A Former Spy Chief on How AI Really Manipulates You

Anthony Vinci — the first CTO of the National Geospatial-Intelligence Agency and a member of the team behind Project Maven — joins Chris Brandt to talk about how AI and China are rewriting the rules of espionage.

We get into why the most powerful manipulation isn't lies but facts, how "LLM grooming" poisons the AI you rely on, the shift from being in the loop to on the loop, and why Vinci argues every citizen now needs to think like an intelligence officer. Plus: his company Vico, which forecasts geopolitical and economic events to help people make better decisions.

Anthony's book, The Fourth Intelligence Revolution, is a Financial Times best book of the year — well worth a read: https://amzn.to/4fgJRD7

Guest links:
Vico: https://vico.io
Anthony Vinci: https://anthonyvinci.com
Book (The Fourth Intelligence Revolution): https://amzn.to/4fgJRD7

Referenced in this episode:
Zero Day Clock: https://zerodayclock.com/

00:00 — The best way to manipulate you is with facts
00:32 — Welcome & who is Anthony Vinci
01:20 — Why he wrote The Fourth Intelligence Revolution
03:51 — Why espionage is now in your living room
05:31 — 2016, Facebook, and hacking an election
09:00 — AI vs. Facebook: targeting you as an individual
09:40 — LLM grooming and the Pravda network
10:34 — Disinformation vs. misinformation
13:02 — Why AI manipulation is subtle
14:57 — We've been lying to each other forever
16:20 — Training everyone to think like a spy
18:51 — The cybersecurity parallel
22:41 — RSA, agentic attacks, and AI in security
24:57 — Mythos, zero-days, and weaponized exploits
30:00 — Should powerful models be released in stages?
32:20 — When one person has more power than a nation
36:54 — The morality question for AI founders
39:11 — Surveillance vs. thought control
40:34 — Could we ever elect an AI president?
41:16 — In the loop vs. on the loop
44:22 — Why competitive pressure removes humans from the loop
45:39 — What Vico does
48:39 — Intelligence as war prevention
51:14 — Where to find Anthony & Vico

Click Here to Subscribe: 

FUTR.tv focuses on startups, innovation, culture and the business of emerging tech with weekly podcasts talking with Industry leaders and deep thinkers.

Occasionally we share links to products we use. As an Amazon Associate we earn from qualifying purchases on Amazon.

Transcript

0:00

The most powerful way to manipulate in you into something is not to tell you lies. It's actually when the AI uses more facts and figures and information, making you believe that this is very, very true. Because you're looking, you're like, it's saying all of these things that sound all very rational and they're all footnoted and sourced, and it looks like this must be a powerful argument. That is the way, the best way to manipulate someone. Welcome to Future TV. I'm Chris Brandt. My guest today is Anthony Vinci. He was the first chief technology officer at the National Geospatial Intelligence Agency, served on the team behind Project Maven, and worked at Bridgewater and Cerberus before this. He recently wrote The Fourth Intelligence Revolution, this book right here, which is awesome and I encourage everybody to go out and read. it's a Financial Times best book of the year on how AI and China are reshaping espionage. Today he's the co-founder and CEO of Vico, an AI company that forecasts political, geopolitical, and economic events for finance, business, and government. Anthony, welcome to Future TV. You know, let's let's talk about your book. You know, and I I did mention I I I read this book. I thought it was really interesting. I loved, as I mentioned before to you, the fact that you you put some great contexts for people in the front part of the book. And then we kind of dig into the like the meat of it and the, you know, sort of AI revolution side of it. Um but can you can you just talk to me a little bit about like what what was your idea in writing the book? What do you wanted to convey? And and and ultimately what is that fourth intelligence revolution? Well, I really appreciate you reading the book. And that bottom line was the purpose of the book was to write something for people who knew maybe a little bit about intelligence, but maybe nothing about intelligence and wanted to know what was happening there. Because I feel like what happens with intelligence is we have these stereotypes. We see these movies like Jason Bourne or James Bond or something like this, and we think that's what intelligence is. And because it's Technically kind of a classified world. It can feel like it's hard to get real information. Yeah. But look, the reality is, is you can actually say quite a bit about it, as I can attest, having written 90,000 words on intelligence agencies. Um, and you can learn a lot about it. And, you know, if we don't know what's going on there, what can happen can be really bad. It can be really bad for society, for politics, for all of us, because intelligence agencies are really powerful. And they do things that are right at the edge of our rights and our responsibilities as citizens. And if it goes bad, you end up with something like what Edward Snowden did and releasing documents or the church commission, where the Senate investigated the CIA for years of misdeeds. And I don't want that to happen again. Right. And so I wanted to write a book and explain to people what was going on. And more importantly than that, where was all going? Because AI and what's happening with China. Is changing everything about intelligence. And so rather than me writing the book 10 years from now, looking back and saying, here's what went wrong, I was hoping to write the book and say, here's where we're going, just so you know. And um, here's why you should pay attention, because maybe working together we can help to we can help to ensure that the right decisions are made, because we all vote for the people who run these agencies at the end of the day. the early days when I was working in, you know, like security, network security and things like that. And you know, trying to explain to people why they should care. They're like, well, I I don't have anything, you know, like nothing I do is interesting to anybody. Like, why would I why would I be concerned about any of this? Why do I have to worry about my computer? Why do I, you know, and and I think um you make a good case for why that you should be worried about these things. Um I mean can you talk about you know, like espionage you you say is in our workspaces and living rooms, right? I mean, can you speak to why people should care about this stuff? I mean, beyond beyond the r the responsibil responsible citizen piece of it, but by personal. For most of American history, intelligence has been something that's done, you know, by spies to other spies in other countries. It happened to generals and politicians and people in Washington and Moscow, and it felt really far away. Yeah. Right. The OSS and World War II were spying on the Nazis, the CIA during the Cold War spying on the KGB. You know, occasionally we might hear a story or two about a mole or something. Um, but Generally, it was something that felt far away. Yeah. Now what's changing is that foreign adversaries are have realized that the weak underbelly of American power is the American public. And that they can spy on American citizens, right? And more than just spy on them, they can try to influence them. And so this is what happened in 2016. The Russians Wanted to essentially hack an election. They wanted to use advertising on social networks and things like that to get people to um have different views on politics. I don't even believe that they were choosing one candidate over another. I think that they just wanted everybody to be angry and feel frustrated and and create disruption in this election and cast doubt on the government, whoever won, right? And cast doubt on truth in general. And truth in general, exactly. At the same time, because we're all now on the internet, because we're all on mobile phones, because now we're all starting to use AI, the Chinese in particular, their intelligence service, the MSS, has realized, hey, wait, we can we can spy on what all of these people are doing. And what they're doing is actually important for us because we're competing with America, not just militarily, we're competing with them economically and socially and ideologically. So Now what's happening is we're seeing countries spying on people, collecting all of that data, and then trying to use it against the best interests of the American people, whether it's through hacking an election or controlling a social media app like TikTok. And so that's that's what I mean by it. And that is really different from the history of intelligence, right? Intelligence agencies maybe used to do things like propaganda. But it was small and it was generally around the war and it didn't really affect people's lives. Now it affects a lot of our lives. And I wanted the American people to kind of see that this was happening and also suggest things that we could do to protect ourselves. Yeah, and and I think I mean you spent a lot of time in your book talking about TikTok, but you know, you go back to twenty sixteen. I think the real center of that storm was Facebook. And I think if an intelligence community wanted to build a tool to manipulate people, you couldn't have built a better one uh than than Facebook was at the time, right? And and and also Facebook, because it's a you know commercial endeavor. was also willing to just outright sell that data, right? Because there's a lot, you know, there's advertising, there's, you know, disinformation campaigns, there's, you know, spying. I mean, there's all these things that kind of rely on the same types of information, right? Um, and and they were really built for that. And it it gives you a scale that wasn't possible. So maybe 50 years ago, you're you're a foreign intelligence officer and you could take out an advertisement in a newspaper, and maybe it gets to some people who read that newspaper. But now with Facebook, you can take it out and millions of people will see it. And it's very targeted. So you can take it out and target a specific ad at a specific type of person, right? Yeah. But We have actually created an even more powerful tool for them to use, which is AI. Yeah. Right. So Facebook was and and by the way, the Russians used all the social media platforms. It wasn't just Facebook. Yeah. They were on Google Plus and and these other ones, but but Facebook is the one we know the most about and well it was seems like might have been the biggest. Still is so big. They're they're they're the biggest player. And they're parasitic, right? They wanted to ride on top of this technology that existed. Why build it yourself? The Chinese built their own. TikTok is like their own that they build. It's a different approach. But now with AI, there is a better tool because now AI, unlike Facebook, yes, you can target to a type of person. With AI, I can target to you as an individual. Yeah. And I can speak to the question you you're asking and I can give you a follow-up answer. Um, so now these intelligence agencies are actually targeting that as well. And there's something called LLM grooming, um, which is when you feed massive amounts of information to the LLMs to poison what's in them. Um, and then that information then kind of retrains the model in a sense. The model thinks that this is real. And then when you ask it a question, it's going to feed you now that propaganda. And so the Russians have set up something called the Pravda network, where they publish hundreds of thousands of stories every day, not for you or I to read, but for the LLMs to read. And then we ask the LM a question and now we're poisoned that way and it goes directly to me. So, you know, the the scheme of cat and mouse continues and gets more and more dangerous. Disinformation in particular is a really challenging thing. And especially when you look at um how it rolled out like back in 2016 on Facebook or, you know, like on TikTok or X, you know, or you know, Twitter back in the day. it it's I think people misunderstand what the intention of disinformation is and how it actually works. It's not necessarily misinformation. It could be accurate information in a context that causes you know, some sort of emotional reaction or or people to make decisions based on, you know, being angry about things, right? Yeah, I mean, there's that old saying, right? Like the best lie is one that has some, you know, seed of truth in it. And so it's the same with disinformation. Actually, the best disinformation or misinformation is actually true information that you have just now kind of maybe either modified or accentuated, or you've subtracted the opposing side, right? Because then if somebody does look it up, actually it turns out it is a fact. So somebody says, um, it is true that this person um from this other country did this bad thing. But then if you lose all of the context that that was one out of a thousand people, um, then uh all of a sudden you realize it uh that that that actually is f it's a fallacy, right? It's just not true. Um that is it and it makes it more dangerous. But here's here's also the funny thing about disinformation in America in particular. Is that even when we know for a fact it's false, and even when we know for a fact it came from a foreign intelligence agency that the Russian GRU, you know, made this piece of information, they put it on Facebook. You and I still have a right to to read it and to access it once it's here. Yeah. And that that makes it really hard to combat because the government cannot come censor it. Yeah. I mean, I don't know if it was Churchill who said it or whatever, but it was, you know, a lie can can go around the world before the truth even puts its boots on. And and and people I think are really attracted to the to lies because it's interesting. Or or or or just th they like the conflict. Because I mean a lot of the the strategy is just like, you're gonna let him say that to you, you know, on there too, just to spark that argument, right? I mean, the argument is the point. The social media has that kind of like accentuation and echo chamber effect on on information and and it gets people riled up in this way that's almost like addicting, like you wanna see it more and more. And that again, uh the our adversaries know this, right? This is why they're why they're using it. Now, AI, on the other hand, doesn't really have that. Um, but what it has is subtlety on its side. The lie in AI can be very subtle. You might not even spot it. And there there have been studies where actually the most powerful, call it propaganda or disinformation on AI, the most powerful way to manipulate in you into something is not to tell you lies. It's actually when the AI uses more facts and figures and information, making you believe that this is. Very, very true because you're looking, you're like, it's saying all of these things that sound all very rational and they're all footnoted and sourced, and it looks like this must be a powerful argument. That is the way, the best way to manipulate someone, um, and then leaving out the other side of the argument and those sorts of things. even though we know this the provenance of this information that that people still, you know, propagate it. I mean, and that and that is not limited to just, you know, people in the country, random citizens. I mean, it you see that happening among politicians, quoting, you know, like everything from, you know, onion articles to, you know, GRU propaganda. I mean, I think and and and you you you talk about how the humanists sort of the the weakest link in all of intelligence. And you know, like I don't know that our brains are capable of you know, because even even somebody like myself who who is familiar with disinformation campaigns and how these things work. And I've you know seen them for years and years and years. But I still feel myself I'm like, damn it, that really pisses me off. You know, and I'm like, I catch myself. I'm like, whoa, where'd that come from? You know, and but but it's hard. It's very hard to to filter through this. I mean so Вот'с топ по? We've been lying to each other forever. Yeah. Right. Even even chimpanzees will kind of lie to each other. So there is something sort of innate in any smart um animal, as it were. And, you know, there there was propaganda. If you go to, you know, Pompeii, there's like graffiti propaganda written on the side of buildings from two thousand years ago. Right. So this has been around forever. And people I think are adapted to both kind of fall into this trap. It's kinda like I think about it like fast food, you know, your body just wants sugar and fat. And so It is exactly you your mind kind of wants like, you know, exciting stories and and and things like that. Intellectually obese and diabetic at this point. There, there we go. But also we have we are actually like pretty good at spotting lies. And like think of think about, you know, if you have kids, like kids who are untrained liars, right? Like they can't, they're actually like terrible at it. They have to you have to actually learn to lie and you can spot it so quickly. They have all these tells, they smile or laugh or look right at the thing they don't want you to see, right? And we have to learn to lie because we're like actually pretty good at at spotting it. Well, there are also classes of people in our society that are sort of trained to detect disinformation. Journalists do it, right? Um, and intelligence officers do it. And you, you know, look, nothing's foolproof, but but you can, I do think it is a learned skill that you can become good at. Um and and and actually in the book, what I recommend, which I I think is a You know, a little bit of a controversial proposal, but I sort of say we should teach everyone to think like an intelligence officer, to in in essence, become a citizen spy. And that is the way to combat disinformation. Because we don't want the government to censor, remember? Like that, that's not gonna work. Just giving another $10 billion to the CIA and the NSA and saying, Go take care of disinformation, not gonna work. It's just never going to work. Nobody wants this. So Even the CIA doesn't want it. Trust me. Like nobody over there wants to to be censoring um what Americans read. Right. And so um I think what we need to do is train everybody and say, look, you we all exist in this world now. There's AI. Everything can be faked. Voice, audio, video, everything could be faked. And there are bad guys on the other side. Like, let's just face it. There are people in China and Russia and North Korea and Iran and so forth. Don't have your best interests at heart, and they're going to use this AI and they're gonna try to trick you and do bad things. They're gonna try to get you. And and even domestically, or companies, by the way, are gonna do the same thing, right? Like, and so you know, they maybe they just want to sell more cars, but like, and so we're all gonna have to learn how to see through this and deal with this ourselves and become resilient. And there are some basic basic things that you can learn, right? One thing is just recognizing. That on the other side of that, that AI, there could be bad information, that there could be a bad person, that the AI can also be hacked. Two is doing those things that intelligence officers do in that situation, which is triangulate information. Don't believe a single source. Even if you have a good source, test it occasionally. And even using the tools of intelligence officers, you can use commercial imagery. You don't believe that the invasion is happening. You can actually look at a satellite image now, buy it off the web and and and do your own research. So becoming actively involved in this, and that's what it is to, you know, think like an intelligence officer. And it can sound extreme, but remember this, you know, you were in the world of network security. If if I was telling you, you know, if I was telling somebody in the audience 35 years ago that we are going to train every man, woman, and child in America. On how to deal with cybersecurity, people would have looked at me like I was insane and be like, what is cybersecurity? And like, why should I care? But we did, in fact. And like eight year olds now know that they should change their password because just like they don't want to get their bike stolen by leaving it on the street, they don't want to, you know, have their email hacked. And they know this from a very young age because parents have the password on their iPad and That's how they keep the eight year old out of it, right? And like they understand cybersecurity and we've trained essentially everybody in the world. I think we can do the same thing for disinformation and AI derived if disinformation. I think in school we, you know, everybody called it uh critical thinking, you know, to some extent, which is uh largely what it is. But um but I I just you know, like Americans want something that comes in a pill or, you know, like is really easy to do, a button to push or a pill to take, you know, like I think learning critical thinking skills and learning to think like a spy, you know, might be too heavy a lift for the bulk of the population. And You know, and and and and ultimately it's always it's always about the numbers at at the end of the day, you know, of of people that you can get over there. I just I just worry that like evolutionarily we are not suited for the world that we're stepping into. And and there's gonna be a big disconnect there. It is critical thinking. It y you know, in the book I actually sort of draw this parallel between philosophers and and spies, right? Like 'cause they're both really cynical types of people who don't believe anything that they're told, right? Yeah. Um, but the difference I think the difference is um it's also recognizing that there's an adversary. Yeah. So it's not just critical thinking 'cause there might be an error because the world's hard to understand. It's critical thinking 'cause there there might actually be somebody Doing it to you. Trying to con you, right? And um, you know, look, we still we still teach it. uh we still teach people how to do this. I do think teachers should should learn. I look, I think you do it in an age appropriate way. And, you know, third graders can be told, like, hey, you're doing, you know, this report. What if what if there was a bad person on the other side? Like, how would that change your mind? Right? There there are and there, you know, maybe one day I'll, you know, maybe the next book I write is a children's story. I don't know. But like Um the the, you know, I I do th I I do have a little bit of optimism that we can do it. And and and especially because we've done it before, right? We've done it with cybersecurity. We actually did it with with terrorism, right? uh post-9-11. I mean, you still, you know, when you're when you're in New York and you're on the subway, you know, there's still the signs that say see something, say something. Right. And we did sort of as a nation realize look, there is this big threat. And we should do something about it. And if there's like some funky looking bag with a ticking sound coming from it, most people will still like telecop and realize like we have to protect ourselves. We also do it with crime, by the way. Yeah. We t we teach children from a very young age like how crime works, that there are criminals, stranger danger, right? And so this is just another one of those things. And by the way, we don't have to get it perfect. Right. We don't get cybersecurity perfect. Do phishing emails slip through? Yes. But even if you just get the eighty percent solution, you create resiliency. You make it that much harder um for the adversary to take advantage of you. We we are entering uh you know, a very interesting new world. And you know, like I was at RSA um earlier this year and uh I you know I was quite taken by the fact that basically, you know, everybody's you know, business had been completely upended uh by AI, you know, and a the the i insertion of AI into security has really, really drastically change things. Um, you know, like we had, I saw everything from uh, you know, deep fake detection tools to, you know, like agent interaction, you know, like uh a lot of around identity access management, API management, things like that to, you know, sort of I identify threats against agents. But now, you know, now you're seeing like these multi, multimodal attacks where, you know, they're launching zero days that have been, you know, Protected and weaponized at the same time, you know, you got an agent doing some sort of social engineering on a human, you got an agent doing social engineering on another agent, you know. I mean, it, it, it's, it's, it's a drastically different world we live in. And at that same time, we have so much data now we're collecting and retaining. It's, you know, like, how do you make sense of it all? And how do you get ahead of that? You know, like adversarial threat when there's such an imbalance, right? write about this in the book and I I think about it in terms of um intelligence right I have an AI spy that I want to use to spy on your uh government well you also have AI and so you're using your AI spy to spy in my government and so I use my AI spy to stop yours. I do AI counterintelligence. And so on and so on. So if you ever read that old comic spy versus spy, they're always trying to one up each other, right? Well, now it's AI spy versus AI spy and trying to attack and defend. And I think that the same thing is happening in cybersecurity. Like, yes, you're absolutely right. I mean, um, what we saw with Mythos, for example, the the anthropic model, where it's able to spot these zero days. It's not that we couldn't spot zero days before, it's just that it took a lot of time. And you had to have like a really good expert to be able to do it. So there weren't that many that could be discovered. Now you've got like AI is great at scaling and it's like pretty smart. So it can just run the you can run at any scale and spot way more zero days. But also at the same time, it's really good at spotting weaknesses in code. Yeah. Right. And and that's actually why it's good. Like I don't think in any way they tried to design an AI that was good at hacking. They tried to design an AI that was good at coding. And when you're good at coding, you spot broken code and and then it could spot that. But that also means you can fix it. Yeah. And so I think what's going to happen is for every downside, there's an upside and vice versa. So yes, you can have more zero days, but I can also very quickly run through my whole network, run through my whole computer system and look for these weaknesses and then patch them very, very quickly. You know, whereas I used to have to put a person on it, now I can, you know, develop, you know, a few new lines of code immediately, maybe in an automated way. I don't even have to have a person look at all. It just kind of cranks through it. Um, similar same thing is going to happen, I think, with information operations, right? And where in disinformation AI, it's already being used to develop disinformation, send that disinformation out to all sorts of different people, but We could also use the AI to help us spot disinformation. Right. AI is great at spotting patterns. Yeah. Including that other AIs made it up. Right. And and and surfacing the provenance of of in of information, yeah. You know, anybody anybody who's written a book recently will probably get the same warning I got from my book agent, which is don't use AI because they use a your the publishers use an AI to spot and see if you used AI to write your book. Because people may know this, you can't copyright something that an AI makes. So if you write a book and AI wrote it, you can't copyright it. Publishers don't want that, right? So so you know, for every Bad piece of disinformation. Maybe we can use AI to help find the disinformation. So I think actually what's gonna happen is that the game is just gonna continue. And now we're just gonna play the game at a higher level with AI going after the AI and so on. Yeah. And I think, you know, when I Mythos came out, and that was a a big, you know, there's a lot of drama around around that. And most recently, you know, the the the consumer version of the Fable Five came out and was instantly pulled from the market. I think I had it for one day. Um, you know, so but I I I think, you know, like it is good at detecting errors in software. Um, but I think the thing that was really you know, that that they I that they called sort of the step function of mythos was really that not only could it detect, you know, bugs and problems and vulnerabilities, but then it could turn around and automatically weaponize those into an attack framework. Right. Which is kind of the big, big difference and could kind of run off and and do it on its own. And you know, when you look at there's um there's a uh you know threat counter out there somewhere that you know like kind of tracks the the timeline between a C V E, which is a uh a vulnerability, you know, document that comes out that talks about a vulnerability. People who are not familiar with that. And uh the the time from a C V E being being uh published to the exploit is now ten hours. Which is crazy to think about. I mean like, you know, when you when the majority of like attacks are coming from, you know unpatched software vulnerabilities anyways. You know, like you don't even have ten hours to deal with something. That's like making a patch, distributing the patch, installing the patch. Um, that's gonna be really crazy worldwide. This is why I I thought that the solution was fairly savvy, which is let's release this early to to the people who want to patch things who we trust. So um, so you know, let's say you're you're Microsoft, you're concerned with, you know, Windows. okay, you release this AI, let me run it myself. Let me self hack um and then patch everything. And remember, it it's finite, right? So in theory, Mythos, once you run it once, or maybe you run it a few times, and it looks for those, and then you've patched all those, eventually it's not gonna find any more. The model will sort of i in theory should hit its like that that's all the at that level of sophistication, those are all the the weaknesses it can possibly find. And so I do think we're sort of entering a world now where you you can no longer just release a model to into gen pop, right? And just let people do whatever they want because there will be these dangers. Yeah. Right. You need I do think this idea that maybe we release it to the government early, let the government first of all find the weaknesses in classified systems and things like this that nobody wants. Nobody wants you know, Sandia National Labs with its nuclear secrets to be hacked. That's like net bad for all of us. So let's let them sort this out, then then from there, okay, now let's worry about utility companies and things like this. And then maybe, you know, networking companies and and and IT companies and software companies. And so if we can do that at every every time, that seems rational to me, right? We again This used to exist. It was just people, right? There was somebody who worked at the NSA, right? And this person was trained over a lifetime to hack into things. that person retires. Um, he goes out and works in private industry. In theory, he could go out and start hacking into American companies and banks and steal all the money that he wanted, right? Yeah. Like that's always been an issue. Now that didn't happen because One, we had laws that that person probably didn't want to break. and and and and two, we you know, like we he's a human being and has morality and things like this. Yeah. Right. And is probably not gonna go do that, even if he could, in theory, go do it. Right. Seems like we need kind of the same thing with AI. Yeah. Like one, we need laws and processes to ensure that it doesn't do it. And two, We probably need something built into the AI so that it doesn't just go out and start harming people. Yeah. And I think kind of both things are sort of happening, right? Where that the model companies are sort of trying to figure out how to do both of those things. And the government is trying to figure out like, how do we not crush these companies with regulation, but how do we figure out some way to like ensure they're not making bioweapons or hacking nuclear labs? Yeah, and and it's moving ab absolutely at the speed of light right now, too, which is makes it even harder. Companies do this as well. It's not just it's just not just nation state actors and things like that. But I mean, the idea that I mean look, look, I mean, some of these companies have resources bigger than, you know, many of our adversaries out there. You know, like Elon Musk personally has more resources than some of our adversaries, and he has rockets. at his disposal. Inter which, you know, I mean, they're like at the at the end of the day, intercontinental ballistic rockets, you know, for whatever purpose you want to make them for. I mean, so, you know, like not only you know these tools I think enable you know more more potential espionage between you know competing states, but you know, like espionage in our day to day lives, um, you know, like between companies, between, you know, very well entrenched wealthy people between you know, I mean there's just the the the attack surface has gotten bigger, I feel now. I mean, would you would you say that's accurate? I th I think that's accurate. I I mean, look, companies have been big for a long time, right? And people worried like in World War One about the military industrial complex and the you know, and there were mercenaries, um, mercenary companies in essence in you know, Renaissance Italy running around like take taking over comp you know, governments. And so we've we have been addressing it, but but it is interesting, you know, what you're saying in that today. It's like these single individuals can rack up so much power. And there are many of them that and and but you know again, I would come back to um this is why we have laws and morality, right? And and e look societies have realized this for a long time, um, that there could be people so powerful that they could actually harm large. portions of society and and maybe even the government itself. And so there there is law. There is law and there's law enforcement. And like, yeah, Elon Musk has you know, rockets. but at the end of the day, one cop can can stop that from h right. Like if he's gonna launch something and a judge says, no, you cannot do that, trust me, a police officer walks in there, gun drawn and says you at the control turn that thing off, they're gonna turn it off. and this is, you know, it's I hate to, you know, I I hate to think about it in such like crass terms, but at the end of the day, you know, um, you know, that's what sovereignty is about. And it's about it's about this monopolization of violence, right? That that that a government really can come in and say no. You just cannot do this. We've gone through the gone through the courts and that's it. but even more than that, there's a morality in it, which um actually in today's world means I think more than it ever has in the past. I do think these guys have a moral duty um to think about what they're making um and how it should be used and could be used. Because there are a l you know, where it gets really hard is maybe you're doing something where there's not a law against it. Yeah. Right. Right. Like there was right. The and but you could still stop it. and you know, social media, I think, harmed people for years and still continues to. Um, and it wasn't illegal, but I do think some of these CEOs could have lessened that harm by making some choices. Um, and I think the same thing is happening AI. What's interesting in the AI industry is a lot of the founders seem to be cognizant of this. Yeah. And and are trying to kind of minimize the harm um where they can. I would hope so. Some are. Some are, some are say. Yeah. Um, I know like I run an AI company today. I think about this. I mean, obviously I don't run very important or something, but like I do actually think about it. And um I know that the developers who who who work for me think about it. And I so I do think there's this morality level to this and we all should take it seriously. Yeah. I mean, I do get concerned that a lot of uh a lot of developers I see in in like the valley have declared P, you know, like P one hundred percent on the Doom chart and they're, you know, like they're starting to bug out. But but I mean, you know, like you look at you look at like that one interview like Peter Thiel did, where um, you know, it was the question was something about like should him humanity continue to exist? And he and it took him a really long time. time to answer that question. And I don't think we got a satisfactory answer out of him either. And, you know, he's one of the architects of this. You know, so like it I think I think it is, you know, like I think we have to I I mean, I guess, you know, the the um the web of threats that we have to be aware of. You know, it's like it's external, but you know, it's the the call is coming from inside the house a bit too. um on some of this stuff. And I'm not, you know, meaning to to spark, you know, AI panic, because I think AI is gonna be an incredibly useful and valuable tool. And I think it's it's it's a it has the potential to be to be one of the most impactful things to humanity, you know, in in our history. But yeah, there there are, like you say, there's concerns. I yes. Um I I think that um the concerns the people harming people part will come long before the AI harming AI people part will come, right? Like like people will use AI to do all sorts of horrible things to other people long before AI is like sentient and evil enough to start harming us. So um I I feel like we should figure that part out first. Yeah. You know, and whether that's look, social media, the threat was surveillance. I think AI, the the threat is thought control, which to me is uh in some ways even scarier than the threat of surveillance or even like the people talk about the like killer robots. Yeah. Like, okay, but at least you know they're killer robots and you can fight back. With thought control, when it's good, you don't even fight back. Yeah. Right. That's what a totalitarian states try to do. And now we're under threat from that even at home. So we should care about it a lot. So that's And it's not even just from the state though. That's the interesting thing. And and yes, exactly. And so there are these things we should worry about today. And then yes, like, look, AI can see seem big and scary, but you know, remember this. We are one president away from the American people being like, Yeah, we're electing the guy who wants to nationalize AI companies. Yeah. Right? Like, and then that happens. And it's like that's it. Right. And so there is a self control built like that's why democracy is so interesting. It's like every four years you get to you get to sort of decide and you're like, Yeah, we're we think this is dangerous and so you need to go in there and stop it and then they stop it. Um, in theory. Or it's also somebody walking up, look, uh, AI is embodied in a data center and you're you're, you know, some wire clippers away from you know, solving that problem if you wanted to, right? Like there are ways that society can control these things. So I'm not I don't think we should be freaking out quite yet. We should more be freaking out about what are people doing with this? And are there people in power um domestically or in adversarial nations that are trying to harm us today? Yeah. I w I wonder how long it'll be until we elect our first AI president and you know, just say thank you, thank you. I I I for one welcome our robot overlords and just go about our day. I mean it's interesting, right? Like I doubt that the I t I doubt that Thomas Jefferson had this in mind when he was writing it, but you know, at the end of the day you have to have been born in American uh in America. And be over the age of thirty five and AI does not meet either of those. So Born in America and soon it will be over the age of thirty five. Yeah. Born. It's a powerful verb. It's doing a lot of heavy lifting in that statement for sure. Um, okay, so I want to talk a little bit about um, you know, like one, you know, the the challenge of sort of these uh getting comfortable with non-deterministic systems in our in our world, right? And and non-deterministic systems sort of uh r running our lives and running our security and protecting us from a lot of these threats. And and you know, one of the things I th I I thought was a great comment you made in the book, you you said, you know, rather than being in the loop, we have to be on the loop. Could you explain, you know, the the difference there? Yeah. Actually what I say is people wanna be in the loop. Yeah. We by default will likely be on the loop, but that actually we will end up in the place where we're not touching the loop. Right. Um I think that's a great Yeah. I mean, in the loop, the idea is that you uh you know, it's easiest to understand it. There's this debate in the military about killer robots, right? And you you you maybe you have a drone that can shoot a missile. It can kill a person. And the question is, do you let the drone just shoot the missile? And in the loop means that at some point a person has to actively hit the button to shoot the missile to kill a person. Right. Right. On the loop means that the person is there, that they are able to monitor this. And they can push the button or they cannot push the button, but they don't have to do either. The the robot will go and can push the button itself, but a person needs to be at least around it. Right. Then there's the idea of just it just does it. Yeah. And it's just automated. And and unfortunately, um, I do believe that we will very rapidly hit a point where we say, Yeah, we just gotta let the systems go. Yeah. Because remember, the bad guy on the other side um also has the drone and they're might say to themselves, well, let's just, you know, I don't care about collateral damage and just start shooting and let this thing go. And now that drone is operating far faster than a person can. And we're there's nothing you can do. And and the proof of this is if you look in finance and you look at hedge funds where um these these hedge funds now make the, you know, quantitative hedge funds they're called, make these um essentially Autonomous bots that do that trade. But these bots are now trading so fast that a human cannot process it. And they let them go, not because they want to kill anybody, but because they want to create more profit. Right. And and if I tried to have a person stay on the loop, then I would be operating more slowly. And the other guy's going to make all the profit. So eventually I say, well, for me to make the most profit. I'm just gonna let the robot go and it's gonna do all of the trading. And maybe I just might have like a circuit breaker if it starts going like completely crazy and losing lots of money, but that's it. Yeah. and and I and so I do think that there is this like competitive pressure to operate faster and faster. The same thing's gonna happen in warfare, the same thing's gonna happen in all sorts of other parts of society with AI, unfortunately. And so I actually I think where you need to operate then at that level. is the design function. How have we designed it? How does it think? Are we sure we trust it before it even gets out there into the field and starts to operate? Yeah. I wanna switch it up and talk about uh Vico because I you know, I tried out Vico. It's it's it's very interesting. Um tell me tell me like you know, speaking about AI and all that, tell me what Vico does because it's this is a really interesting tool for making better decisions and I Yeah, we made a system that can forecast the future and it can forecast it better than people and prediction markets like Polymarket or Calci or something like that. And basically what we did is we built a system that takes in tons of information um from the internet, you know, reads, think about something that's reading all the news and every newspaper in the whole world every day, every minute, right? And then um we built an AI system. It's actually sort of a pretty sophisticated system, it's not just an LLM. Because as you sort of mentioned before, LMs are a little too non-deterministic to be pretty, you know, particularly good at this. There's a lot of mathematics and data science and actually modeling and simulation that goes into this. But we built a system um that is very good at thinking like a person would think to forecast the future, taking into account historical patterns, taking into account um causation and all of these factors of how you would actually predict what will happen. And then we unleashed it and we let it. forecast things and and it learns over time to get better and better at this. And we back test it so we know how well it performs. And it's actually doing pretty well. So I mean we forecasted the strikes on Iran. We forecasted the closure of the Strait of Hermuz. We forecasted that um the the Virginia Supreme Court would overturn redistricting. Um we forecasted um inflation in in April and May this year. Um, and we did it with high confidence and in non consensus. Like these were hard forecasts to make. Right. Like people did not think the Strait of Her Moose was gonna close. That's why oil doubled in price. They really As and I was telling you this earlier as a political science major. I I find that so bizarre that like that but I mean I think that speaks to like the the volume of information that everybody gets. It's har it sometimes it's like, you know, more information isn't necessarily better. It can be hard to weigh it, to analyze it, to do it, and to do it every time and do it without bias. And I think that in many ways is the secret of our system, is that we as people fall into these biases, either our the way we think as an individual, or because they're, you know, if you read the news, it's all argument and politicization and all of these things. And the the system doesn't care. It's just looking for truth. It's like a perfect philosopher. in that sense. And it's so it's able to outperform people. And then you can use it to make decisions, to make better. That's what I care about. Like like forecasting the future is cool, right? But what I actually care about, what I did as an intelligence officer is just try help people make better decisions. And that decision might be how to trade a stock. That might that decision might be um how to ensure a war doesn't start. That decision might be a personal decision, like should I buy a house today or should I wait six months? Right. And we're trying to do that. And that involves providing forecasts, yes, but also scenarios. When you make a decision, you want to run a scenario, not just like, you know, w what will happen in this election or what will happen in the Strait of Hermuz and whether I should buy oil or not. But what if this happens? What if that happens? That's how we actually make decisions. And so we provide a series of tools. to help people think through decisions. And we, you know, we service people in finance like hedge funds. We service, you know, I still think about the missions. So we try to help, you know, the Department of War and the intelligence community. And ultimately anybody can sign up and use it to make their own decisions. You know, maybe President Vico is gonna be the Well, I do I I do really think that presidents are gonna use AI to help them make decisions. I do. And I I actually I think it will make the world a safer place. Like I agree. What a lot of people miss about intelligence is to bring it right back around, is that and what intelligence officers do is try to keep us out of wars. Right. The military is there to fight and win a war. The intelligence officer is there to help you never get into the war. Because you see what's happening coming, you understand what's happening, and you buy space and time for a decision maker to try to avert that war. Well, I think the same thing, you know, happens with something like Vico, right? Like my goal is like this thing helps us not get into a war or this thing helps you make that better decision so you don't lose money on your house or choose a bad college for yourself or something like that, right? Like let's uh that's like my real goal. I had a professor always says that, you know, war is sort of the outcome of bad politics, right? And you know, like understanding in a historical context, you know, the the the changes that are going to be created by any decis political decision that gets made. And at the end of the day, wars have to be diplomatically resolved anyways. So like, you know, if we could get to the point where you could say, you know, in seventy years You know, if you if you overthrow this government in 70 years, we're gonna have this huge crisis here. You know, like that's that's the level I would love to get to. And I I think this is a good start you got you got going here uh with Vico. So I'm excited to see uh it grow and and and be used in uh in decision making. Thanks. Appreciate that. That is the goal and I think what we go to work every day trying to do. And so and by the way, if if anybody wants to try it out, um it's at Vico dot IO, V I C O dot Io. and just mention you heard the podcast and you can try it out. if people wanted to get in touch with you and and you know, you've got quite a quite a past too, um, how would how would they reach out to you? Yeah, can check me out at Anthony Vincy dot com um is is is my website where you can reach out to me or you know, check me out at vico dot io and hit me up there or just hit me up on LinkedIn or X or something like that. Well, Anthony, uh really appreciate you being on. I I got about three more hours worth of questions I would love to ask you. So maybe we maybe we have to do a part two at some point. But uh thanks so much for being on. I'm excited to to play more with Vico. I've been sort of gaming some things out myself on there, which has been interesting. So uh thanks so much for being on. Thanks so much for sharing Vico and good luck with everything. Thanks so much for having me on. Really appreciate it.