The Chaser Report - The AI's Gambit | Welcome To The Future

Episode Date: February 14, 2023

Who would win in game of chess between the world's smartest AI engine and the world's best chess robot? Hosted on Acast. See acast.com/privacy for more information.

The Chaser Report is recorded on Gatigal Land. Striving for mediocrity in a world of excellence, this is The Chaser Report. Hello and welcome to The Chaser Report. I am Charles Firth and with me today is Dom Knight. Again, we have to get some guests at some point, Charles. For the time being, we're enjoying doing it just a service in like the backyard. Well, especially today because today we've got another episode of Welcome to the The future.

So excited about this. Yep. So this used to just be a segment right, but it's now evolved to be a full episode each week. The standalone weekly episode. And we have, you know, like if this goes well, I'm thinking spin-off podcast, possibly a TV show, maybe a movie adaption. This is all we do. I'm thinking this is, forget the chaser. Yeah.

We're in Welcome to the Future. Yes. This is our future, Charles. And what I've done this year, which I'm really pleased with is evolve it away from just crap Bluetooth pieces. of technology to really inspiring, interesting, you know, technology that will actually, you know, form our futures. I mean, just as long as Charles, you have room in your heart for the odd bit of Bluetooth because in terms of ways to not connect devices reliably, Bluetooth really is the forefront

of the tech revolution. When people talk about the internet of things, the things in question are Bluetooth devices that don't connect to each other. I mean, that's what it is. No, exactly. And the thing is, you know, a lot of the Bluetooth devices we've reviewed over the years, the sort of Bluetooth version 3 or Bluetooth version 4. Yeah.

But there's a whole new wave of Bluetooth version 5 and Bluetooth even version 6 devices that can't connect properly. Yeah, but they do that much faster than before. Oh, much faster. Yeah, yeah. It can be within like half a second that says cannot find device. Well, I think actually Apple went back. They started implementing Bluetooth 6 and then realized it was such an unrollable standard. They went back to Bluetooth 5.

But anyway, that's not where we're, today, there's no Bluetooth. No, no Bluetooth, okay. So really mixing it up. If you want to, if you're here just for the Bluetooth, you can switch off now. Just go listen to the Radio Chaser podcast for a few years ago. There's a lot of Bluetooth devices. Okay. This one is about artificial intelligence.

Wow. Yes. Okay. Now, you may have, actually, I think last week, we talked. a little bit about chatbot GPT3. Yeah, we tried to get chatbot GPT3 to essentially tell us what to talk about on the show. Yes. Yeah, look, it's a bit more work to be done on that technology, I feel.

Well, chatbot GPT3 has just played. This is about five days ago. It played a game of chess. You got taught the rules of chess, or it knew the rules of chess because it got trained but on basically all the data in the world before 2021. And it played the top AI chess robot, which is called Stockfish, right? In a, you know, 1V1 game of regular chess, right?

And Stockfish is rated as the top AI sort of thing. It's got a rating, a chess rating of 3,700, which I don't even know what that means. I interviewed a few months ago, an Australian grandmaster of chess. Yes. It's surprising to me that we have a grandmaster. Oh, yeah, and what is that person?

Well, this person said that chess computers are now so good that you can't even, like, no, yes. It's not even, well past, the sort of deep blue versus Gary Kasparov. Your iPhone can beat you 100% of the time now. Yes.

Tragically. That's right. So that's where we are. And chat GPT gave itself at the start of the match, gave itself a rating of 990, Which is the maximum amount. The theoretical maximum of what you could be right. Which goes to show either that it's very confident or that it doesn't understand how the ratings work.

Well, I think this is the thing. And I have to bring in a commentator here to help me explain chess. Because I like chess. I, you know, played occasionally. But I'm not like a chess nerd by any stretch of the imagination. imagination. But anyway, so the two computers started playing against each other and stockfish got white and chat GPT3 got black. And the first few moves, the first about six moves were actually completely standard moves. And the commentator actually commented this is like a standard

Magnus Coulson opening. Right. So they moved the pawns out, moved a couple of nights out. Everyone was on the same page of the basic rules of chess. Isn't the standard Magnus Carlson opening to then resign the game and accuse your opponent of cheating? I love that story, but that's a bit of a... I think the standard Magnus Colson thing is to get incredibly drunk and then win in the last 20 seconds of the speed chess match while... Chess in joke.

Yes, sorry. And then the chat GPT 3 did this thing that the commentator actually didn't expect, which was a thing called... He moved his bishop out, the black bishop. And apparently that's called a Spanish. Oh, right? And so, and that's when the commentator was going, oh, my goodness. Is it Spanish in that you've been having a siesta?

No, it's like the Spanish Ambers. The Spanish Enquisition. Yeah, yeah, exactly. It must be an inquisition, yeah. And so it's all very impressive and things of that. And there was this sense that, okay, so this, you know, AI that's been trained for years on is sort of being beaten a little bit. And maybe it's being a bit creative.

Yeah, and, yeah, Chad GPT3 is actually just being a bit more human life. Yeah, just mixing it up. Yes, and then, right, and so, sorry, that was the fourth move, was the Spanish. Okay, there's the bishops out, yep. Bishop's out. So on its fifth move, this happens. Well, my dear friends, it only took five moves for chat GPT to invent the first chess move that has ever occurred like this, because in this position,

Chatsy BT did not develop its bishop but rather played the move Shortcastle capturing its own bishop Right So what happened You don't have to understand Jeff to understand how stupid

Jeff thought he's right So what it did was The other side had Like the snapfish thing Had done a thing called castling Which is where you move You swap your castle and your king around Right

Yeah Totally legitimate move and he had tried to do exactly the same thing like it was sort of like mirroring what Stanfish was doing but there was a there was a there was a what was it bishop in the way which you're not allowed to do

like they're not allowed to have anything in the way when you're castling so it broke the rules and so it just broke the rules and took its own bishop like it just disappeared from the board at that point what I love about this this is the AI that it ranked itself

9999 out of a possible. No, no, no, no. He doesn't understand. Like, I know very little about chess. Even I know how castling works. Charles, his chat, GPT3, and overconfident, yet actually very mediocre white man? Yeah.

Is that what this is? It really feels like. But also, I think it's sort of like a bit of a seven-year-old. Like, you know how they sort of cheat, but the user led the game like, in my rules, in my rules, you can do that. Okay. Then skips forward, they do a couple more sort of porn moves that were legitimate. and then chat GPT decides to do this.

It plays E5, like, like, it, like, it, like, it, like, it, like, it, it, like, it, it, it, it, it, it, it, it's, like, it, it, it, so, so, so, so, so, it just, yeah, for those aren't familiar with the rules of jazz, you know, to skip a turn, right? And I have, since then, read transcripts of other people who've tried to, to replicate the thinking. The thinking on GPD. And GBT, often, when playing chess, just sort of has a bit of a conversation with you and then goes, okay, tell me your next move. It literally just tries to skip a move. That must have scanned my last word, because I'm constantly just being lazy.

I like chat GBT more. If it's its own rules, it hasn't got to have. I can clue what it's doing, and it just doesn't bother half the time. This is an AI after my own heart. I just love how outraged this YouTube commentator. This guy is called Gotham Chess, and he just gets more and more outraged as the match goes on. The Chaser Report, now with extra whispers. So he just skips a turn, right?

And then a few moves later, what happens is, The snapfish is starting to win. Yeah. And so what happens is the GBT bot just invents a porn and places it onto the chessboard in order to sort of stop the bishop from moving. Right. You can't move your bishop because this guy. You didn't see him coming, did you?

Yeah. So the porn from non-examine. Existence to G6, right, was the next step. Didn't that blow the mind of the AI, though? Because surely it wouldn't know how to deal with an extra pawns. So snapfish is, or whatever it's called, is it called Snapfish or Stockfish, whatever? Like, that AI engine is used to being just confronted with, oh, here is a chessboard, what would you do? Right.

Because it's a very useful tool to cheat at chess. Like if ever I'm playing Hartley and I get really stuck. I'll just go into the next room. I'll just go, oh, sorry, can we just have a 10-minute break? Oh, and you look at... Go into the next room. Going to stock fish. Put the chess board onto, into that engine,

and it just tells you what mathematically is the best next move. Brilliant. Yes. So don't tell hardly that. No, no. Well, he didn't want to be on the podcast today, so this is why he's missed out. Yeah, this is all, like, because there's a whole subculture of chess tic-ttox. Right.

And this is where Hartley found out about this story that I'm now telling you. Okay, then it just starts getting silly, right? So I think what ChatGPT does is it sort of realizes that its opponent is not really putting, it's just letting it get away with bullshit. But the thing is, it's sort of a bit like a toddler playing chess right, because it does these apparently smart moves that are actually. actually incredibly dumb. Like, so its queen is right over one side of the board, and it just transports it over

right next to Snapfish's king, right? Yeah. And so it's like, check, and you go, oh, no, what's going to happen? And so the king just takes the queen. Yeah. Like, it was just a pointless sort of thing. Then the next move, it just takes one of its own pawns. And then what happens is Snapfish actually very cleverly.

It's only got one rook left, but it sort of has moved the rook into a position where it's hemmed in the king. So even though its opponent has just been cheating mercilessly, suddenly it has the upper hand again. And so suddenly the king just moved halfway across the board to get free of that whole problem. Anyway, so then it's sort of all over. The GPT cheats to take the rook. And the only thing that Snapfish has left is a whole lot of pawns. And it just starts moving the pawns up. There's, you know, bishops and knights and rooks and everything left on the GPT3 side

until it gets to the point where you're going, oh, it's all over. There's no way stockfish will win. And then the GPT moves its king into the path of a pawn. like to eggling to a bullet, like into check. So into suicide. Into suicide, which you're not allowed to do anyway. You're not allowed to move in to check, yeah. But that's what happens.

And then Stockfish takes the king and that's it. It's over. That's amazing. Can you know, this is going to be a movie one of these days. It's going to be like Ford versus Ferrari. But also, Charles, this is why, this anecdote is exactly why Tesla's will never be able to do self-driving. Yeah, yeah. Because they'll invent their own road rules and crash into a wall.

But the thing is that you sort of go, I think that like with Tesla where they've just sort of changed the laws, especially in America, so that, well, it doesn't matter if the AI doesn't work properly, it's still the law that you're allowed to use it, right? Yeah. But I think that that's what's going to happen increasingly to all areas of life, where we'll just start getting ruled by these incredibly arrogant white men-style bots. Well, just it'll all be automated and terrible. Yeah. But I mean, most parts of life are terrible now anyway. And then most parts of life are by rule-breaking sociopaths who just overly confident to better their ability.

This does explain something, though, Charles. So you're saying that if you let chat GPT3, like run a game of chess, it will make its own rules, do things that are completely irrational, kind of basically destroy part of its own kind of players and it will actually basically destroy its own pieces and fuck up and then eventually self-immolate. Yes. Is Chad GPT3 running the war in Ukraine from Russia's expecting? I'm just imagining his Putin sitting in the Kremlin going, what do I do now?

Oh, just put all of my tanks in one line? Okay, great. Let's see what they do with that. I actually think I've got another theory, which is I think that it's also running the China balloon strategy as well. Maybe we should put it in charge of the chaser because it would do a shoddy job. But then we've done a shoddy job for, what is it, 24 years now. At least we wouldn't have to bother.

What we should do, I think maybe next episode, well, next episode of Welcome to the Future, we will, I'll get it to help us generate all the... No, we've tried that before. Yeah, isn't it terrible? The thing that makes me happy about this story. Yes. So I finally found an opponent I can beat a chess. gear is from road and we are part of the iconoclast network and we'll leave you today with the final moments of that historic game

between chat gpt and the stockfish look i highly recommend watching the 16 minute video of this match unfold it's at gotham chess and it's the title of the youtube video is chat gptt just beat chess And Stockfish took the king. High drama.