A screwdriver that talks: Mookie Tenembaum on AI, war and power

This conversation with Mookie Tenembaum — writer, essayist, and analyst of artificial intelligence — was recorded at the Solanas Convention Center, Punta Ballena, Maldonado, Uruguay. It takes up AI’s role in security, medicine, warfare, disinformation, and authoritarian control, with particular attention to the technological triopoly that defines the global balance of power.

The interviewee’s opinions are his own; the Conversations section does not imply doctrinal agreement.

Jimmy: Hi, Mookie — how are you?

Mookie: Very well.

Jimmy: What a beautiful spot we’re in.

Mookie: Beautiful, really.

Jimmy: I want to have a conversation with you about artificial intelligence, looked at from a few different angles. Maybe to start, just take five minutes to pour out some ideas — whether it’s worth calling it intelligence at all, or whether it’s an algorithm with a very sophisticated technique behind it that lets us generate a very useful tool. And also whether it’s really a technology that learns, that understands, that has even a little bit of consciousness. Then we’ll move on to more present-day matters.

Mookie: Let’s start with this. Artificial intelligence is a machine. It’s a screwdriver that talks — well, not much more than that. Very sophisticated, very interesting, but it has as much consciousness as that tree over there. None at all. It’s just a language model, one that took an enormous amount of time to develop, and it is never going to have consciousness — no matter what we do, no matter how sophisticated we make it. Because to have consciousness, you have to suffer.

The basis of everything we humans do is fleeing suffering. But it’s suffering that sits behind civilization. If we hadn’t suffered we wouldn’t have risen; and if we hadn’t risen we wouldn’t have built — and wouldn’t have destroyed, either. We wouldn’t have done any of the things we’ve done. They are all the result of it. Otherwise we’d have ended up like zebras. We’d be nothing.

Jimmy: You mean zebras?

Mookie: I mean any animal. No animal has consciousness. Animals are, in my view, biological computers — nothing more than that. They have a program, they run that program, and they don’t know how to do anything except run it.

Jimmy: I think they have a little bit of consciousness.

Mookie: Well, we can disagree. To me, animals don’t have consciousness — if they did, they’d have civilization, they’d have something. They have nothing. They feel pain, they don’t suffer.

Jimmy: They feel pain, they don’t suffer?

Mookie: When someone gives you a vaccine, it hurts — but you don’t suffer. When a masochist gets hit, it hurts him, but he doesn’t suffer. Pain and suffering are not the same thing. Animals feel pain — of course they feel pain — but they don’t suffer. Because to suffer you need distance, you need to be able to watch yourself do it, which an animal can’t. We’re not better for having consciousness, and we’re not worse. We’re different. It’s another thing altogether. That’s how I see it.

As for artificial intelligence — should we call it intelligence or not? I think you can call it intelligence. But it has one flaw: it never says I don’t know. And that’s strange — it’s one of the things you notice. If it doesn’t know something, it ought to tell you so.

Jimmy: I know people like that, too.

Mookie: Sure — but in this case, maybe it really earns the name. It’s spectacular. Look, every job that’s done sitting down has vanished. It doesn’t exist anymore; AI is replacing it. It’s a matter of time — and not much time.

Jimmy: Good. Let’s go to the questions. I wrote them out so they’d be precise. The first: a doctor with artificial intelligence should replace a doctor working alone — but artificial intelligence on its own shouldn’t replace the doctor.

The second: the Hamas attack of October 7th exposed the limits of security systems based on artificial intelligence — they can be deceived, blocked, or destroyed. The attackers exploited blind spots, jammed signals, and destroyed cameras, which left the automated detection systems inoperative. How can AI systems in security and defense be improved so they’re less predictable, more resilient against cyberattacks, and kept under effective human supervision?

Mookie: Let’s start with the doctor. They should replace doctors — the sooner the better.

Jimmy: Sooner the better?

Mookie: Of course. Things are judged by their results, and by nothing but their results. Artificial intelligence has just passed exams that doctors can’t pass.

Jimmy: How is it going to do the work, though — the back-and-forth, the bedside method we have?

Mookie: That’s a good question. But there’s a machine that works, and works beautifully. And yes — it should replace doctors, engineers, lawyers, and any line of work where knowledge is the base of the profession. Knowledge — conocimiento; notice that in Spanish the word carries cimiento, “foundation,” inside it — knowledge is the foundation of the profession. That’s my opinion. Let me be clear: it’s my opinion. I’m not saying this is how it is. It’s what I think.

As for October 7th — artificial intelligence played very little part on the defensive side. In fact, thanks to AI, Israel took Hezbollah apart in three weeks. Here’s what happened. You saw it, when the beepers exploded.

Jimmy: The golden beeper — the one [Netanyahu] gave Trump the other day. Did you see it?

Mookie: They can be set off, you know — switched on.

Jimmy: Ah, I don’t know if he’ll switch it on. It hadn’t occurred to me — he’s a great friend, I don’t think there’ll be a problem.

Mookie: When the various beepers went off — and the beepers themselves, of course, had nothing to do with AI; that was brilliant work by the operatives — there were drones flying overhead. Ours, I say — Israel’s; I say “ours” because, truth be told, I feel like part of it. Those drones picked up the explosions in different places at different moments, all that information was fed into the AI, and now it was easy to find people, because this one was with this one, who was with that one, who was with that one. In a matter of hours the AI prepared an attack plan, and three weeks later — sure, there are still some men with revolvers, but Hezbollah as a dangerous armed force against Israel was finished. Eighty percent of the missiles were gone. They had a thousand missiles pointed at Israel.

Jimmy: And the missile-defense system — does it run on AI?

Mookie: I don’t know much about it, because honestly they don’t tell me. But it clearly works on AI. What was on the border, though, were people — the girls on watch were the ones at the screens. If anything, it was the opposite of your question: it wasn’t automated enough. And of course mistakes were made. There are some odd ones, too. You could think there was a spy, because the information those men had was simply too good. Or you could think we just fell asleep at the wheel, the way we did in the Yom Kippur war. But things happen that are unforeseeable — and that’s exactly why I think AI may fail to account for those cases, because in the end it works from patterns drawn from huge amounts of information. New things, though —

Jimmy: New things—

Mookie: New things are always old things remastered. What is analogy? How do we learn? By analogy. We see that something works this way, and we try to put it to work somewhere else. That analogy — that, basically, is intelligence. And that’s what AI does: it learns from here and applies it there.

Jimmy: And if it’s so good already — do we have to wait a couple of years for it to improve?

Mookie: No doubt about it. But it isn’t going to replace — the armies are going to use AI alone.

Jimmy: But it’s hackable, isn’t it?

Mookie: Right now, AI — no one knows how it works. It’s a black box. You know that pigs go in one end and sausages come out the other; how it does it, nobody knows. There’s a whole project trying to figure out how it works, and so far without success. So if you don’t know how something works, you can’t hack it — it isn’t easy, at least as far as I know. I haven’t yet seen a serious AI that’s been hacked, with evidence that it was hacked.

The system runs on something called weights — pesos, in Spanish. Fascinating.

Jimmy: And the people on the other side are going to want to get hold of a lot of pesos, then.

Mookie: Ha — right.

Jimmy: It’s all about the millions.

Mookie: Exactly. The weights — and it’s the weights that mean you can’t know what’s going to happen. On top of that, you add biases — because you don’t want it teaching people to build a bomb, you don’t want it antisemitic or racist or whatever. So we stick our fingers in — and our fingers are pretty fat — to make it more predictable. You follow?

Jimmy: I do. But it has no information about the future.

Mookie: Because the future doesn’t exist.

Jimmy: Once it’s fed on the information of the next hundred years, it’ll be able to give us answers about the years to come.

Mookie: No, I don’t think so — that’s why I say new things can knock it off its axis. That’s why the human role —

Jimmy: That’s why I think the human role is important.

Mookie: But — see, I’d say there’s no role for the human at all. My opinion: we’re going to go from being the protagonists to being witnesses. It’s AI that’s going to do it to us.

Here’s the example I always use — so this gentleman has heard it and already knows what I’m about to say. Say you want to build a building. The one I love is the Tokyo Metropolitan Government Building. So tomorrow Tokyo wants a new building. They open a tender. They call in the ten biggest firms in the world; the various architects present ten projects. The city government splits into two committees, each reviews five, each picks one, then they choose a finalist. That’s how an AI would already work today — except there aren’t ten, there are ten thousand. And now two things happen. First, with ten thousand, we get diluted. We’re not going to disappear because Terminator shows up to kill us — we’re going to dissolve. It’ll happen by dilution. We’ll have less and less say in everything that happens. And then AI just — you know they put two AIs together to communicate with each other quickly, and they invented a language nobody could understand? That’s what’s going to happen. They’ll talk among themselves. Remember when we were little, and your parents would talk with the aunts and uncles about what they were going to do — they’d take you off to one side and you’d be standing there going, where are we going? Well — that’s what’s going to happen to us.

Jimmy: And it’s a good thing?

Mookie: It’s a good thing.

Jimmy: In what sense? It’s a little frightening.

Mookie: Yes, but people were frightened when the car came out. They said it wouldn’t catch on, because people wanted to stay with the horse — the horse, there’s a relationship there. This happens every time a technology appears. Except this technology has one very special thing about it: it’s the only technology that comes with a loaf of bread under its arm. You know how they say a child is born with a loaf of bread under its arm? It’s the only technology that can correct itself — the problems it brings, it solves itself. A good example is energy. We’re going to need a huge amount of energy for AI, right? We’re hearing it everywhere. But who’s going to solve that problem? AI. It’ll find such efficiency with energy that we won’t need so much. That’s what makes it unique — as long as it has information, a lot of information on the subject. Otherwise it can do nothing. It needs the information first.

Jimmy: And we’re running out of data.

Mookie: You know that today the AIs already have no more data to take in. The only way left is what’s called synthetic data — and this is genius. We’re eating our own tail. What do they do? They take data the AI itself generates and feed it back as data. So basically the data being created is synthetic data, not real data. It’s finished. The AI already knows everything; it already has access to a hundred percent of human knowledge.

Jimmy: And if another Einstein is born?

Mookie: If another Einstein is born — what role does he have? I don’t know. I don’t know what importance Einstein has for us. Einstein matters because he gave us a window onto a particular world. But for that to happen again, and to compete — remember, [the human] won’t disappear, no. People will say, ah, but human creativity — we’re animals. We’re arrogant animals. We’re — sir, he said, we did three things, we grabbed, we shoved it all into the AI, and suddenly it does everything. Look how good we were. No. In a little while the bastard — as I sometimes call it — learned everything, fast.

Jimmy: Fast, fast.

Mookie: Good. Let’s move on to AI in war and military conflict. Artificial intelligence is revolutionising warfare by allowing real-time decision-making — you touched on it a moment ago — and autonomous targeting. But the ethical and legal implications of these decisions remain unclear. Should artificial intelligence have the authority to make lethal decisions in armed conflict? And how can accountability be guaranteed in the use of AI in war?

There’s no way to stop progress. Someone is going to do it, and you’re going to have to do it too — because if you don’t, they’ll eat you alive. So this will happen sooner or later. What’s more — people ask me, where’s the money in AI? It isn’t in the language model that teaches you to write an email. The twenty bucks a month — that’s not where the big money comes from, la guita grande, as we’d say in Argentina. It’s in the military. That’s where it’s going. There’s a company, the best known in the field today — Palantir, in the United States, which sells to the American government. Then there’s Anduril, another famous one, in military matters. That’s the direction things take. That’s the big money, and it’s going to happen — it’s already happening, I have no doubt.

Jimmy: They want to keep the final decision from being made by a person.

Mookie: I’m not sure a person is going to be better — especially the people I happen to know, the military. Once the AI gets to killing on its own, it just kills, and kills, and kills — on its own, of course. Otherwise — the speed. Imagine the speed. War — you always have to remember — has no referee, no card to show. It’s not a game. And here’s the one good thing in history: wars are going to start being fought with robots. In the end there won’t be people on the other side. That’s the good news. AI will, of course, build its own robots — it’ll build them itself — and those robots will kill each other. Fine: we send them off to some corner of the earth, let them kill each other on their own, and we live in peace. We won’t be giving the orders. They will. The AI. I told you — you’re going to be a witness. We move to Mars.

Jimmy: Could we get to a stage where it’s two robots squaring off, David and Goliath?

Mookie: But seriously — look, AI can, and this is exactly why I say China is finished. Finished as China. It has no access to sophisticated AI, because it has no access to the chips, and it never will — not because the Americans don’t want it to, but because the Chinese can’t. At some point they tell you, for instance, that China has the largest navy in the world. But a ship attacked by ten thousand drones, perfectly synchronised by an AI, lasts fifteen minutes. They take it out instantly. Either you have someone facing you with that, or you’re done. And it won’t even need to happen — it’s enough that the other side wants it to. In the end, fear is just that: believing that something is going to happen. But it doesn’t have to. Trump understands this clearly, and that’s why he decides the way he does — and yes, the world order changed.

There was an English philosopher, Thomas Hobbes. Hobbes said that people’s lives are basically miserable, and that man, left to himself, ends up killing everyone — and the only solution is to have one strong man who holds it all, so that people have security. Something like a Bukele — but from the sixteen-hundreds.

Jimmy: Easy there, Tom—

Mookie: Easy, easy — no, it slipped out. Anyway, Thomas Hobbes. We’re living Hobbes’s dream — we’re going to live it. What is Donald Trump doing? Trump has done genuine realpolitik. The way I put it: if I don’t give you two slaps, the message doesn’t land. He’s not far from saying it himself — it’s not as if I have to read too much into it. Panama Canal — I’ll take it. And that’s that.

Jimmy: While some countries use AI to improve services, others are turning it into a tool of surveillance and repression. Authoritarian governments use artificial intelligence to track dissidents, censor information, and manipulate public opinion, consolidating their power through technological control. To what point is AI strengthening authoritarian regimes, and what measures can be taken to keep it from becoming a weapon of repression?

Mookie: Let’s go back to the authoritarian regimes. The authoritarian regimes are China and Russia — the big ones — and Iran. Then there’s North Korea, but honestly the rest are small fry, the odd one here and there. Those are the authoritarian regimes. And they don’t have AI — so we can relax, for now. Let me explain how it works, because otherwise it sounds like I’m talking magic.

AI needs particular, special chips called GPUs. The GPUs are made by a company called Nvidia — a company that at one point became the largest in the world; it goes back and forth with Apple, they fight it out. Nvidia’s chips are two to four years ahead of any competitor, and of course it doesn’t sit still waiting. To make those chips, though — Nvidia designs the chips; they have to be sent out to be made, and one single company in the world makes them, only one, called TSMC, in Taiwan. That company makes ninety-two percent of the world’s AI chips — no more, no less. TSMC is probably five years ahead of any competitor, and pulling further ahead. But to make the chips you need machines, and those machines are built by one single company, in the Netherlands, called ASML. ASML makes machines that only they can make. They sold these machines to everyone who made chips — China had machines, everyone had machines. ASML is twenty to twenty-five years ahead of any competitor. Nobody is going to catch up. And, once again, ASML doesn’t stand still. These machines are insanely expensive — three hundred million dollars, five hundred million — and they’re installed in the various places where chips are made, including China. To make chips, the machine has to be serviced once every two weeks. If you go a month without servicing it, it’s scrap. Scrap. They’re a million perfectly synchronised mirrors — something genuinely hard; they work at the level of an atom, a millionth of the width of a hair, using light, which is the only way to do it.

So this triopoly — ASML in the Netherlands, TSMC in Taiwan, and Nvidia in the United States — guarantees that whoever controls the triopoly controls artificial intelligence. Biden — I’d say the best thing he did in his presidency was to squeeze and take away China’s ability to manufacture AI chips. AI chips: the smaller, the better. They’re measured in nanometers, a nanometer being a millionth of a meter. Right now they’re making three- and two-nanometer chips; those are the ones in use. Anything above seven nanometers is no good for advanced AI — fine for this little gadget that helps you write emails, but the AI of industry, the AI of the military, the AI that’s going to deliver efficiency across everything — that needs seven and below. Until now China had access only down to fourteen, and now, with the latest turn of the screw, it seems only from twenty up. The machines it has, they’ll stop servicing — and in a month they’re good for nothing, just somewhere to set a vase on top. The Chinese know this, and they’re desperate. So they’re using two paths. One is to throw a lot of money — and a lot of people — at reaching that technology themselves, which of course doesn’t work, for many reasons. I won’t get into China or I’ll never stop, because it’s a subject I know well, but it isn’t tangential to AI. The other path is to tell stories — we made it, we did it, here comes DeepSeek. DeepSeek. I’ll use a slightly rude word: DeepSeek is what it is, which is nothing. DeepSeek is marketing, to win attention. The other day they said they’d got a chip running eight hundred times faster than normal. That doesn’t exist. It’s not that it isn’t true — the technology doesn’t exist. They get carried away. And there’s a lot of internal consumption, mind you — because if the people realize they’re going to be left behind, that their economy will keep struggling, the people will rise up. In China there’s a tacit deal between the government — the Communist Party government — and the people: I give you prosperity, you don’t get involved. Prosperity is ending. Imagine what could happen.

Jimmy: Artificial intelligence is capable of generating disinformation at scale, erasing the line between truth and manipulation. This raises serious concerns about its impact on public conversation and trust in the media. If AI-created disinformation is inevitable, how can societies adapt to mitigate its impact and strengthen trust in information?

Mookie: This one’s on us. If you decided you wanted to know more, and you read this paper or that one, sooner or later they’re going to sell you a mailbox — con you — because that’s their business too, selling mailboxes, and they get paid to sell mailboxes. People know what “a mailbox” means. And this is the important part: people, first of all, don’t want to know the truth. People want to not suffer. And if, in order not to suffer, they have to believe in a fairy godmother, in a God, that they’ll be immortal — that life is whatever they want — they’ll take it, and rightly so. Mind you, this isn’t a stoic view — grit your teeth and suffer — no, no. You’ll flee suffering anyway.

Jimmy: It happens to you too. You — what is it you want?

Mookie: I’m saying you don’t want to suffer.

Jimmy: No—

Mookie: None of us wants to suffer. It’s just that my suffering doesn’t run through there. Believing in a God or a fairy godmother does nothing for me — I want to know. And look, I feed on the information that’s out there and I try to cross-check it. That’s my thing. And then sometimes I want to spread it, so I do — I write articles about it; you know I sign everything, las cosas como son, things as they are. But people don’t want to know how things are, and they’re right to. What advantage is there in knowing the truth if the truth hurts?

Jimmy: I didn’t follow that — and now you know the truth, so what, the anthem should play?

Mookie: All that epic drama, on the one side — yes. On the other, we have confirmation bias. What’s confirmation bias? The Nobel laureate in economics said it — Kahneman, an Israeli. Kahneman uncovered confirmation bias. Confirmation bias says: I’m always going to pay attention to what confirms my idea. Again — so as not to suffer. Otherwise you’d have to live with two ideas crashing into each other. What for? To feel good, not to feel bad. That escape just keeps going — and life isn’t going anywhere; it’s chasing you from behind, the whole time. You’re not heading anywhere. When you arrive, you think, ah, I wanted to get here — but the truth is you were running away. And in that flight, which we all share, I told you — neither the truth nor any objective matter [pulls you], unless that objective matter makes you feel good. It makes me feel good. But that’s just how I am, like everyone else — and it doesn’t make me superior to them. In fact I’d say it makes me inferior: I’d love to have something that solved it all for me, so I didn’t have to get up at five every morning to read.

Jimmy: Right. So we’ve reached the subject of democracy. AI systems are designed to be persuasive rather than truthful, optimizing for mass reach instead of accuracy. This has deep implications for journalism, education, and politics, where the manipulation of language can distort public discourse and alter the perception of reality. How can a society make sure AI doesn’t weaken trust in the truth and in democratic institutions?

Mookie: A bit repetitive — but you used the word education. There’s not going to be any education. It’s over, for several reasons. First, you don’t need it: all the knowledge is already here. It was always in the phone — it’s just easier now. Now you can ask it, instead of hunting around on Google. So education and schools are glorified babysitters. School is a babysitter — they park you there so the parents can go to work, which they won’t have to do either, because they won’t have to go to work. So what for? And more than that, everyone will want the kids at home.

The pandemic was a rehearsal for all this. We went off to work, the kids stayed home, the information was a disaster — because it was a disaster, remember, people said all sorts of outrageous things. And what happened? It went fairly well. We got a vaccine in a year, solved the problem, went back to where we were.

I think we need to look at a few things. Look — humanity eliminated hunger. There’s no more hunger, truly. There are five guys who are hungry; the rest of the world isn’t. Food isn’t scarce — there’s a surplus. In fact we now have an obesity problem everywhere, not only in the developed countries. I was reading the other day that Italian children are the fattest in Europe. Every animal is built to deal with scarcity, not abundance — and we’re pretty bad at abundance, like the rest of the animals. So what we’re seeing is a string of achievements, and instead of saying hey, nicely done, it’s no — look, we’re destroying the planet, killing the animals, that whole laundry list — and if that’s not enough, asteroids are coming, plus a tsunami, and who knows what else. Turn around. Stop for a minute. Look back. Pretty good, isn’t it? Fine — you and I didn’t do it, that’s the truth. But you see, it’s the people who look just like us, who also walk around with two eyes and two ears and say things — I think sometimes that’s exactly what’s missing: look how well we’ve done. Look at the things — how good they are, while everyone’s talking about how bad things are.

Jimmy: I suppose there’ll be people who unplug, go off to the countryside.

Mookie: If I’m right — that’s the thing. If I’m right, people shouldn’t listen to my theory, because listening to my theory does harm. When I say we flee suffering, think about what we’re saying: we’re saying we’re all cowards. And what does it mean to be a coward — you know it’s the one insult that exists in every language. Coward. He flees suffering. So people don’t want to know they’re fleeing suffering; they want to believe they’re heading toward the grandchildren — and they’re cowards and fools. Fools. Leave them be. But we are cowards. All right — so we can’t own our cowardice. If I accept that all I do is flee suffering, and I feel like a coward, I’m going to feel bad — so I have to not believe myself, I have to not believe what I’m saying. So I’m in a perfect spot: if they pay attention to me, I’m right. And if they don’t — I’m right.

Editorial note

Two of the speaker’s attributions are reproduced as spoken: confirmation bias is generally credited to Peter Wason rather than to Daniel Kahneman; and Thomas Hobbes (1588-1679) is a seventeenth-century figure, not a sixteenth-century one.

Recorded at Solanas Convention Center, Punta Ballena, Maldonado, Uruguay.