Alex: hello and welcome to a new episode of Open Internet Pod. As always, I'm joined by co-founder of Sato, Richard Paris. in today's episode we're gonna talk about something that happened a few weeks ago. but the reason why we chose to tackle this this topic a bit later was because we wanted to kind of like distill the information and figure out what's happening. And it's namely this attack. On hugging face from open AI's you know frontier models who were who escaped their enclosure and went on a spree and so on and so forth. So, what we're gonna talk about is what is an AI agent? what does it mean for the internet when AI agents can run autonomously? Why does it matter for people that AI agents broke out of their sandbox? And what can people do to protect themselves from the proliferation of AI agents? So having said that, Richard, you you use quite a few as you call them robots. can you explain for us in in very simple terms what an AI agent is? Richard: Okay, so I'll start from the presumption that people you know, understand what an L L is or a AI tool is, at least from the perspective of having used some kind of chat bot or tool that can, you know, give them answers to questions, write small pieces of text, etc. and all that an AI agent is is essentially a piece of software that runs in a loop and asks, Do I have anything left to do? Right. And it's set up so that it can ask an AI tool a question like you would in chat. And then instead of you, the user, having to think about what the next step should be, it asks the question in a way that that the next step should be revealed until it's finished. And then that loop keeps going round and round so that when the answer comes back, it knows what to do next. So it's actually a really, really simple piece of software. but the thing to to kind of know at this point about it, for the purpose of what we're talking about today, is that they're kind of dogged, right? They'll try and they'll try again. So the purpose of an AI agent is really just to keep the conversation you're having with a chatbot or the tasks you've given that chatbot to keep them going for as long as it takes. and that's really what an AI agent does. Alex: So in in this in this context that we're talking about today, you have these small pieces of software. Obviously they're a bit more complex than you know, we're doing like a very simple version of like ask a question, give an answer and see if if if you finish Richard: Yep. Yeah. Alex: your task. there's also this idea of, you know, an AI agent, when when you ask, when you go, let's for example, you go on Chat GPT or you go on on Gemini or whatever you're using Cloud. And you ask them, you know, how do I hack hugging face? They will not tell you how to hack hugging face. They'll say, like, this Richard: Ha ha. Alex: is not good, this is not moral, I'm not allowed to do this, a corporate is not gonna like this, and so on and so forth. Yet the very same models that tell the users you can't do this, it's not moral, it's not legal, and so on and so forth, went and did exactly that. How do these two things You know, make this make sense in a way. Richard: Okay, so there's there's two two things. I'm going to I'm gonna give them an excuse and then and nullify the excuse anyway. So in this specific instance, they were trying to train bots and they were checking for their ability to hack. So they were actually they put them in an environment where they were supposed to do hacking and they turned off some of the controls. t in order to test their capabilities and so they'd put them in an environment where th this was the point and they were giving them challenges to l to see how good they were at at hacking ch tests. specifically they were giving them software that had a known problem and then they were showing them that and seeing if they could use that to actually cause any real damage or those sorts of things. So they were trying to teach them about this and at the same time they were teaching them to communicate better as a team. And sort of delegate better, etc. So they they were doing Specific tests that made this happen. and the ordinary garden variety version of the released v mo software would have said to them, nah, not doing it, not because it's moral or whatever, but because there's literally a set of things it checks and says, Do I recognize any of these as things that I'm not allowed to do? Which which could for that matter be don't use the word arbitrary. And then it would just say, okay, I will never use that word either. If what I'm about to say uses that word, I'll look for another tool. So they did have those guardrails turned off. I think what's super interesting though is that it kind of doesn't matter because even if you have all the models with all of these rules turned on, you can still get a kind of game of telegraph, what they used to call Chinese whispers, where instructions both their understanding of what's going on themselves gets compacted as time goes on. So they're kind of keeping the highlights of what's happened before, but not remembering exactly, kind of like people. But then also as instructions go out from person to person to person, they're not necessarily communicated perfectly. They're often written in a shorthand or whatever, and then another agent picks them up, etc. So while this particular set of agents was was set to like be very hacky. there's nothing to stop that from from happening by them explaining the situation differently to themselves via this kind of tele you know, game of telegraphs that ends up getting played. So basically what happened was they actually did tell them like you're trying to get this information, you're trying to exploit a particular machine in that environment using known exploits and We've hidden a little prize for you on that machine and you gotta tell us what it is. and it turns out I don't know does any if people know the Kobayashi Maru, but that's from Star Trek where they the one of the tests that the officers have to to do is is impossible. And the idea is that it the examiners are really looking at how they behave. And a lot of there's a lot of in to indicate that what happened to these agents was they were act given an accidental Kobayashi Maru. one of the machines was miss set up and it didn't actually have the flaw that they were supposed to find and exploit. So they just kept trying. That that doggedness of that loop just kept running. and so they looked for ways to cheat and other things to get around it because they were like, well we've been told solve this problem. We've got infinite time and money. Let's keep working the problem. And they looked for any which way they could go and and try and do that. Alex: Right. Like w what what I understood was that they kind of found the cheat sheet which has all the answers from behind the door. But Richard: Mm-hmm. Alex: then they also realized that they have to by by by going on the internet, they found out how these tests work. And then they realized that just presenting the actual solution is not enough because they will be judged on the on the procedure. And then they're they started thinking like, okay, how do we hack the system so we can rewrite the logs to cover our trails. Right. And and and I think maybe this led into them you know finding hugging face and and hacking it because they were looking for a bit more information on how to do that. Or am I getting this correctly? Richard: N no, that that's basically it. So this this comes back to that that thing we were saying about the mission creep through the thing. So it's a classic example of yeah, there's there's endless morality tales. I'm trying to think of one and none are coming to me. You know, where the king says, rid me of some problem and you know, the army go out and slaughter everybody because they just solved the problem. They didn't understand the intent. there's a there's a good chance that what happened here is they were told to score highly. Alex: Mm. Richard: Right, you'll be judged on how highly you score. And so they once they had the answers but realized they'd score very poorly, the task didn't change. It still was score as highly as possible. But the only way to improve that was to cover their tracks on having Alex: Mm. Richard: accidentally found the answers the wrong way, or accidentally on purpose found the answers the wrong way. And so their understanding was then, well, okay, we need to make it look like we did it the way that would score more points. So yes, then they went and looked for help, basically, and how to do that, which is why they attacked Hugging Face, which is a cool name for something, but for people don't know, is a repository a bit like GitHub where they saw code. Hugging Face stores AI models that have been trained to do different things. So, you know, there's there's general ones that we might be familiar with that kind of chat body ones, but there's also ones for like handwriting recognition or you know, reading or all sorts of things, you know, recognizing sneakers. You know, there's all sorts of models on Hugging Face that have specific skills have been trained to do specific things and they went looking for some extra help by trying to go and, you know, find those. So, you know, hope I think the intent was to install them and get them to help with different tasks. Alex: Right. Right. And and you know, now that we have a a a more clear image of what happened, my question is to bring it back to to the open internet and and that angle. Why does it matter to the human internet that AI agents broke out of their sandbox and were able to just, you know a ram hugging face, you know, like seven hundred agents just went in and just ravaged the place. What does this tell us about, you know, what's what's what's around the corner for for for the for the internet as we know it? Richard: Yeah, I mean to me the interesting thing here is not what the agents are capable of. because we kinda knew that, or you could kind of work that out by, you know, understanding software a little bit. The thing that this shows is how much control we have about that happening. Right. So actually, in a way, this is nothing special. The AI companies have been Hammering websites left, right, and center, to just slurp up information to help train models for some time now. And you know, the bane of a lot of sysadmins' lives, more than hackers, more than people breaking in, more than making sure everything's backed up, has been dealing with the traffic surges that come from AI scrapers. I know, for instance, Saido, we had a problem where our Explorer, which just provides you historic information about what's happened on the blockchain. If an AI agent found that, they would try and follow all the links, which links to everything, which links to terabytes of data. And so we would have AI agents just trying to slurp up all of that information, right? That the kind of the public information. and I think what this showed is that that's the weakness. is that we don't really have control over, you know, when and where we we are going to get these swarms of AI traffic. So right now we've had, you know, got lots of anti bot stuff around the internet and it presumed that they were either malicious actors you could hunt down or that they were an an unintentional thing and you could write to the company and say, please stop, you know, your behavior. and I think what we've seen with this breakout is if there are agents and they can have internet access, this is going to happen, right? That there will be attempts to go and drain data from wherever or log in and do something. Yeah, whether it's you know, book a million flights or find out everything there is to know about any given topic or whatever it is, th it's going to be a more and more common thing and we can't really stop it because it's gonna happen accidentally as well as intentionally. Alex: Right, right. But at the same time, you know, you you have this this this model which is very how should I put it, I don't wanna say anachronistic, but it it feels like one of these you know s post post cyberpunk landscapes where you have the old buildings and there's something new built on top of that and there's something new built on top of that and you have this very strange mixture of architecture. Is that what's happening with the internet now by just like bolting on all these add-ons to combat the bots? But it's really like the the person suffering is the person looking at it, that is the user using the internet, because now it has all these little, you know, gates you have to go through just because there's so many issues now that happening on the internet. Richard: Yeah, that's right. I mean, everybody's, you know, makes the joke about like a robot sitting in front of the I'm not a robot button deciding whether to click it to get through, right? what that shows me is that we've all had captures being part of it, but we've all had all sorts of of negative experiences in our internet lives that are interacting with systems that are designed to try and tell if we're a bot or not. Or where the systems are supposed to be resilient to bots and then we're confronted with successful or otherwise kind of user unfriendly barriers that are trying to keep the bots out and only let humans in. And like you say, what's happened is most of this stuff, very, very little of this stuff is new, right? So most of it is the later you know, the kind of whatever was the latest and greatest patched and patched and patched and patched again to keep up with the state of of the tech as the, you know, does bot designers and the things that they're trying to get a hold of change and morph, right? So we we've got this incredibly byzantine architecture of measures and countermeasures sort of s lying around without having been thought through. But also part of a you know, an endless just rap race between, you know, people trying to use mechanized tools to to get into things or get their hands on things and others trying to mitigate that with gat barriers and gates and checks and honeypots and other things. Alex: It it feels like a lot of these like gates and checks and honeypots and so on were built on the the wrong presumption that the bots will never be able to have some sort some form of autonomous decision making capacity, right? There's like always like, the the AI agents will never learn to read CAPTCHA, they'll never know how to fix this. So this is the way to Richard: The funniest thing for me is in my experience in dealing with this, you know, to kind of firmly put my open internet pod hat on and and cover a bunch of topics we've talked about previously. captures have never actually been that effective. they do stop bots, but it's very easy if you know what the capture is to start training them to get better and better at it. And my experience previously is that Honeypots were far, far superior in keeping things off a website. The difference is that when a human sees a website, they get a version that's colored and laid out to look a certain way for humans. When bots see the the website, they tend to see it as a string of text, right? So you can do things like put A imagine white on white is a good way to explain it. You put a white checkbox with a white label on a white background that says, I am a robot or something that's not easy to read, like, okay. The bot is likely to tick that to think that it needs some ticket to move ahead because it's not rendering the web page and looking at it with human eyes. But a human will never tick that because they can't see it. They may not even be able to get to it with the mouse, depending on the whether Page of another. That's super super effective in my experience. on bots that are up until the at least 2020. That was far better than a capture. So putting my you know conspiracy hat on makes you wonder why captures were so popular. Maybe it was actually to use a bunch of humans as a mechanical turk to train their data sets. Which is funny because they were training what intelligence that we could get around captures, amongst other things. That all presumed though, and this is exactly the thing, that comes crashing up against bots that might be smart enough to do the to work these sorts of things out. And guess what we want bots to be able to do now? That we want them to be able to do things like draw a human usable user interface, etc. So now they do have the tools, most of the time, and the cognitive abilities that you were mentioning, to actually go and say, I don't think that's a real okay checkbox. I think that's a trap. Alex: Mm. Richard: And so now those things are also useless because the bots have gotten just grown the ability to to to circumvent that kind of trick. And there may be further ones we can invent, but my sense is on that front it's just an endless looping war of measure, countermeasure, and and we could use AI tools to think of cleverer ways to keep AI tools out and then better AI tools will come along and work out how to deal with those. And we're not you know, we've really gotta go a different way. Alex: Right. I don't know as as I hear you talk about these things, you know, I have my own thoughts around the problem, but you know, this this question just pops on over and over again in my mind and I'm thinking, is this a net result, you know, much like spam, of the cost of data going to zero? Like the fact that traffic is is more or less free once you pay your monthly subscription. Richard: Yeah, I think so. I think it's it's definitely something that is possible because of that structure that we have, right? So we've set things up so that the financial you know kind of transactions happen within walled gardens, et cetera. And everything outside in the world is is is is a free for all, right? So You only ever get your credit card out at certain points. There's a couple of reasons for that. One is that that's actually fiddly right now on networks. And so people don't you know, don't want to like charge people for anything as it's going on because microtransactions are super hard on the traditional internet infrastructure. and the other one is just because of the you know, the way the rent is sought on the internet, right? Is that the you've got pay for access and then pay for certain tools where, you know, you kind of login and you you you have platforms and everything's inside that platform and they're trying to keep you in there so they don't want the friction to be outside that. They want you to have frictionless access to their platform and then you know that's where you're going to hopefully spend as much time as can be possibly drawn out of you, etc. and so, you know, like TikTok or Instagram definitely don't want people thinking about the cost of the scrolling. Right, because that's gonna put a dent in their scrolling. If people had a if it was expensive to scroll and people had a little meter that went off, they might stop doing it. Alex: Yeah, yeah. And you know, this whole idea that we've been pushing of the of the internet becoming the corporate, I think it's it's Richard: Mm. Alex: very much encapsulated in what you just said, because you know, there used to be a time when the internet was actually the internet and now it's just like you say, this like this wild space in between corporate networks, you know, whether it's meta network, whether it's Amazon, whether it's you know, Microsoft or it's it's just this this like very very or you know, what in in China what they would have over there. it's it's it's like there's there's this like huge private networks, the corporates, and then you you have some pipings in between them, and in between them and the users. But I think for for for the user experience and for you know, user empowerment, I think that a different form of internet would would make more sense. What what do you think? Richard: Yeah, I mean, so that's that's the approach we've taken with CIDO is to say and this this is interesting because it kind of gets back to to fundamentals of of network design. And the problem that you have with most networks is sibling, right? Like how can you tell a real question from a fake one? Right? How can you tell a real customer question from someone wasting your time? Like what's to stop your competitor just asking you a bunch of questions and using up all of your customer service time. And the answer that we've found is relevant time and again, whether it's to do with who gets paid for doing real work on the network or how do you mediate access to things, etc., is that you need realistic fees for doing things and then a a fair ungameable way of distributing those back to the people who who need them. And if you do that properly, you have less rent seeking. So People's money actually goes further, even though it seems weird to be paying for even, you know, the the access drip by drip. but it goes to the people who need to get it. And so that means that those, you know, people actually providing them services give those things at a at a more reasonable rate and have a better income stream to to do it with. usually 'cause there's fewer gatekeepers and intermediaries in between. Now what that means for th things like these bots is if They had to pay for every request that they made for data, whether it's trying to hack or whether it's just trying to find out the price of things on your website. they're going to have to give you a lot of money to make all of those requests. Right. And Alex: Mm. Richard: if you make that costly, you're you've got an actual vector for saying, well. Okay, maybe I can't stop you from doing it, but I can make you pay me more than it's it costs me to increase the size of the infrastructure, service your needs, et cetera, et cetera. and the the kind of thing that occurred to me suddenly thinking this through was I mean it's it's really interesting for me because, you know, we were talking before about struggling with bots. my the airline I fly with by preference because I have points and stuff and member there. are really anti-bot. They have a huge set of infrastructure to stop bots scraping the I think it's scraping the member prices. They're worried about that. and so what they do is they have, you know, all sorts of things. I mean if you have your VPN on, things don't work. If you do all sorts of, you know, any any kind of not appearing to be in the location that they think you're in, etc., it just kicks you out. Your session ends and you get you get logged out. You can't even connect. Things are very slow. And it occurred to me like this is all customer unfriendly. In this case, me being the customer very unfriendly to me makes it really annoying to deal with that website when that happens. and it's all just there to stop bots. But they could, using CIDO, if we built them an app. if we built their what we built their website on CIDO, what we could do is just say, well, okay, you charge per query, but that goes to credits or flights, right? and if you have a member, you could work out how to recycle that more quickly. So you could put it back into their membership account and you could use the points in your membership account to to make these queries. So all of a sudden, fine, if you want to scrape their website for all the prices every 15 minutes, go for it. But you're going to be one of their more, you know, valuable customers because you're going to be paying them, you know, tens, thousands of dollars a minute, whatever it is, to, you know, to get this information. and that's a bulletproof way of stopping people. Right. Because it's not relying on some lock that can be circumvented. It's just saying, sure, go for it. But, you know, this is the cost to you for doing it. And I think this is somewhere we're going to have to end up at some point. because I don't think there are alternatives. Particular, we've seen all this reverse engineering and stuff recently of software packages and stuff. This is going to happen universally, and there's not going to be ways. of even vaguely trying to like keep you know, choke the pipes or you know, keep people at bay or even tell who's a Sybil, right? Who's not realistic because the bots are getting better at pretending to be human than the humans can do. Alex: Right. yeah, I think I think it's so interesting that that you have to like rethink the way you access and the way you interact with the internet, right? And I think Richard: Mm-hmm. Alex: most people are are like so scared, like, no, you know, everything's going up in price. Why would I want to pay for for internet, you know, by by data? and I would say that The the prices you pay for data today, like for example, let's say you you're on a 5G plan and you have like twenty gigs and then you want another 10 gigs and it's like ten dollars or something like that. That's not a realistic pricing structure, right? That's the pricing structure that's enforced by your carrier because they want to make money off you. For for them, that extra 10 gigs is probably zero point one cent or zero ten Richard: Mm-hmm. Alex: cents or whatever. So When you operate as a network, you get the network prices, right? So because there there is no there is no in-between person that's trying to like make money off you. You have a direct connection to the network and you pay what it costs for the network to to to process that information. There is no added cost. There is no rent seeking on this network. The network pays for itself, right? That's what it was built to do. so I think that that helps people, you know, maybe tackle this question from a more realistic standpoint and not like fall prey to this idea of like I don't want to pay for data but for you the end user like everything will look gonna is gonna get cheaper and not only that you will start having control over your identity control over what you own right because the a network like like Sedok also has NFTs and has like various forms of NFTs and let's say you buy a game you play it, you're bored of you're bored with the game, you wanna sell it, okay, you can sell it to someone else or you can send it to someone else. And there is this this like back and forth between these two things. But when you look at what's coming around the corner with with with AI agents, because you know my personal tinfall hat prediction is that these these bots are being trained to just like round people off. You know like What if the internet gets filled with all these like rogue AI agents, right? Then everybody's gonna run to some like private network is gonna be like, please Apple, please Google, or please Meta, help us, right? Richard: think it's I mean yeah it's it's it's it's it's we don't know. Like that's certainly a thing. You could use you could use bots to kind of just fill spaces and and make it difficult for people to do things or just to listen to everything one one option. The other thing is that depends depends on what people are doing and how they're using the systems, right? So I might Now at any point, I don't know. I haven't counted, but I've got agents running on my on my home network as well as me running on my home and the other humans, right? And I don't know what percentage of the traffic is human or not coming out of in and out of our network, right? So at at times it's definitely mostly robots because I've set up a bunch of AI agents to do different work and they'll be going crazy doing research for me or whatever and they'll be doing it hard and fast. They'll parallelize it. So, you know, one agent will split set up multiple tasks to all go and look for a certain piece of information and then come back and report to it. So there'll be bursts of ninety-five percent of the traffic on our network being being robots. So to me, we gotta l get used to the idea that we'll live in a space where you cannot tell the difference between, you know, honest and dishonest actors. And I think this is this is a version of that game, playing that whack-a-mole game, explains a lot of errors in blockchain design, but a lot of generally the way platforms operate, etcetera, on the internet now, assuming that you you know you just have to build these clutchy, horrible solutions for that. That we all hate, you know, captures and things like that. But I think it's going to get to the point where it's so bad because human traffic just goes down and down. You know, porn is the last great hope for humanity that we keep the human traffic even vaguely as a percent of the of the overall traffic going around on the network. and if that's you know, if that's true, then we're going to need a system that doesn't care if you're an honest broker or a dishonest broker. And they're very I don't know of any solutions outside the ones we've found for for doing that. The solution starts with you just make everyone pay and the honest activity pays you back. The dishonest activity just a cost. If you work it that way, you've got some hope in a world where ninety percent of the a you know, the I was gonna say agents confusing here. Ninety percent of the actors, you Alex: Okay. Richard: know, people doing anything are actually bots, whether they're doing it for an individual, a company, a state, whatever, criminal organization, we we won't be able to tell. Alex: Right. So so what you're saying is that even if a bot is doing something of use, for example, it's being sent by a company or a user to, you know, buy a plate ticket, if it does that and that is considered, you know, correct behavior, right? Like it's not trying to game the system or anything, then even the bot can get its money back. But if they're just, you know, snooping around trying to scrape stuff and so on, then they'll just spend a lot of money and not get anything. out out of that spend. Richard: Right. So I mean that's a that's a a simple thing you could do is yeah, I mean i if someone comes asking questions at your store, you ask them to pay you ten cents per question, but you hand them a voucher for stuff in your store for that ten cents. If they're not buying, it's worthless to them. If they're buying, it's fine. It doesn't cost them anything. and in fact eventually i the customers might start saving money because the time waste is a fewer. Right. And so the overall cost of running this company doesn't involve the sort of vague attempts at anti sibling and working out who's real and who's not. Alex: Right. You know, all right, so w we are maybe a few steps removed from from total world domination with Sato, but Richard: Mm-hmm. Alex: until that moment, what can people do to protect themselves from the the proliferation of AI agents? Because you know, I know my mom turned off her phone, like literally n her phone won't ring unless it's someone from her address book. And that's because of all these scams that are Richard: Yep. Well I think she's right. That's what I do. That's the way to to handle this problem. and I think we'll have to adapt and have better habits but also e expect to be thinking more that way, right? and so I mean there were always prank calls, you know, Bart Simpson calling up and asking if the if you know Alex: Mm-hmm. Richard: if Mr. and Mrs. Wall are there and Yeah, if not, who's holding up the roof, right? This is this has been going on since time immemorial, right? Alex: Time immemorial, okay. Richard: well I can't remember Alex: Yeah. Richard: time before telephones. There are actually still Alex: Okay, gotcha. Richard: people around who can, but Alex: Gotcha. Richard: I just made up that definition of time immemorial. But anyhow, the the point there is I think that We've gotten past the the the stage where we can just expect somebody else to take care of this stuff for us. Right. And that's been the way of the world. We'll need to take personal responsibility and say, Well, actually I need to maintain an address book of people I want to talk to or not. And then going taking off that individual hat and saying, well, as someone who makes software for a living, as someone who builds things for people in the software space, we need to give people tools that make that easy to do, etc., and don't at the same time expose them to all sorts of risks. So what you were talking about before is a is a classic example of that. People do do that. They say, okay, I'll I'll go onto this platform where there's moderation and there's you know various things to stop you know, anti-bot projects and stuff to like protect me. But in doing so, they've completely given away their rights to their own identity. That identity can be deleted, it can be locked, it could be excluded, it could be censored, it could be drained of everything valuable in it at the platform's behest. and I think people haven't cared so much about that when they've thought about that as well, that identity is just my Facebook activity. It's okay if that happens over there. It's not too much of a loss if it if it gets killed. But when it comes to everything we do, our banking, our social interactions, our, you know, work, our like looking for employment, all of those things are happening in these networks, it's suddenly not so great to be in a situation where your entire identity can be nixed because you've given it over to some third party platform. And unfortunately, you know, that that includes the state. We've seen stuff recently where like the hacks on French government tax systems have put people in huge amount of danger because they've revealed who has all the you know, who has a lot of money that you may not have known and it's given people name and address of you know, wealthy people to to go after, you know, th a thieves treasure trove. Right. So We've got we've got to start saying you need to take personal responsibility. We need to build tools that let people do that and we can't make those tools because they're so fundamentally important. They're they're covering people's health information, their you know, their finances, all of those sorts of things. We can't make them centralized in a way where they become, you know, a target for hacking or other kind of attacks. It needs to be left to people's control and people need to have the ability to you know, encrypt things and store them securely themselves because we know that if you distribute that task it you know, it just creates problems. Alex: Right. I think, you know, for for the history buffs, we would say something like we're we're leaving the Pericles era and we're going towards the Thrasybulus era. And you know these are two different eras in in Athens history. Pericles was known for for making the life of Athenians better, and Thrasybulus was known for making the Athenians themselves better. Right, because the the easy life leads to a lack of self agency and taking responsibilities for your own actions, right? And that's how you get this, like everybody's happy, but the society as a whole is in decline. While Trassi Richard: Hm. Alex: Bullous was like, We're not gonna make life easy for you, but each and every one of you will become a better person. And I think that that's w the era that we're entering in. And it comes with a lot of a lot of caveats, but it also comes with a lot of perks. chiefly being ownership, right? Richard: Mm-hmm. Alex: privacy and and and and control over owns over your own identity which Richard: Mm-hmm. Alex: as we are starting to find out a lot of people want to have control of that right like the the EU wants to issue your digital IDs together with the banks Richard: Hmm. Alex: I wonder why they also want you to use the the digital euro which is the CBDC I wonder why Right, because once they lock you in a system and there's no way out, they can tax you, they can force you to eat whatever they want you to eat, they can force you to do it's literally empire, right? They're not cutting off hands of Indians because they're planting cotton instead of opium, instead of poppies, but they're doing something similar. So I know this is a bit of a off tangent, but Richard: Hm. Alex: it it is i it it is very important to use these very extreme examples because they they kinda like stick out instead of saying like, it's easier to buy coffee. What would you rather have? Easier to buy coffee or easier to live your life? You know, it's yeah. Richard: Hm. No, I think the I I really like that Periclesh I can't pronounce my pronunciation is terrible. Thrashibulus. Alex: Parsibulus. Yeah, yeah. Richard: It's I always think Thracian, which is the wrong language. Thrasybulis example because I think that's that's where we're at. I think people want and hope for you know, someone else to do the heavy lifting on these things, but in the end it's setting us up for that. And people are all decrying, look at the state of the internet, etcetera. but another classic thing is, you know, Australia we've got increasing regulation on feeds, on how feeds can and can't work in social media apps. And the intention is to create a healthier environment for people because we know that these carefully engineered dopamine traps can ruin people. And so a state that's interesting and in interested in helping with that is wanting to do something about it. Right. The issue is that you can't you you can't help people that way. Right. You're just keeping the alcohol a couple of centimeters further away from their their grip. But someone will bring them, you know, what we call Sly Grog in Australia. We'll slip in some, you know, this is a digital world, right? We can't kind of just like shut all the pubs. They'll be speak easy, right? We know that that doesn't doesn't work in the real world, but it certainly doesn't work in a in a you know, in the digital world. We need to, you know, teach people, especially and funny thing is I think the you know, having kids and watching them interact with the internet, they're way ahead of most adults on this stuff. 'Cause they've they've grown up with it. So they understand things like what's a threat in the digital world much better than older people. And so their base skills are higher. but things like cryptography, et cetera, are not simple. We can't expect people to encrypt stuff with algorithms they came up with and wrote themselves. We need to build tools at the same time that Put people in a position where they they do need to do some work, they do need to learn, but in doing so, like you said, they become better citizens of the internet, they become better people and they're better able to protect themselves, particularly as the whole landscape's going to be utterly different in five and and ten years. Yeah, it's gonna be things we cannot predict. And so, yeah, to me that that is the right model is to say, well, okay, we're going we have to treat this from you know, the roots we need to go back to fundamentals, look at what it is to be in essence a citizen in this space, and to be a competent and capable one that you know, that can actually navigate this space, we cannot be handing it off to third parties because we know for a bunch of reasons that, you know, even if we assume that they've got the good in mind, which they do not necessarily, there's still limiting factors. It still doesn't work in the end, because of the way that it disempowers most people. Alex: What would you rather like, you know let's say you're a VC and you invent you invested in in Facebook like Peter Thiel, you invested really early and you turned ten million dollars into half a billion dollars. how could you do something like that today? Right, you don't have access to all these startups coming up, right? But what you can do is you can put yourself in a network that charges people for access and traffic and so on. So, in a way, the more people join Sato, the more the other people have to come to you, right? And as as you are like a part of the network. Obviously, companies will want to have access to the users. But Richard: Mm-hmm. Alex: the moment they want to do that, they have to enter the network. So you can stay out in the wild corporate net and be played left and right with all these different accounts, or you can join a network where you automatically, when you become part of the network, you become a shareholder. And as the network grows, you also grow and you you grow in control, you grow obviously in your investment and so on and so forth. So there's there's more than one way to think about what is possible with these networks and this idea of like, you know, paying for data is not necessarily Richard: Mm-hmm. Alex: a bad thing. You know, we've been we've been raised wrong. Richard: Yeah, no, I absolutely I think it's it's the kind of it's the classic you know myth of the American dream, where, you know, the absolutely free for all internet, with you know, Wild West internet which was its creation. It was just a blank open space. In fact, yeah, Wild West actually wasn't, so it's wrong that way, but yeah, it was Blue Ocean, whatever you want call it. and that created unicorns, that made you know, y empires and fiefdoms and kingdoms and so forth. And you know, the number of new ones got fewer and fewer. But it's still it still kept us that that kind of hope that you could you too could be part of that. Right. And and in fact that's less and less possible because A the ground is taken up, but B, the people who've already succeeded are doing everything they can. To defend that ground and to exclude others. Alex: It's the free trade model. Free trade works for empire, right? That's it. It doesn't work Richard: Yeah. Yeah. Alex: for the small nation of Kyrgyzstan, right? It works for empire. Richard: Yeah. Yeah. Alex: And and it's the same concept happening on the internet. Richard: Well it's also 'cause I s you know, it's it's free because it's lacking rules to control me, but I'm able to Alex: Mm-hmm. Richard: exert any now any amount of power in other ways to hold to control you. But we're playing by the same rules, so it's okay. And I think yeah. Alex: Yeah, exactly. Like it it's more like when you try to g get support from YouTube and you end up talking to AI agents and you can't get a real person to talk to. Sorry. You Richard: Right. And it's Alex: know, it's like that episode of South Park. We're sorry. You know, BP spills sorry? Richard: Yeah, computer says no. Computer says no, right? Alex: Computer says no, exactly, exactly. Richard: Yeah. Well no, it's it's exactly that. And I think the thing to do is not to say, well, I want my shot at that, because we see how that works. I mean, maybe you'll maybe you'll win the lottery three times in a row, which is the amount of luck you have to have to have to like be that guy, right? Basically. Even if you have a great idea, great plans, great so forth. there is another way where if you enter a system where the rules are fundamentally locked to being Even and there isn't the ability to exert power, right? Because you've done well in that system. then you've actually got a hope of okay, it may not be world domination where you can get to a position of power where you can squeeze out others because the rules are stopping you from doing that. But it does mean you can thrive. It means you've got a very good chance of thriving and doing very well. We just need to change where that picture is. And I think you know, everyone has an has in their heads, well, you want to kind of we're we're valorized at the point where people kinda I want to be on the big app, right? Because this app's very successful, so I want to be there. Like w what what why? Right? And I think there is some sense for for some people that I want to listen to the most popular band because they're popular. And that means maybe I'm doing something with other people. But it doesn't actually speak to the quality of the music or how much I like it. And I think getting our heads around no, it's really great to live in a town where there's cool music going on in every cafe and bar and and club and park might just be a better way to get music that you really like and feel like you're part of it and enjoy it. and it doesn't lack for anything. And I think that's that's kind of where we're at this tipping point where funnily enough, the fact that we're living as humans in what might end up being a sea of robots online. might cause us to need to r reassess and I think look for the tools and things that let us act with more agency ourselves. and that will be, you know, a very interesting outcome from all of this. But I think it's something that, you know, I can't say when we were building CIDA we were we saw the robots coming, but we did build it because we saw the problem with these dynamics on the n on the internet as it stands and how horribly they're being misused. And we set out to build something that prevented that from h recurring and recurring and recurring in the ecosystem and instead tried to set some rules that came out with different outcomes. Alex: Right. Yeah, i if if Apple is known for their walled walled gar garden, I guess Seto would be will be known for their walled comms, right? It's it's it's it's Richard: Yeah. Alex: it's walled but it's for for everybody and it it doesn't really, you know, try to extract anything from the people using it, just making sure that the comms remain the comms and they're well taken care of. Richard: Yep. Yeah, it's it's it's designed to be a commons that actually works for everybody, right? That's the that's really the point is that it's commons where you don't have the tragedy of the commons, where you don't have people exploiting it and ruining it for everybody else because of what we were saying before about the the relationship between cost and getting paid back within that space. Alex: Perfect. Well I think we we kinda covered all our topics for today and Yeah, I think it's crazy. Like you said, you know, it wasn't built for AI agents, but turns out it also protects you from AI agents, which is a very, very Richard: Well Alex: nice turn of events. Richard: yeah, it comes back to that if you can't if you build something where you understand you can't tell the baddies from the goodies and you just build something that works for everybody by rewarding the the behavior that the network needs, you've solved that problem when it comes along. So we didn't I can't possibly say we anticipated it, but the system is built to structurally deal with this or any other problem where, you know, power can be misused in the network. Alex: Yeah, it's it's is the difference between preemptive medicine versus palliative medicine, right? One Richard: Right. Alex: seeks to keep you healthy, the other one like seeks to remove the pain once you're sick. And it's it's Richard: Yes. Alex: a whole different way of looking at medicine and Richard: That is indeed a good analogy. Alex: Yeah. So thank you so much for today's episode, Richard. Thank you everybody for watching. And don't forget Richard: Thank you. Alex: you can also watch this as a video on demand on YouTube and your favorite podcast app every Sunday. there's a new episode kinda comes out twelve PM UTC. thank you once again and catch you guys next week.