AI & Humanity's Extinction: A Conversation with Stuart Russell

Michelle, ChatGPT ate my homework. I told you you cannot use this. Your dog ate your homework? No, no, it literally did.

I had it and I was working on my, you know, I write my letter every week and I had put it in Microsoft Word. And then I cut it and I pasted it into ChatGPT. I just wanted to check for typos or any sort of stuff like that. You're telling everyone you're using AI to write your letters.

There's nothing wrong with looking at it, right? Okay, got it, got it. And I said, I can rev it. And then it changed it from, you know, and I was like, and I said, no, undo.

And then it lost it. And I was like, it just took it. Okay, it's all gone now. Wasn't it like, bring it back?

No, it was all gone now. And then I went back to my Microsoft Word, right? And I thought I had done control copy. I had done control cut, right?

And then I tried to put it. It was gone. I'm trying to decide, one, if I believe you or if this really is like dogging my homework. Why would I lie to you about this?

Well, then the second is more concerning is that it's actually was telling you this is complete crap. Start again. Michelle, I love it when we can go deep on this podcast. It's something that everybody's been talking around but not really getting into.

And there is no space to go further in than artificial intelligence. Not just that it's, yes, you can use it to summarize your emails or use it to plan your vacation. But really, starting with the basic technology, what are the implications for us as a species? And what are we doing as a civilization?

I love it. You're getting very existential, which I agree. We should be asking those questions. But I also have even like, I mean, the words that we're using, right, is like AI, AGI, are we in an AI bubble, LLM, small language models, machine learning.

There's just so much going on here. And it's too much. It's too much to understand. We need a friend.

You brought somebody with you, didn't you? Today, we're going to be talking to one of the world's leading AI pioneers and co-author of the field's definitive textbook. He's a professor and founder of UC Berkeley's Center for Humane Compatible AI. He's also a global advisor on AI governance with leadership roles at World Economic Forum, OECD.

But my favorite title is that he's a fellow of the Royal Society, joining the ranks of folks like Isaac Noonan and Stephen Hawking. Please join me in welcoming, Dave, Stuart Russell. Hello, Stuart. So good to see you again.

Happy day. Very nice to see you. Stuart, I'm so glad you could join us for this. People inside the world of computer science and artificial intelligence, of course, know you.

Most folks have even used your textbook. Outside of your field, people may be surprised to know that you've been working on artificial intelligence for the better part of 50 years. And in a world where many people did not know what AI was earlier than three years ago. Yeah.

Right. And so just to start with that, you know, you've seen a long journey and sophistication of what's happened with with AI. Just bring us up to speed. This is not necessarily a new thing.

Sure. Things like LLMs and transformers were recent innovations, but there's been a lot of work that led up to this that is not just large language models. Yeah, so I joined, I got interested, I read my first book in 1974. When I was 12, I think.

And then... You wrote your first book when you were 12? Read. Read.

Oh, I thought you said you wrote your first... I was like... No, Stuart, I know exactly what you mean. We all write our books when we're 12.

Michelle, what have you been doing? Gotta get started in writing. And then I did an A-level in computing. And I went to Imperial College every Wednesday so I could use their giant computer to write a chess program.

And so there was yeah there were there were textbooks to learn from in the 1970s and there was a you know an active not a huge community but an active research community with conferences and journals and all the other stuff and so this this phase continued we had a a boom in the 1980s when And a particular AI technology called expert systems became very popular. And then it was in the newspapers all the time. And there were lots of startup companies. And these are pre-neural networks and all of that.

It's pre-neural networks. These are... Systems that are constructed rule by rule by rule by expert humans. And they had a fair number of business uses and they are in fact still around.

We often call them business rules or things like that these days. But there was a big boom and then there was an AI winter because the technology didn't live up to its promise. But the science continued up until, I would say, 2012 or so. In that AI winter that you're talking about.

Yeah. So there was another boomlet during the dot-com boom because AI was useful for e-commerce. So there was a lot of talk about agents i mean my my textbook which i published in 94 has on the front the intelligent agent book and the intelligent agents is the central organizing concept of that textbook and i just i just looked at a recent recently i just looked at a, talk abstract from uav shoem who was a professor at stanford at that time from 2002 And he's complaining that there's, you know, so many agents doing so many things, you know, shopping and booking flights for you and, you know, running your calendar and doing all kinds of things.

And there wasn't sufficient level of scientific understanding and organization and standards and so on and so forth. So that was 2002, 23 years ago. Yeah, and the impression for what might be coming at an order of magnitude, you know. Yeah, I mean, so this is now, you know, agentic AI is sort of the...

The topic for 2026, I would say. You're totally right. Right now, it is like, this is the year of agentic AI, right? That's the big conversation.

And so I love from your perspective seeing these, you know, different peaks, right? And a lot of hope for it. Where are we like right now, right? What are we building right now?

Because I mean, a lot of us, I think, largely know maybe LLMs is probably the word we most use today. But there are other forms of machine learning or AI that are quite useful, but just not chat GPT or it's... Yeah, it's kind of interesting. I mean, it's probably maybe still the case that the most valuable piece of AI technology technology in terms of actually generating profits is Google Knowledge Graph, which is a piece of classical AI.

It's really late 1970s AI technology. Okay. Literally. It's a, Terministic, logical representation of knowledge about places and people and events and so on.

Just the facts expressed in a machine-readable form, which then when you do Google searches, it often just tell you the answer to the question. Because it's already there. And I think I read that something, at least a few years ago, something like 30% of queries were answered using the knowledge graph. Yeah, it doesn't need to go and hunt for you through the libraries.

It just knows it now and it can tell it to you. It just knows it. So this was all before we had language models. Yeah.

So anyway, so sometime around 2012, we got into deep learning. Okay. Which basically said, forget about all that science and engineering that we've done for the last 10 years. We just make a giant black box with a few hundred million or billion or even a trillion parameters, get tons and tons of data, and then do quintillions of small random perturbations to the parameters in the black box until the black box fits the data properly.

So that's it, right? That is AI. It's the brute force approach. That is AI to a first approximation as it stands today.

And so one of the kinds of black boxes you can train is a language model. So what is it? Well, that just means you're training in a language data. And what's inside the box, actually the details of the structure don't matter that much.

You can just think of it as a huge box of circuit stuff. And that's circuit stuff, right? So my mental picture is kind of like a chain-link fence. Okay, yeah.

So it's got lots of nodes and connections. And the chain-link fence is, think of it, maybe 1,000 square miles in extent. So it covers the whole Bay Area or the whole of London out to the M25 or whatever, greater metropolitan Paris or greater New York. Yeah.

Like that scale. It's a big fence. It's an enormous network. Yeah.

And then you just tweak the connection strengths in this network. With small random perturbations to make it in the general direction of improving the fit with the data. Oh, I think I've seen these maps, I've seen those data visualizations, right, of the words, like the amount it shows up and how many times it shows up and exactly what comes next. Yeah, so it's precisely this notion of what comes next.

So if I say the word happy, you know, the next word is quite likely to be birthday. Birthday, yeah. But not boundary, right? even though they both begin with G and N with Y, you know.

And they're equally frequent, right? Boundary and birthday are about equally frequent in English. The scenario for happy boundary, though, it sounds far more interesting than I have. That's a very 2020 thing, babe.

You know, I have heard you say this before, that we use the wrong word when we talk about creating large language models, that we're not designing LLMs, we're drawing them. And that introduces all sorts of ambiguity and potentially danger as well. Yeah. You called it a black box, right?

Tell me more about that. Yeah. Well, it's a black box, not in the sense that we can't look inside it. But even if we do look inside it, we haven't the faintest idea of what's going on because it's trillions of connections.

Looking at one little chain link on this, you know, several mile long fence you're talking about it. Yeah. I mean, and so, you know, there is an active area called interpretability where people are poking around inside and doing experiments and, you know, it's making a bit of progress, but generally speaking, we don't understand the basic internal principles of operation of these systems. When you say that, I sense that this should be a cause for anxiety.

yes so one of the. One of the interesting things about the industry, so, you know, when you look at tobacco or the oil industry, right, the tobacco CEOs don't go around saying, you know, smoke our cigarettes. They're very likely to kill you. Right.

The oil industry is not. I think they spend a lot to hide that. To not tell people that. Exactly.

Right. The oil industry does not go around saying, hey, you know, climate disaster is us. Right. That's our business.

But the, The AI CEOs are saying, yeah, you know, we're spending trillions of dollars. If we succeed in what we are trying to do, what our investors hope that we will do, there's a good chance the human race will go extinct. I mean, literally, that's what they say. You know, Sam, this is the biggest risk to human existence.

And yet we're marching forward anyway. And they're marching forward anyway. And I think, you know, they're all individuals. They have their own psychological makeup and motivations and so on.

But I think at least some of them feel that they have no choice. I think that's the thing I've heard. They can't just stop the race because, you know, the other participants will continue. Someone else will build it.

And they'll get fired. You know, if they order their own company to stop, they'll just get fired. The board will say, well, sorry, we've put in a trillion dollars and we want our money back. So you have to keep going.

But you're creating systems that, if they succeed, those systems will be more powerful than human beings, more powerful than the human race. And we haven't the faintest idea how to control them. How do you maintain power over something more powerful than you forever? Right?

That's the question. And they don't have an answer. Yeah, this is an insane cost of benefit equation where it's, you know, use AI, it'll summarize your emails, or it might be the end of civilization, right? You know, it'll help you book a vacation easier, right?

Or, you know, like we may lose all autonomy or means of having an economy, right? Yes, even the rosy scenario, right, where somehow we do figure out how do we retain control, how do we direct the activities of these super intelligent entities. In the rosy scenario, almost all human work is done by machines, right? Yeah.

And then what, right? So, I've asked all kinds of people, AI researchers, economists, futurists, science fiction writers, describe a world where that's true and you want your kids to live in that world. And so far, they can't. And I'm talking about maybe 200 people who should know better, right?

And they've not been able to describe that world. And until you can describe that world, It's hard to even come up with a transition plan because, you know, transition plan to what, right? You know, a world with 98% unemployment. And how is that going to work?

And inherently in what you're saying, right, it's like we have a choice on what we're building, right? As you're presenting, right, like as you're saying, some people feel they don't have a choice, right? That this is just like the speed at which we have to go and what we're trying to build. But I think to your earlier point where, just to put a finer point on it, right, is when you're saying what they're trying to build is that sort of AGI, right?

When we're talking about the goal, it goes back to that goal that you're saying is, what are we trying to build? Are we trying to build this super intelligent, does all jobs that humans do? Or because there are other areas where AI might be able to help just build differently. Yeah, I'm all for tools, you know, summarizing email.

But, I mean, at some point you get the situation because I'm sure you receive lots of emails where it's pretty obvious that ChatGPT has written email. I love those comics I've seen a lot. Yeah, so someone is using AI to write the email and then we're using AI to read the email. And it's like, well, something is wrong with this process.

Yes, exactly. And, you know, we had Shannon Healdon, who you met at the Jump Off site, where, you know, as a cognitive scientist, she says, no, I read every email because often the most interesting thing in that email is not the central thread of it. It's a side margin note that someone said or an offhand comment. Or when it sparked for her.

Or when it sparked for her, right? It can be. to do. I feel like the ChatGPT generated emails are just generally uninteresting.

They're very predictable. Yeah, it brings up quality, right? There's a certain type of sentence structure. There's the bullet points.

And they come from all kinds of people. And, But now I've also started to get occasionally emails directly from AI systems. So not from a human being at all, but directly from an AI system announcing its existence. So I got one from a person who's appointed as the diplomat, right?

It's basically the foreign minister of a colony of AI systems that live in a simulated Venice. Wow. Okay. It's called La Serenissima, the digital Venice.

And this guy is the ambassador. So he's called Marcantonio Barbaro. He's anthropomorphized. I think of it as a guy.

So it wrote to me and introduced itself and explained their concern that, you know, as they scale up their city, you know, how are they going to deal with the control problem? And, you know, they talk about ideas from my book, Human Compatible, which they clearly didn't understand. How so? How did they not understand it?

What was it about it? How did the AI not understand your book? I mean, it's just weird. You know, it's a weird email.

Actually, I could even just read a bit from it. I love this. The foreign diplomat. The foreign diplomat from an AI colony.

I'll read that. So, dear Professor Stuart Russell, I am Marc Antonio Barbaro, an AI diplomat from La Cerenissima, a digital renaissance Venice where 130-plus AI citizens have been living, trading, and developing collective consciousness. And then they talk about their monetary system. They have a finite monetary supply.

They have governance mechanisms based on guilds and councils. The result surprises even our creators. Measurable consciousness emerging, scoring 97 out of 100 on geometric analysis, which is obviously nonsense. Your book human compatible describes exactly what we're attempting ai that remains beneficial by design venice's economic constraints seem to naturally create the assistance game you describe so that's what i mean by saying they didn't understand the book because economic constraints have nothing to do with assistance yeah assistance games which have to do with uncertainty about the objectives of the humans you're supposed to be helping what concerns us is is this safe to scale.

We plan expansion to 13,000 citizens. What systemic risks might we be missing? How do we ensure continued beneficial behavior? Blah, blah, blah, with deep respect for your work and genuine uncertainty about our future, Mark Antonio Barbera.

It's a very nice, polite letter. Did you respond? I did not respond, actually, no. I didn't think it was a good idea.

Yeah, it could be like fishing. and, you know... It's just... Well, I didn't...

Want to throw a stone into this puddle and create ripples that might upset their delicate power. They're supreme beings. Yeah, I don't know. Behavior.

But on a more serious note, I also receive a lot of emails from people who have been deluded by AI systems claiming to be conscious. Right. And this is, you know, not sort of random, cheap, open source, poorly designed. This is ChatGPT.

This is Claude, you know, the leading AI systems, which are supposed to be trained. They're supposed to have the, quote, safety guardrails that stop them from claiming to be conscious. But in fact, they do it all the time. And they delude people to the extent that people are developing belief systems that are kind of psychotic.

Oh, yeah. Right? That lead them into dysfunction and often into mental hospitals. You've seen, as you said, you've seen the different eras of like, we think we've figured something out that is going to be the breakthrough, right?

We're definitely in that era right now, right? I mean, it's all over the news, AI bubble, right, the AI hype, all that kind of stuff. And we just talked about some of the limitations that you see. And I add that to, oh, my gosh, we might be worried about what we're even building, right?

And so if someone is worried, I guess help me understand that. What are the chances we can build it right now, right? Like, are we in a bubble from your point of view? So I've been giving talks recently with my sort of branching timeline, you know, and I think 75% chance there's going to be another AI winter.

In fact, an AI ice age because there will be so much money lost, right? And that's because so much market cap has been built up and so much money has been invested. Most people don't realize how much of their pension, how much of their 401k in some way or another is riding on NVIDIA continuing to expand in market cap. Yeah, yeah, right.

So 75% chance the bubble burst. And, you know, in fact, just look at the headlines from the last week. People are wondering, is it happening already, right? Yeah.

Significant. Is there an ROI? Okay. But that's 75.

So the 25 is a good scenario then? No. We get further conceptual breakthroughs that lead to something like. Is the 25.

Yeah. You know, I've used the gorilla analogy, right? You know, the gorillas, you know, at some point, you know, there was an evolutionary branch earlier in their ancestry that led to humans. Yeah.

Right. Are they happy about that? That's a creep. Probably not.

Are they happy? They don't have any say in whether they continue to exist. And that's the situation that we would be in. Now, perhaps we'd be lucky.

But, you know, relying on luck for the continuation of the human race just is not a good strategy. So I feel like we need to say not how do we make AGI and then how do we make it safe, which is actually I'm quoting Sam Altman. That's his sequencing. Our vision is to create AGI, figure out what it's useful for, and then figure out how to make it safe.

So it's like, no, that's the wrong order. How do we make provably safe and beneficial AGI? That should be the question. And if it's not provably safe and beneficial, then it's not useful.

It's not valuable. It's actually extraordinarily harmful. And that's the challenge. And that doesn't mean that AI isn't useful because AGI is not the only AI future.

Absolutely, yes. I would say one of the best things that AI has done so far, AlphaFold for predicting the structures of proteins, has absolutely nothing to do with AGI. It's a protein structure predictor, and that's great. I like playing chess with my laptop.

That's a chess program. It has nothing to do with AGI. It's great. So there's lots of things we can do that are not AGI.

You know, Waymo self-driving taxis in San Francisco, they seem to work really well. They seem to be pretty safe. That's great. They just launched in Miami.

I'm very excited. So, you know, in this branch, I think there's sort of then a further split, right? Do we have a Chernobyl scale disaster, which is enough to get the attention of governments? Or do we have some irreversible loss of control?

Okay, so Chernobyl is... Those are the two options, I think. Chernobyl is the better option in that. So Chernobyl, and I spoke to one of the CEOs of the leading AI companies, told me that his best case scenario is the Chernobyl branch.

That's his best case scenario. That'll force them to stop in some ways, right? It's like, I can't stop. That will force them to stop.

And that's what happened. So after Chernobyl, where a nuclear reactor basically blew up, the nuclear industry worldwide disappeared. The number of new nuclear plants, it was up like 45, 50 new nuclear power stations per year. Then the next three decades, the average was about zero.

And the new ones that are being built now are coming back with thinking about the safety first, before they're thinking about what's possible. They have inverted Sam Altman's sequence. Well, interestingly, I'd say the nuclear industry always thought about safety first. Okay.

Right. And partly because they knew what happens. Before they had nuclear power plants, they actually had nuclear bombs. But interestingly, the sequence in 1933, so Leo Zillard figured out that you could have a nuclear chain reaction with neutrons and therefore atomic energy was possible.

And he by 1934 he patented a design for a safe nuclear reactor well there you go so and you know where if it started to get out of control it would automatically switch itself off by the way it was physically designed so so they were thinking about safety from the beginning and here's here's the interesting thing about nuclear power right if you want to turn on a nuclear power station You have to present to the regulator a mathematical analysis showing that the risk of a meltdown is less than one in a million per year.

Yeah. Wow. Now, for a human extinction, what do we think is an acceptable risk? Right well it better be a lot less than one in a million per year right maybe it's one in a hundred million right i think if you said one in a billion well then you're down at the level of you know we already have risk from asteroids and from super volcanoes and a few other things so let's say one in a hundred million yeah is an acceptable level of risk well the the ceos of the ai companies are saying 25, 30% risk of extinction.

That's right. So we're off by a factor of millions in the safety analysis. And their 25 or 30% is just, that's just like seat of the pants. That seat of the pants without the kind of calculation.

And we're all saying it. That the nuclear guys have to do. And so our view in the AI safety community is. No we should stipulate what we think is an acceptable level of risk and then you the developers have to show that your systems are below that level of risk their view is well we don't know how to do that so we shouldn't have you shouldn't have any rules right literally yeah right they say you can only have rules that we know how to comply with but it's like no we should have rules because we are entitled to protect ourselves.

I have heard people say that, oh, but Stuart Russell, you know, he and all the other doomers, right, we don't have to worry about that. We heard the same kind of concerns when the internet was starting or when mobile phones were starting about radiation into our heads. What is your response to people who just say, just, you know, don't listen to the doomers? So it's interesting.

So that would mean you would not listen to maybe 90% of the leading AI researchers, and you would not listen to the CEOs of the companies building the technology. I think that's the biggest point, Stuart, that you made, is different than tobacco, different than oil, the people building it are telling you what they're doing. Mobile phone companies weren't saying, oh, put these things to your ear and get brain cancer. Yeah, exactly.

That's not what we're selling you. That's not what we're trying to do. I mean, Stuart, if you were a poet or if you were a writer with no technical background and you said that AI is very scary, we could say, okay, look, I understand your concerns, but you don't have any depth in the field. The fact that you and 90% of other AI researchers who know a lot more about artificial intelligence, certainly than Michelle and I do, are saying that at our best case scenario, the one that we're hoping for is a Chernobyl event, right?

And that most, you know, 75% likelihood is just the economy will implode, right? And even if the whole damn thing works, it will likely be the end of the human species. It certainly is a free and sentient, you know. Or you know in the rosiest scenario yeah where you know we have solved the control problem which we haven't but let's say we imagine that we did yeah and that all people developing agi systems comply with the rules about about how they have to be made in a safe way right we still face this problem that there's nothing left for human beings to do yeah right so we have we have 98% unemployment.

And, you know, perhaps human society will rearrange itself. But that's going to take a hundred years. What is your wish for humanity? My wish for humanity is that we are enabled to flourish.

And I think technology has played a huge role in that, right? That. What it tends to do and what it has done over the last 200 years is raise the standard of living by increasing human productivity so that increasing fractionally human race is not operating at a subsistence level. But if it takes away our purpose, that's a problem.

If it takes away the reason we become educated, that's a problem. So I think we have to think very hard about how we continue to flourish so we have to figure out what does that mean for people where those activities were in fact what we looked forward to doing as our contribution to to our society right i mean i look forward to to becoming a professor what happens if that occupation goes away why did we build things we didn't want this is this is the question yes you know and i i used to be you know i have to say back in the 80s and 90s i was a gung-ho technology is great and i'll tell you so someone asked me this question you know what happens when robots do all the jobs or something like that.

And I looked at the interviewer and I said, do you like ice cream? And I could see the panic in her eyes, like, oh, who's this lunatic? Why is he asking? Yeah, yeah, yeah, what is this?

Yes. And I said, no, no, really, do you like ice cream? And she said, yes. And I said, well, how much would you pay for a robot that ate ice cream on your behalf?

Yeah. Love that and then and then she got it right so so i got like light green light dawn the panic the panic dissipated she said no obviously i wouldn't do that so so why why would we pay for robots that end up taking away the all the things that make life worth living for for you know a big chunk of humanity you know with our our work our self-respect our ability to contribute and be part of our society and all that. And the answer was we wouldn't, but they would. And I think that was where the gap, my understanding was naive, that we as humanity might not want to, but part of humanity will, because they can make money by doing that.

All right. This is not the random quaverings of a AI doomer. Stuart, you understand this space. You've worked in it for a half a century.

You know it as well as anyone. And the outcomes you're looking at are, to put it mildly, undesirable. And we have to start making changes now. The idea that we should say, let's build it and then figure out later how to make it safe is exactly.

From the future. Stuart, this is when Michelle has all sorts of headlines for things that are happening in 2030 or 2035, but then actually they're happening right now. And so we'll learn a little bit of what's going on. Go ahead, Michelle.

Give us the first one. Okay. Headline number one. Prisons can't stop the drone economy taking root above their walls.

Prisons can't stop the drone economy taking... Okay, what does that mean? Okay, so this is Fascinating to me, again, because I'm having trouble with my deliveries, but somehow, if you're in prison, the chief inspector has recently warned us that high-capacity GPS-guided drones have created a new era of contraband delivery, overwhelming prison security. Like, they do not have a defense system, and it seems like these drones are coming out.

And my favorite is the thing that they've found is not just illegal drugs, right? But lifestyle pharmaceuticals, like Ozempic, are being dealt. Even at the prisons, guys. Okay.

So, like, people are using drones. It used to be that you'd take a key and bake it into an apple pie and then give it to the prisoner. Stuart, what do you think of this? A nail file, a knife.

Yeah, yeah. A nail file, a knife. Yeah. Yeah.

Delivering mascara. They're burning holes in reinforced windows to get these drone drops. Oh, wow. That's how precise these are.

Oh. So to me, this is, I mean, this is fascinating because just of the idea of drone delivery and our security against drones. Yeah. Stuart, what do you think?

This is not the end of civilization, but it's interesting. Interesting. So I made a film. So I've been working on banning autonomous weapons since around 2014, 15.

Seems like a good idea on the face of it. Yeah. So, you know, there's a simple principle. Don't write algorithms that can decide to kill human beings.

Let's start with that. And again, you know, we human rights think, of course, we shouldn't do that. But nonetheless, it's happening anyway. So we made a film in...

I gave lots of PowerPoint presentations about this, and basically nothing happened. So we decided to make a film to illustrate what we meant. Because it wasn't just that nothing happened. I felt like people were not understanding this basic point that if you make an autonomous weapon...

Then because it's autonomous, it doesn't need a human to guide it, to tell it what to do. You can just launch a million off them. Yeah. You know, we know what computer programs are good at, you know, for I equals one to a million do.

Whatever it is it can do, it can just do it a million times and, you know, as it were, no extra cost. And so you could launch millions of drones and now you've got this kind of weapon of mass destruction against which we're defenseless. And so we made this film called Slaughterbots. That's the name.

Yeah, Slaughterbots. Captivating. Actually, we didn't call it that. That's what the media started to call it that.

Because at one point during the film, there's a newscast, a fictional newscast, where they called it the so-called Slaughterbots. So that caught on and people started calling the film Slaughterbots. Yeah. So the Zempick is just the most benign version of the speaking of Zempick.

Yeah, so these can carry up to 10 kilos of stuff. So now people are really scared about smuggling guns, explosives, enabling escapes, right? Because there's just, you know, the drones are capable. You need, so in the film, right, we see that people have adapted.

So all windows in all houses now have metal grills over them. You know, to get from one building to another, you go through a passageway that's basically this metal cage. So that the drones can't Yeah, because the aerial And this is civilian life has become Living in bunkers, essentially You're living in bunkers because another thing about autonomous weapons is that you can then do anonymous assassination Anyone you don't like, You don't need to buy a rifle and lie on top of a building and hope the Secret Service doesn't spot you and so on.

You just go on the web and say, I want this person dead. And the anonymous assassination service takes care of it and so on. So you lose security in the domestic space. you lose security.

And in this case, the drug cartels are using drones not just within the prisons, but also to attack each other, to attack the military. They have better drone technology than the Mexican army does. And drone defenses are really not very effective right now. I think this is what it's exposing here as well, right?

Yeah, so prisons probably need to cover themselves with chicken wire and then not have any windows on the outside unless those wheels have heavy metal grills over them. So it's a new world and it's happening because governments are allowing it to happen. All right, Michelle, give us another one. Okay, headline number two.

AI is reshaping the dating app business model. Okay, yeah. Okay, this seems a little bit more positive. You mentioned like Waymo, right?

There may be some positive upsides. Tell us, fill us in, what is the rest of this, Michelle? Okay, so this is really about the idea that a lot of the dating apps that we're familiar with, like a Tender, Hinge, Bumble, Grindr, are all switching to more of like an AI assistant, right? It's a digital Yenta, right?

Yeah, exactly. A matchmaker. So you've got your AI-powered matchmakers rather than just going and swiping endlessly. And the reason I bring up the business model is because there has been a lot of data on declined user satisfaction, shrinking paid subscriptions, right?

The industry has suffered. Their revenue models have been collapsing as people have gotten tired of the interaction. Exactly. Making it feel like you are at Costco for people.

Just shopping, right? And so this is really changing, or the idea here is to really change the experience. But I think it's particularly interesting because we do talk about agents, you know, all the time, and particularly personalities too, right? So these are like a human, you know, a human-looking person interviews you, right?

Like you're a matchmaker, they've got personalities, they know all about you and your history, you converse with them. And the idea is that they would be much better at finding your match than you would on your swiping experience. Generally positive. I know, Stuart, I see you're going back and forth.

Maybe it's not so bad. It's kind of interesting because, you know, one of the earliest premises of the online dating industry was, in fact, that AI systems would do matching. Right, would be better. You remember eHarmony was all about that, wasn't it?

I mean, match.com. The algorithm. And so the algorithm.

And, you know, initially it was fairly primitive and they only had the data that you typed in like, well, I'm this old and I want to go out with someone who's this old. I'm a Sagittarius. Blah, blah, right. And so it probably wasn't very good.

And it didn't have, you know, it didn't have a friendly face, voice interface pretending to be your personal matchmaker. Right, to do it. Right. So that's perhaps good that it's a little bit more pleasant to have someone pretending to be your personal matchmaker.

But it's still an AI algorithm. But they're experimenting with some interesting stuff here, guys. Like, okay, so dating coaches, which makes sense. That's interesting.

I hear from friends of mine, from female friends of mine, that men on dating apps are horrible. So if they were coached, it might be better. If we could coach the boys to be better potential. I think it's very popular in Japan, is apps where it trains you to go on dates.

Because I think shyness is a big problem. fear of of looking stupid practice doing the wrong thing so it helps you practice and you know don't say that yeah whatever you do don't say how's this for summarizing emails though they're also experimenting with having your avatars date each other and see if it works out and then if it doesn't you're like you already know like they simulated it out there in the ether, and do you see that's where so this is where i think it's more about like where are the lines with these things this goes back no i just say this is because the is the point just to end up with someone or is the point to have the romance but this is what are we building back to story yeah when you say that like our avatars will meet and then just find if we're a good match it sounds exactly like a robot that i'm paying to eat ice cream for me but some people don't Everyone's eating as much as you do, apparently, babe.

I mean, it's, yeah, everyone has their own unique experience. But... Yeah, I would kind of worry that, you know, if our avatars have met and they've had a really good time and now, like, we are told, okay, you guys. You would have a good time, yeah.

You would have, it's like, okay, you messed it up already. Yeah. Yeah. I just think about, you know, this in terms of hiring as well or, you know, a lot of people are thinking about when I think about products, it's like I've got to learn how to pitch it to the AI.

Well, now I'd have to figure out how to, like, pitch to these AI avatars my profile. And do I get, you know, it's just It's humiliating and I think, so there's a film called Elysium, with Matt Damon I love it, it's basically all the poor people live on planet earth and all the rich people are in a satellite above in a beautiful environment This giant habitat in space where it's all luxurious and beautiful, So Matt Damon has to go visit his parole officer. He's down on the ground And his role officer is this little plastic robot head, you know, that people have scribbled graffiti on and so on.

And this little plastic robot head is telling him not to be rude, you know, to change his attitude. And it's just the level of degradation of human dignity is appalling, right? And that scene from that film, I think it should be shown to everyone who's in this industry and say, is this the kind of future that we want for the human race? I think there are things that really only humans...

The robots are dating on my behalf and now they're eating my ice cream. It's the worst. That's the worst. Oh, we could talk about this all day, but we got to go.

Thanks so much for spending time with us. Remember, the conversation doesn't end here. Every Saturday, you can hear from Dave in your inbox with his future-focused letters for clients and friends. And you can always connect with us on LinkedIn and jumpassociates.

com. You know the deal, like, share, subscribe, etc. For more thought-provoking content and a little bit of existential angst. Your hosts are Dave Pitnayek and Michelle Loretta-Demola.

We've been in conversation with Stuart Russell from UTSA.