldn't comment on that. But, well, no, Elon is, you has signaled publicly an interest in spending way more than that next year through Twitter.

So or through X, I suppose. Yeah, so the budgets keep increasing exponential. Oh, my gosh. So tell me how far away are we from, you know, a China-like system but run by AI? How long before, how long do we have before a government can just say, lock it down? I mean, you were talking about, you know, enforcement of the law, and, you know, we assume that not all laws are going to be enforced every time. Were you implying that AI will be able to catch and enforce every single time? I think they'll be a lot better than it at humans because they are sleepless. They can process way more data. It takes us a long time to read a hundred page document. It takes them, you know, less than a second. So if there's a lot of information they can process and they can

spot it with in the future, higher reliability than people. So I think they could really beef up a lot of enforcement regimes to unexpected levels. I think the, it seems pretty technologically feasible, as I was mentioning before, to do a lot of this stuff now. but it would require more expertise, the technology and it would be easier to use. So it might take a while, but yeah, we do have a lot of the keys to a much scarier regime

already available. It's more of a question of implementation. Right. When you look at the future, how do you prepare? How does the average person, what do you study? What do you do? Because we're in this place where everybody's saying, well, there won't be any jobs.

So what do you study? What is like the last to be eaten? I don't know. I think physical labor might take a while longer. Digital labor is seeming a lot easier for these AI systems. robotics, so robotics might take longer, so maybe after this, maybe I'll go do carpentry or something or construction. Yeah.

Right. But even then, robotics is moving along fairly quickly now too. Earlier, just a few years ago, you couldn't get humanoid robots to walk across a variety of environments. Now it's a lot of them can do that. Right. So I don't think that there's a very robust occupation, Alibald. there it's such a general technology and maybe there's some that specifically involve a human

touch where like if it's specifically a business where it's human therapists and there are no AIs, maybe some people want that novelty or something. Right. But a lot of them, a lot of people like for medical diagnoses, they might like it being a human but they also want you know a lot of efficiency and if they can just ask an AI system on their computer to diagnose them. It's just a lot quicker and cheaper. So it'll be nice to have,

but maybe there'll be a few companies that just really try and claim that this is providing a lot of value and it's a luxury good. Yeah, I've said for years that there's going to come a time to where your doctor will come in and say, you have cancer and I think, and the person will just say, what did the AI say?

What's my diagnosis from that? because they'll just have all this massive information and the latest breakthroughs and everything else. How long before we're there where there's not that it's the expert in very important things that you, the average person, would have access to? I think that this partly already is happening. It's just that they're not overt about it. So, for instance, in law, there have been many instances where people

can find out that the briefs that the attorneys wrote for them for their clients is actually just written by an AI. So we don't necessarily catch it. For medical diagnoses, maybe they'll go off in a different room and just sort of ask the AI system and then come back with a diagnosis. This has also happened even in just creating data for AI systems. We used to have human annotators constantly work and label a lot of data. But then they just started using AIs to label the data.

And it took AI companies a few months to recognize, oh, we don't need to be hired anymore. So I think in society, we may also just have attorneys, for instance, may just screen the contract with an AI. And it'll save them a lot more time than reading the whole document. And they won't necessarily tell you about it. So I think this is a way in which AI will propagate throughout the economy, even if people aren't necessarily wanting it, even if there are rules against it. If it does provide, if everybody's using steroids, then they will need to end up using steroids or AI assistance. So, I mean, how old are you? You're young?

I'm 28. You know, a lot of 20-somethings are very pessimistic on the future. You have a reason to be pessimistic because you know what the potential is for this in a relatively short period of time. as far as man's, you know, life goes. What, are you an optimistic guy? What, what, what do you, how do you look at the world and not say, we're doomed? What, what gives you one thing is getting, um, I think actually the public, the, I think the public gets it. I think a lot of more elite decision makers are, well, we have these financial interests to keep making this go on.

And well, we need to wait for some analysis from, and this will take three years before we can talk about any sort of solutions. I think like looking at Congress on this, and there are people who are trying for it, though, but there hasn't been really any substantial efforts there. and it seems pretty unlikely for anything to happen. But the public, I think, generally gets at that. This is a likely threat to my livelihood. And having some bought and paid for scientists say, oh, no, no, no, no, it's hundreds of years away before it'll be able to do anything. They're not buying it. So I think if people make it clear to their representatives that something needs to be done and that this is a priority,

then I think we'll be in a much better situation. So that's been, I think, the biggest surprise. A few years ago, you know, this was a very low salience issue. Nobody talked about it. But it's emerged to the fore again. And I expect that this will just keep ratcheting up. There'll probably be another big AI upgrade maybe in the next six months, late this year, early next year.

and that'll make the public go what's going on and start having some demands of something to be done about AI. How's that going to manifest itself in the next six months? So I make this prediction largely just based on the fact that it took them a long time to build their 10x larger supercomputer to train these AI systems. And now they're basically built. And so now they're training them. And they'll finish training around the end of this year or early next year.

be released then. So the exact skills of them are unclear. Each time, each 10x in the amount of power and data that we throw into these systems, we can't really anticipate their capabilities, because AI systems are not really designed, like old traditional computer programs. They're more grown. We just let them stew for some months and then we see what comes out. Wow. And it's like magic. Kind of. We have. We have like extremely huge sources of energy, or we have substantial sources of energy just flowing directly into them for months. Right. Right. To create them. It's alive. Yeah, yeah. So I think they should probably get a lot more expert level reasoning, whereas right now they're a bit shakier, and this could potentially improve their reliability for doing a lot of these agent tasks. Right now, they are close. closer to tools than they are agents.

But what's a difference between an agent and a tool? Yeah. So a tool, it's where it's like a tool being like a hammer. Meanwhile, an agent would be like an executive assistant. You, a secretary, you say, go do this for me. Go book this for me. Arrange these sorts of plans. Make me a PowerPoint.

Write up this document and submit it, email it, and then handle the back and forth in the email. I think those capabilities could potentially turn on with this next generation of AI systems. We're already seeing signs of it, but I think there could be a substantial jump when we have these 10x larger models. Wow. I mean, think of it this way. Just in terms of brain size, imagine like a 10x larger brain should have to be a lot more capable. At some point, you know, it's going to snap the neck that larger brain. brain so I hope that doesn't happen soon. Dan, you're fascinating. Thank you. I, 15 years ago,

I was looking for the people who had some ethics that, you know, were saying, wait, let's slow down. Let's, we should ask these questions first. And I didn't find, you know, a lot of philosophy, you know, behind the, behind the progress. seekers. And it really frightened me because at some point, they're going to say you can live forever, but it's just a downloaded you. And if we haven't decided what life is, you know,

we can easily be, you know, taught that, no, that's grandma. I mean, you know, and then what value does the actual human have, the body,

if it's just downloadable. So I appreciate your look at safety and what you're trying to do. Thank you. And thank you for bringing this topic to your audience because it's important and it still isn't discussed about. Oh, yeah. Yeah. Thank you. We'd love to have you back.

Thank you. Yeah. Have a good day. Bye. Thanks. Thank you. Bye-bye. Just a reminder,

I'd love you to rate and subscribe to the podcast and pass this on to a friend so it can be discovered by other people. When you were little, you were trying to negotiate, always in trying to negotiate, to exchange these cards of hockey, the bonoom,

these bracelets, even of the collation. You know that each thing has a value, well, before to have to be counted. And, and the things have not really changed. Negotiate-titre T-D you can't renew with your instinct of negotiation. With, without operation gratuit,

no amount of minimum and no fretsmanual. You're made for negotiate for and the appellee negotiate-tit-titre t-ed-ed-d and fete for you aid.

Telecharge it right now.