Slashdot Mirror
AI Can Beat Humans Only One Game At a Time (axios.com)
An anonymous reader shares a report: Despite all of the potential for artificial intelligence to solve our most vexing problems, it's still in a primitive state, according to a new report by Stanford University. But a separate paper, this one by Alphabet's DeepMind, suggests again that it has made some of its best progress in the narrow realm of games. Why it matters: Those advances are important, but life isn't a game. AI progress outside of these areas has been harder to define and track. "The most important thing for AI is to go from exceptional promise to use in actual everyday life," Martial Hebert, director of the Robotics Institute at Carnegie Mellon University, tells Axios.
Not nearly enough on sexbot technology.
let's play Global Thermonuclear War!
... desperately hoping for A.I. that will do it for them so they never have to. This won't end well.
Just sayin'.
"Not to mention all the idiots who use words like boxen."
Anonymous Coward on Monday August 04, @06:49PM
Certain genres, at least, are more like "real life" than e.g. image classification and other tasks commonly approached using machine learning. A program that can play Starcraft, or GTA, or Grand Turismo, or Counterstrike using the same inputs and outputs as a human player is a lot closer to programs for real-world tasks like driving a car, than a program that identifies handwritten characters or plays Go.
.:Semper Absurda:.
Yes indeed it is. You can't take an AI, of any form, raise it from a blob of generic neural nets, teach it english, then teach it to manipulate it's own robot arm, show it the chess rules wikipedia page, and have it play a game. Even if such an AI existed it might very well be a shitty chess player.
Current chess AI are very much tailored to playing chess, where the inputs and outputs are simple.
Someone will probably accuse me of moving the goalposts, demanding an ever better standard for what constitutes "real" AI. I personally have always maintained that human-level or better intelligence would satisfy my definition.
>life isn't a game
It can certainly be treated as one; the goal is to have the longest uninterrupted chemical reaction (I'm at around 4 billion years, personally). You can narrow that down to a 'minigame' where the goal is for an organism to successfully replicate (I'm losing there, since I've only managed replacement at 2.0 children and generally you want some redundancy just to be sure). And that game can also be divided into a number of mini-minigames.
A game is a contest with rules, goals, and a scoring system. In chess it's to checkmate your opponent and avoid being checkmated using a variety of pieces that move in certain ways on a limited checkered surface. In life it's a bit more complicated, but that doesn't mean treating it like a game is a flawed strategy.
But AI's got what plants cra...oh wait, wrong thread.
vs. Stockfish?
Why do I think Stockfish was running on a 4 CPU box and AlphaZero running on something a few orders of magnitude greater?
except as a platonic ideal in scifi-land. In real life, it doesn't for some very fundamental reasons. But most journalists are writers, so they see 'AI' and think Asimov.
I'm not. I'm a rugged individualist.
--
roman_mir
There is no AI
Intelligence is insight. The ability to apply one experience or idea to a different and unrelated concept. "the eureka" moment, the creative spark, curiosity... .
Today's computers do only task programming. A system is programmed to perform a specific task with predetermined parameters. I submit that using today's technology, AI is not even possible. Look at the hardware layer. A "massively parallel" system has what, a few dozen cores limited by bus of connectivity.
A brain, even an lower mammal has billions interconnected neurons, each capable of independent firing and the inter-connectivity is dynamic as it changes over time, experiences and developing needs and/or skills.
Till the medium of computer development changes, AI is over-hyped task programming.
A smart machine will first consider which is more worth its while: to perform the given task or, instead, to figure some way out of it.
Stanislaw Lem, The Futurological Congress (1971).
USB, USB, USB!
"AI Can Beat Humans Only One Game At a Time" headline straight after a story about AI beating various other human-beating AI's in several different games.
Yes. Several different games, played one at a time. The only thing that "Does not compute" is your reading comprehension.
>life isn't a game
It can certainly be treated as one;
No way. If life can be treated as one, then where's my wall hacks and OP cheat codes? And don't tell me it's still loading or it bugged out with the stupid memory leaks.
/joke
So you can play chess, go, and bunch of other games simultaneously against multiple grandmasters and win all of those games?
It's only a matter of time (and not much time at all, I expect) before someone decides to train an AI on a range of board games together. Nothing very difficult about that, nothing fundamentally different, just a larger input space. It will be interesting to see how the different "algorithms" will end up sharing certain neural pathways between games while others will be specific to certain games.
And even more exciting will be to see how much time it takes to learn a new game after having been trained on other games. I can totally see this experiment happen any day now.
+1. Unfortunately life is not a limited set of possible conditions and rules. Most things aren't. Games are, which is why computers are good at them.
So you can play chess, go, and bunch of other games simultaneously against multiple grandmasters and win all of those games?
No, I can't. Neither can the AI, which was the original point.