私たちの文章でAIは作られた。そして今、それを借りている（第320回）

Thank you for listening to Data Science at Home podcast with Francesco Gattoleta. You are about to get cutting-edge insights from the people who are reshaping the world of technology with machine learning, data science, and artificial intelligence. It's time for Data Science at Home. Welcome to the show. So let's play a quick game. Imagine your hospital worked just like your phone plan, right? So imagine you have like three tiers, you have a chatbot in the waiting room. If you pay 20 bucks a month, you get an actual doctor. Or if you pay enough, you get the doctor who actually reads your file before walking in, right? Sounds insane, right? Well, people would be rioting right now. We would call all this like a broken healthcare system. And believe me, it is. And in fact, a few countries are basically already leaving it.

Now, swap hospital for AI and swap doctor for the AI that actually gives you the good answer instead of the stupid one, right? Well, you just described the pricing page of every major AI lab in 2026. So today's argument, and I'll be very careful about it, you know, not to sound like a rant, probably as usual. AI is, you know, I go without, this goes without saying, creeping into every single part of our lives. Pretty much the same way electricity and water did. And it should be treated the same way. Something everyone gets a baseline of You know, not something that you kind of rent A bit like Netflix, right? So, there is also a US senator I spent already a few words in a previous episode about this Who just proposed, well, recently proposed a law

To literally take half of big AI and hand it to the public And I'm talking about Bernie Sanders wants to do this in my opinion he has he had a great instinct when he did this bill though it could be the wrong tool so by the end of this episode you'll see exactly why at least this is what I personally think and I try to give a sound explanation about an interpretation about what a bill like this can do and what we should do instead but of course I'm no politician I'm just an engineer who sees things and, well, is watching the game, being played again and again and again. Welcome to Data Science at Home. I'm Frag. Let's get into it. So, let's define one word people throw around without ever explaining it, which is public good. In fact, it's two words, but you get the point. So, a public good, what is this?

Well, it's something like a park or like clean air or a lighthouse in your city if you're of an arbor or something. So essentially, it's everyone gets to use it and can use it. So me using it doesn't stop you from using it. You know, think about clean air. How can you prevent people from breathing and quote unquote using clean air, right? Even though I pay for it, you know, that doesn't mean anything. Other people can and will use it, quote unquote, like breed. Now, AI, the way it's sold to you right now, in fact, it's not that. It's just a login and a credit card away. So economists have a name for that as well, which is called club good, you know, like in a golf club. You pay the membership and you walk through the gate. You don't pay the membership and you stay on the sidewalk. And the only thing you can do is watching the game. So what AI was actually built from

Is about as public as it gets So think about ChatGPT, Cloud, Gemini All trained on Wikipedia, forums Centuries of books, government records Even your own conversations If you ever hit thumbs up or thumbs down on a reply Sam Altman, the guy who runs OpenAI And probably reads most of the data more than everybody else, literally describes his own model as trained on, and I quote, the collective experience, knowledge and learnings of humanity, which is just a fancy way of saying, built from everything we all wrote for free without anyone asking, right? So, look at what actually happened. We, like all of us, collectively, wrote the raw material or created the raw material of these models, right? a handful of companies now spent a lot of money it's a billion so far

turning it into a product now they sell it back to us in the subscription tiers the same folks who actually contributed to generate or create that raw material many times without knowing so of course nobody stole anything legally speaking and I'm not here to re-litigate copyright law There has been already a few cases about that. But it is a pattern. And we have seen this movie before. That's why I say I'm watching the game being played over and over again. We got that with the oil, with land, the radio airwaves. Something that belonged to everyone gets fenced off. And eventually society decides that it needs some public rules. Not just whoever had the most money first gets to keep it. Right? so this is why at the beginning of this episode I made that example well that analogy with healthcare and education in fact, which are the right comparison

in my opinion, not just you know, an analogy that I made for the sake of clickbaiting both are things where if you are locked out the gap doesn't stay small but it actually snowballs so for example, a kid without healthcare early in life pays for that for decades during its life later. Or a kid or a small business or even a public hospital stuck with the free worse AI while everyone rich gets the good one You know you have the same snowball there So we are one generation away in fact from finding out exactly how big that gap gets And by the time we have the data to prove it, I'm afraid that the gap will be already permanent. So Bernie Sanders wants to nationalize half of Big AI. And there is the actual plan. And in fact, there are three points that he made. And I'm oversimplifying. I'm absolutely not speaking legalese today.

I'm just oversimplifying and give you really the too-long-didn't-read version of that bill. This episode is supported by OutShift, Cisco's incubation engine for frontier technology. Multi-agent AI is in pretty much every enterprise roadmap right now. But there is a problem with that. and we are all facing it. Agents can pass messages, but they cannot think together. So these systems silently underperform through cognitive failures, things like misreadings, unverified claims, and of course, false consensus. You know, all the stuff that conventional monitoring cannot detect. OutShift by Cisco is building this fix, the Internet of Cognition, which is an open source foundation for building multi-agent systems from design to production. With shared contacts, shared memory, and guardrails to drive the results that we actually expect from artificial intelligence.

So read the paper, experience the demo, grab the code from outshift.com. That's O-U-T-S-H-I-F-T dot com. So, three parts. One, any big AI company, meaning it makes at least $200 million a year from AI stuff, has to hand over half the company, right? Not half its profit, half the actual stocks, right? One time straight into a fund the government controls. That's step one. Step two or point two, if a company does AI and something totally unrelated, for example, an AI company that also happens to sell, I don't know, toasters they have to split the business lines so that the government can only take a slice of the AI side and point three that giant pile of stock pays a yearly dividend, like pretty much any stock does, some of it goes

straight to Americans as a check okay, something like a thousand dollars a year as an estimate basically the same idea as Alaska, where residents get actually an annual check because the state owns a piece of the oil. The rest goes into health care, education, housing, and so on, you know, all public things. Now, a seven-person committee picked by politicians controls how all that voting power actually gets used. That's all in the bill, right? So add it up and you get roughly $7 trillion, a fund bigger than every country's sovereign wealth fund on earth, except probably not always, created by one bill. Now, let me be straight to the point, because this is less crazy than it sounds at first. OpenAI has floated a public wealth fund tied to AI growth. And so did Antropic, very similarly,

has talked about national funds with AI stakes. Even Elon Musk has said the fix for AI taking jobs is basically free money for everyone. so Bernie Sanders isn't actually pulling this out of thin air he's talking what DAI companies themselves have already said in interviews, in blog posts, etc and he's actually writing it into a law but that's pretty much it now, what these companies have said is disputable or it can be interpreted even the reason why they did that was probably for marketing or for, you know, whatever. Did they really mean that? That we cannot tell. But, you know, it's hard to fight a bill when the industry already agreed, at least with the idea. Now, the mechanics and the specifics probably need to be polished, but the idea of redistributing

AI, let's say, with the people, is pretty much there. so why do I still think it's the wrong tool let me give you three reasons not one reason one rich companies are extremely good at making money disappear on paper and we have seen this many times you know the second this bill gets serious expect really an army of lawyers to spend a decade or more arguing about what actually counts. Is this subsidiary really an AI company or, you know, it's just because the toaster is a bit smart doesn't make you an AI company. You know, there is so many things that lawyers can find and can appeal to. Is that one conveniently just under the 200 million line? You know, they will start playing with the thresholds in order to disappear from that category. You get the point, right?

There's even a rule in the bill to stop companies from just moving overseas to dodge it And that rule alone will get fought in court for years The people this is supposed to help will not free a dollar until all the lawyers are done eating We have seen this before with big international cases This is a preamble to another one The second reason, and this is kind of the bigger one a check doesn't really fix the actual problem. And I want to be extremely clear about this. If good AI becomes the thing that decides who gets hired, who gets fired, who gets a fair diagnosis, who gets decent legal help, whatever, mailing everyone a thousand bucks a year while the good AI stays locked behind the paywall is like giving someone cash for their illness instead of just letting them see the doctor. So you solve the money problem but you didn touch the who owns who actually gets to use the good tool problem right The third reason and again nobody wants to touch this one

who actually deserves the check? Because in the bill, they mention, you know, the dividend goes to Americans because it's a US law about US companies, which is fine. You know, I get the legal logic there. But the thing being taxed wasn't built by Americans. It was built from books, forums, conversations from basically every country and language on the planet. The reason why any large language model can deal with more than 100 languages, while an American can barely speak one, makes you think that probably the data come from outside of America, don't you think? So most of those people, in my opinion, will never see a cent. if Sam Altman is right that this is the collective experience of humanity and I quote him when he said that then why does one country get the check for something humanity built together? That would be a fair question. Nobody has a clean answer on this.

Well, nobody has ever asked this question. So yeah, there's no world government that can tax a company for scraping the whole planet's writing. So, realistically, if this bill passes, which I believe it will not, but even if it passes, other countries can just copy it for themselves, and we end up with a stack of separate national piggy banks instead of one fair global one. It's worth saying out loud so nobody pats themselves on the back for solving the global problem with a very local check, right? Now, there is a bigger point underneath all of that. this bill treats AI just like another wave of automation, you know, like the assembly line or the printing press, where the old fix from, you know, a century ago was tax the winners, hand out the winnings, right? That worked okay when the technology was a tool

that helped regular workers do their jobs a bit faster, you know? Think about the printing press. it works a lot worse, of course, when the technology is becoming the thing that every other system, hospitals, schools, courts, the news, everything runs on top of, right? So nobody fixed only rich people own printing presses by taxing the presses and mailing everyone a check. You know how they fixed it? They fixed it with public libraries and public schools, okay, that taught everyone to read, no matter who owned the press. That's the fix. It's not a check. It's a library. And so I see the analogy there. And this is the only thing that even Bernie Sanders is not talking about. And I'm fairly impressed because I consider Bernie Sanders a very smart guy. All right. So what would you actually do, you may ask? Well, that would be a fair question.

National SDI companies is a bumper sticker. It's not a plan, in my opinion. So that would not solve the problem. You know, it would just keep people busy thinking like we did the right thing. But it actually doesn't solve any problem. I'm an engineer. I try to find solutions to problems that exist or don't exist yet. But this is one that does exist. So here's an actual plan, how it would look like, at least to my eyes. And, of course, I'm very open to, actually, I would be extremely glad if you jumped in and criticize this or point it out something that doesn't make sense to you, opening some kind of debate that is a nice commenting system on YouTube, but you can also jump on Discord and we can have a chat about all this. So let's go back to the actual plan, how the actual plan looks like. And, you know, it's quite different from what Bernie Sanders proposed. So the first thing that I would say is that it's about public AI computing, computing power, you know, like a public utility.

So governments already do this for supercomputers. Europe has one as well. It's called, I think, Euro HPC, high-performance computing. It's real. It exists today. So just extend that to AI specifically and let public hospitals, schools, researchers use it at a cost, not at the Silicon Valley prices you know, nobody thinks it's radical for a government to run a power grid same idea second problem second point, sorry a public open AI model and when I say open AI, I mean open space AI, okay, it's open as an attribute, not as a company the real thing open AI model that's good enough to be a safety net not the best AI on earth we know, just a floor that everyone stands on every school, every small clinic every small town hall should have some kind of baseline AI that is not rented

from a company that can change the price or the rules whenever they want and this already exists in early form in France and Germany they both have public AI projects and there is an open project called OLMO, O-L-M-O built specifically to be transparent. So the idea already works. It just needs more money and, you know, definitely political will, which are the only two things missing, in my opinion. Third point, give the public actual rights, not just a tax and check, okay? The way we already regulate electricity and water companies should be replicated for AI. If an AI company's product runs a hospital, a school, a government office well, they should have to show their homework where the training data came from let regulators check under the hood and offer a fair price to the public sector instead of coaching it the fourth point, a public data trust if the public's own writing and conversations

train these models, the public through some kind of independent trust not one government office that can be bought or bullied, you know, as it is now, should get an actual say in how that data gets used. That fixes the we all wrote the textbook and someone else is selling it back to us Problem Instead of patching it after the fact with a check The fifth point, the slow but real one, teach people to actually use this stuff. This is kind of already mentioned in the EU AI Act. If we are worried about AI taking jobs, well, the real fix is treating knowing how AI works and how to use it well, the same way we treat literacy. A basic skill that every kid gets for free in school now. Of course, it will take a generation. It's the slower path. Slower than a wealth fund, of course, but it's the one that actually compounds

the same way education always has. Now, none of this means shutting down private AI companies. It's a good thing. And I'm not saying that. they should keep racing to, you know, build the best stuff. That competition is, in fact, genuinely part of why progress has been so fast. The point is not no private AI. The point is, don't let private AI be the only AI. And, you know, the same way we don't rely on private hospitals only, or private schools only, or even private electricity only. You know, we have so many electricity providers for many reasons, you know, to maintain business continuity, to create a marketplace, to choose for the best prices, to choose for the best quality, and so on. I want fossil fuel or I want fossil energy or I want renewable energy and whatever are my ideals, I can choose. a public floor

with private companies still racing above it is a good thing that's a realistic version of this argument not a 5 year plan of course ok now where could I be wrong I would be a hypocrite if I made this whole case and pretending there's no downside so let me be honest with you governments are slow terribly slow AI moves at an insane speed as you probably have noticed A public AI project could easily end up two generations behind the good stuff Funded by budgets that get decided at every election Not every time the technology changes That's the real risk Europe in particular has a track record of big public tech projects Running years or decades later And I'm not going to pretend otherwise There's also a trust problem A government-run AI system isn't automatically more honest or fair than a private one. You know, you're just trading one group you have to worry about for a different one.

That's what you're doing. Public broadcasters have absolutely been hijacked by politicians before and lobbyists and stuff like that. And funny enough, Bernie Sanders' own bill hands control of $7 trillion in voting power to seven people picked by politicians. That's not really solving the too much power into few hands problem. That's just moving the hands, right? So there is a fair argument that private money moving fast is exactly why we got so much progress so quickly these last few years or decades. Governed contracts crawl while private investors don't. Well, none of that kills the argument for me. It just means don't go all or nothing. Keep the private race going at the top eventually, but also build a public floor underneath so nobody falls through the gap. But if you ever hear someone make this argument without admitting these downsides, well, be suspicious.

That's not analysis. That's probably a sales pitch. So, here's where this lands. Nobody, I say nobody, argues that AI shouldn't be regulated or that people who lose their jobs shouldn't get help. Of course they should. The real question is much simpler than that. And it's this. Do we treat AI like electricity, water, and schools, something everyone gets a baseline of, no matter the bank balance, or do we treat it like a Netflix subscription with a sorry for the inconvenience check mailed out afterward. Bernie Sanders is right that regular people shouldn't be shut out of something built from everyone's own words and effort. And he's not even making that up. The AI companies said it about themselves first. So they are saying the same things, just in different lingo. But a $7 trillion tax that pays out a national check for something the industry itself calls

a collective achievement of all humanity. Well, that's the wrong tool, in my opinion. It's a right instinct, as I said, but the wrong tool. You don't fix only rich people own the printing press by mailing everyone a check for the newspaper they, by the way, can no longer afford. You fix it by building a library next door and letting anyone walk in. That's how you fix that problem. So that's the episode. if this made you think or well a little annoyed at me or at Bernie Sanders probably more to him well that's good because it means it worked jokes apart you can find me on discord and of course on datascienceatone.com and don't forget to subscribe to the official YouTube channel from which you are probably watching this episode but if you're not of course feel free to jump on the YouTube and subscribe there, that would help me a lot. And comment and like and share again. Thank you so much.

Hey, quick one. I wrote a book, The Mythical Token Month. Here's the idea. AI coding agents feel like magic, but they don't remove the hard part of building software. They just clear out the busy work around it, as we all know. What's left is indeed the actual thinking. Now, I have to say that no AI wrote a word of this book. I have to say that. it's a self-claim but you have to believe me it's out now as an e-book NordiRM the website mythicaltokenmonth.com links in the show notes thank you