OK, but Lee is an extreme outlier in this regard. His is a sad story indeed, because the game lost one of the greatest players ever in his prime. But the vast majority of pros adjusted to the initial shock of Alphago and now quite happily use it as a tool for improving their play. - ThreadSky

siegeweather.bsky.social • 235 days ago

OK, but Lee is an extreme outlier in this regard. His is a sad story indeed, because the game lost one of the greatest players ever in his prime. But the vast majority of pros adjusted to the initial shock of Alphago and now quite happily use it as a tool for improving their play.

Reposted from Paul Roquet

Buried amidst the platitudes, a bleak follow-up to this story. Career over, creativity destroyed, and can’t even enjoy the game anymore

www.nytimes.com/2024/07/10/w...

Comments

siegeweather.bsky.social•235 days ago

And unlike the amateurish dreck AI developers want to replace art/artists with, Alphago immediately + unambiguously outperformed human players. In fact, as Lee himself admitted, it played beautifully and elegantly. But this was only possible due to the rule-bound, zero-sum nature of the game.

timlubin.bsky.social•63 days ago

I find it quite encouraging that he won one of the four games. Perhaps he lost mainly because the machine did not play like a human opponent, and he did not have time to adjust his strategy.

siegeweather.bsky.social•63 days ago

Absolutely. It’s unclear to me whether Lee understood how AlphaGo’s decision trees worked going into that match. If he knew that playing a tactically sound but puzzling move (Match 4, Move 78) would take it so far out of game prep that it would “freak out,” then he did actually adjust his strategy.

siegeweather.bsky.social•63 days ago

He is on record as saying that it was the “only possible move.” It’s not the only move—even an amateur player can see that—but if that’s really the way he saw it in the moment then it is just further proof he is one of the greatest players ever. Nobody else would have seen it that way.

khakhalin.bsky.social•235 days ago

> But this was only possible due to the rule-bound, zero-sum nature of the game.
This, sure, but also because the algorithms were customer, clever, self-improving etc. etc. It's AI as we knew it before the name was culturally appropriated!

siegeweather.bsky.social•235 days ago

Interesting point. I admit I only understand the go-side of AlphaGo. I take it you are saying that it had the capacity to improve in ways that weren’t a direct result of simply playing more games and remembering which variations led to weak positions and which ones led to strong positions?

cmbnugent.bsky.social•235 days ago

Isn't it a back-propagation neural net? If so, it's really only going to learn by playing more and more games and learning which moves work in which situations (very different from Deep Blue, which wasn't a neural net). Playing itself speeds that along, but it's the same process.

cmbnugent.bsky.social•235 days ago

(Though it’s almost 30 years since I worked with neural nets so what do I know?!?)

khakhalin.bsky.social•235 days ago

I would argue that once you self play its not the same process. But also they used a very smart lookahead decision tree. It was a feat of engineering, really cool and nifty!

khakhalin.bsky.social•235 days ago

It played with itself from the get go; no external data at all. As as games are very long, turns are very local, and exact positions almost never repeat, it was really tricky to link the effect of one little turn with hundreds of possibilities to winning or losing the game

khakhalin.bsky.social•235 days ago

That's where the smart engineering went.

But also I just love when beautiful things emerge from simple rules, like ice on a window, like clouds, like flowers :)

khakhalin.bsky.social•235 days ago

(I'm not even sure it is a zero sum game, now that I think of it :)) maybe nominally, but not emotionally, philosophically!)

Comments

Posting Rules

Reply