k4ir0s   Canada. Mar 12 2016 07:19. Posts 3460
Anyone following the alphaGo challenge? next match tomorrow 2-0 for the AI
it's really fascinating what they're accomplishing. 1 year ago they solved Atari games using self-learning AI with 'sight' being it's only input, and now GO. The founder even mentions in an interview that they might try Starcraft in the future. They recently announced that their next challenge is a 3D game..
Yeah bot is insane. I need to catch up with game 3 though, will update comment for more.
It is cool because there is finally AI that can solve this problem better than human can with the same time constraints (real game clock). There are times you see alphago just sat there and thought for a long time before making a move.
Go is a deeper version of chess where good players (9-dan pros) can (and need to be able to) walk through the game tree about 50 moves beyond. Seems like alphago is thinking deeper than human now (maybe alphago can traverse hundreds of levels deep in the decision tree now), hence beating human.
Just checked score, alphago is 3-0 now. Jeez
no pain no gain
Last edit: 12/03/2016 11:11
okyougosu   Russian Federation. Mar 12 2016 17:19. Posts 962
If they can make AI to dominate Go top pro 3-0, crushing starcraft players would be super easy after modifying this AI thing for Starcraft games.
Just look at how does programmed script control zerg army:
TimDawg   United States. Mar 12 2016 18:00. Posts 10194
I find this more scary than interesting tbh lol
I'm pretty sure I heard they are going to try to make an AI for SC1 actually
online bob is actually a pretty smart person, not at all like the creepy fucker that sits in the sofa telling me he does nasty shit to me when im asleep - pinball
yea i was following it as well and wondering a bit about the low viewer numbers and stuff.
never really had contact with go before but this american go pro is doing a good job explaining stuff. im not super surprised that a machine can calculate deeper and faster the trees than a human but its still awesome to see how they did it and how its succeeding because of AI leaning but at the same side I agree with TimDawg that its a bit intimitading.
Im not sure how this will develop in the future (maybe like this I-robot movie with smith and shit) when machines learn how to learn on their own and enslave us all but hopefully that is still a bit away but I fear that its actually all not that far away anymore.
okyougosu   Russian Federation. Mar 13 2016 09:01. Posts 962
No worries, the definition of self-learning for bots like alphaGO is to exclude situations where it needs to calculate deep into "bad" tree nodes, by remembering and cutting those at self-learning phase. Most likely some more advanced game theory stuff added to limit the amount of future calculations, like in chess programming. Anyways, the "learning" part is heavily pre-programmed by human being and is way too far behind from human learning ability.
okyougosu   Russian Federation. Mar 13 2016 09:09. Posts 962
Once a computer bot win a match by killing his human opponent by a high-voltage discharge, that would be first generation evil robot
Daut   United States. Mar 13 2016 10:05. Posts 8919
the end is near
NewbSaibot: 18 TIMES THE SPEED OF LIGHT. Because FUCK YOU, Daut
vasoline73   United States. Mar 13 2016 10:11. Posts 791
I watched game 3 and 4 live.
Game 3 was depressing but good on team human to win game 4! 1-3 bby. The end is less near than initially assumed.
These matches have definitely made me want to try to learn Go. Super interesting stuff with commentary.