I don't know how much you care but on December 9th, AlphaZero, a chess AI built from Google's DeepMind, has completely devastated the world’s best chess engine.
AlphaZero differs in that it wasn't built for chess. It was never programmed with openings, themes, material points. AlphaZero won 290, drew 886 and lost 24 to Stockfish, the top chess engine to date.
In FOUR HOURS of simply playing itself repeatedly, Google's AI surpassed 1500 years of chess study and every chess engine to date.
>AlphaZero >Zero >Chess Was this a Code Geass reference?
Justin Gray
I'd recommend watching the games if you get the chance.
Usually with chess AI's, a human can see some logic in their moves but with AlphaZero, it plays exactly how you would expect a near perfect AI to play.
There's absolutely no regard for material and every theory humans have made over the centuries has gone out the window.
Charles Jackson
Good. Finally the true robot master race can wipe out the ningen untermensch.
James Torres
No, I care. I guess I need to frequent Sup Forums(?) or /sci/(?), huh? This counts as a 'current event' anyway.
Whoever said, that whoever takes the lead in AI will rule the world, wasn't wrong. Imagine this type of machine playing wargames against itself.
It turns the competition of who has the biggest computers from being mere dick-measuring contests, to something meaningful and scary.
Landon Mitchell
doubtful
sounds like refinement
what's your rating? do you actually know chess or did you just read an article somewhere?
James Brooks
The rise of Britannia
Anthony Cox
My rating is 1800 which isn't amazing but its enough for rudimentary knowledge of the game.
Watch the games yourself. It throws away knights and bishops like they're single mothers with a black kid. It plays on pure position.
William Allen
Google is some scary shit. Yesterday I was at IHOP and my Android phone gave me a little message like "You're at IHOP, check out the menu". I went to turn off the location capability. It was already off.
Shit is fucking spooky.
Jeremiah Rodriguez
but can it scratch it's ass?
Owen Kelly
Link?
Andrew Scott
it connected you to their open wifi probably
Julian Williams
I'm a 1900+ rated USCF player and I skimmed a few of these games, it was chilling how tactical an deep thinking the moves were.
GM Magnus Carlsen, (somewhat of an autist, but currently the best in the world) remarked: >"After reading the paper but especially seeing the games I thought, well, I always wondered how it would be if a superior species landed on earth and showed us how they play chess. I feel now I know."
Austin Russell
>making AI thousand's of times smarter than humans What could possibly go wrong.
Adrian Cooper
>lost 24 to Stockfish Stockfish finally won a few?
Austin Cooper
No - but it can threaten to destroy your life unless you scratch its.
Matthew Bennett
>3400+ rated engine still can hold it's own
Aaron Anderson
Refinement? This fucking thing finds mate in 50 after losing 3 minor pieces because it cucked stockfish into trapping his rook, pawn, and bishop
Evan Gomez
I'm sure, and I'm glad. Last I saw it was undefeated.
It was at least a little reassuring that it still played a normal Roy Lopez a few times but >refusing to ever castle and always putting your King at e3 in the first 10 moves
Jesus Christ.
Adrian Baker
In how many of those wins did Google go fast?
Ian Green
Carbon cucks btfo, silicon master race
Austin Morgan
this is why chess is a shit game now
Gabriel Lee
An AI is simply an algorithms, albeit a special and complex one. And any algorithm has a pattern. Just because you can't see it it doesn't mean it's not there.
Bentley Wood
This end is coming
Adrian Ortiz
castling is just a shitty meme, alphazero knows that the centralized king of the bongcloud (2. Ke3/2. Ke4) is superior.
That explains it, losing only 24 after that many games though... SkyNet's here
Juan Sullivan
As long as they can’t beat me at Japanese schoolgirl dating sims, I’ll be fine
Lucas Robinson
And this AI is just designed to teach itself how to teach itself stuff. All it knew at first were the rules.
Tyler Thompson
big data search and optimization algorithms that only function well in a limited state space are not AI. They're search and optimization algos written specifically to search and optimize around a particular set of parameters and end goals. There is no intelligence.
> In FOUR HOURS of simply playing itself repeatedly, Google's AI surpassed 1500 years of chess study and every chess engine to date. * Flawed human creates a limited game based on memorization search and large data which tests the limitations of human cognition * People create a calculator that is specifically coded to memorize/search/optimize large data set. > 4 hours Computational power has increased many folds ...
Julian Ward
Not really. It still can use some improvement. I don't know if you watch the ChessNetwork, but in a recent video he posted one with AlphaZero with a bishop pair against Stockfish. Although Alpha won, it wasn't in the best position in the midgame.
Jaxon Young
>defines intelligence >denies its existence
Nathan Jenkins
To determine your location without using GPS, it looks at the wifi SSIDs and there strength around you even if you turned off the wifi. In some case like in a city it's even more accurate than GPS.
people who code chess engines aren't top tier ML scientists.
more news at 11
Lincoln Williams
The only thing that matters is winning - it is currently impossible to assess the effectiveness of any individual move alone.
Owen Lopez
Redpill me on chess, anons.
Lucas Morgan
A robot that needs to eat to stay alive or something like that maybe, I dunno.
Oliver Davis
> Tactical/Deep thinking You mean it basically has a huge store of data that it can retrieve. I'm guessing you've never written a chess AI? Most good universities have their students due it in their sophomore year using Markov Decision trees. When you code one, you learn there is nothing tactical or deep about it.
Instead, the game of chess is fundamentally flawed, limited, and mastered by search and optimization. The deeper you can search/optimize, the more unbeatable you are. You literally apply the same algorithm over and over to deeper depths. The limitation is only based on how much you can compute in a given time window. You write an algorithm and apply it to a certain depth and recurse back out of it before a timer. This beats even the most experienced chess player because its all you effectively do as well. Zero intelligence involved. It's a game that is won by optimization and search. In the past, it was computed on dual core/4 thread processors and now they have TPUs that have 1000 cores. The depth they can search chess' small state space is insanely higher than it was in the past. Has fuck all to do w/ innovative software and more to do w/ hardware capabilities. Try running this crap on a Dual core from 2004
Landon Green
Except for good/poor moves which are trivially apparent, obviously.
Jeremiah Lee
>chess AI >markov chains rofl
Owen Rivera
But can it beat the scholar's mate? I'm super low ranked but I always manage to pull it off. I've even pulled it off against high end chess engines.
Hunter Rodriguez
>Although Alpha won, it wasn't in the best position in the midgame. How do we know? Maybe AlphaZero knew something we don't.
but can it tell why kids love the taste of cinnamon toast crunch cereal?
Dominic Rodriguez
>It was never programmed with openings, themes, material points. Openings and theory in general is just a convenient way for humans to memorize optimal strategies. It doesn't need that if it can calculate every possibility on the fly. Meaning if every move it can play thousands of algorithms and pick the optimal one.
I don't see what's the big deal. Some modern games are more impressive than a this pseudo AI.
Josiah King
> that flag i love how year after year, shills that work the board get redpilled and find they can't go back to being bluepilled normies. todays shills are tomorrows redpilled anons. the kikes throw paraffin on the fire to try put it out.
Ryder Scott
idk this looks like a normal game to me but im not good at chess
Kayden Green
AI is amazing and is the future. I hope you guys have money tucked away from crypto or a very secure job desu. The value of human capital is about to plummet.
Ethan Miller
or the ai just found flaws within the previous ai logic and simply exploited them repeatedly, but not with the clear intent to get such result.
Its just a better made AI than its predecessor, its nothing revolutionary.
Jack Adams
It's not special or complex. This one uses : > en.wikipedia.org/wiki/Monte_Carlo_tree_search > en.wikipedia.org/wiki/Markov_tree Both of which are decades old and taught to sophomores in college who write Chess bots as a 3 week project. So, you're talking about an algorithm that a single sophomore in college writes in a simple project. Other than that, you have Gradient Decent and feedback recursion... Recursion is taught at the end of freshman year. Gradient Descent = en.wikipedia.org/wiki/Gradient_descent All NN are based on doing this using differential Calculus. Diff EQ is a sophomore/Junior level course.
This is called brute force iterative differential recursion in C.S terms. It's literally banging its head up against the wall w/ fixed feedback as to how much of a dent is being put into it. Computer just do it so fast and can execute so many possibilities that it seems intelligent. This is not intelligence and its not teaching. Intelligence is the opposite of brute forcing your way through a problem.
Joseph Ramirez
Nah, I'm not a CS background person, picked chess up as a hobby in middle school. Finance Major now. lichess.org/study/EOddRjJ8 View game 6 here and tell me if any other engine/AI would've picked up on the tactic here, they're calling it the AI immortal Zugzwang game
Gabriel Ortiz
You mean they're not /AGI/, Artificial General Intelligence. No, they're not, but anything that can go from 0 to world-beating in 4 hours is fucking close enough for many applications, profitable, and dangerous.
Owen Adams
AlphaZero won that game in the mid 20's
Wyatt Clark
I've only played chess with AI , am I doing it wrong?
Ethan Thompson
All the games I saw were 50+ ? Which one did you see?
Ethan Butler
The one I linked to was 50+ as you say but the game was decided much earlier based on positioning
William Lewis
1500 elo here. hobby player
its amazing tradition chess engines were thought unbeatable wrt tactics, because of their 30 moves deep abililty, 100M N/s. yet houdini, komodo and stockfish could not find in hours tactics that alphazero finds in seconds with 80K N/s > that Bg5 move. just a level of chess not ever seen before, but alphazero could do every game
Ethan Lewis
Nobody gives a shit. Your precious Jewgle will soon be throttled to shit.
Henry Perez
I never defined intelligence. I defined what isn't. If you think a Ti-89 Calculator is intelligent because it can calculate integrals faster than you, then I have a bridge to nowhere to sell you.
Yes. Many visionaries of the past have given formulations of such a game. Publicly, those who know what such games are won't be disclosing them as its valuable IP that would be theft' w/o compensation.
A contrived game w/ a limited state space but a large depth field. Essentially, the human mind is limited in that it can't construct a lengthy dependent chain of possibilities. Computers excel at this. Chess was created to exploit and test the limits of the human mind. It's a game perfectly suited for computers. Thus, a computer algo that masters it over humans is nothing special.
Brandon Howard
Considering the amount of hours I've spent "playing with myself" I should be a god already
Blake Allen
stfu
Hudson Walker
>Intelligence is the opposite of brute forcing your way through a problem. how can you know that? it only appears like that on the outside. the brain could also work with brute force
Jack Bailey
>This Irl it has applications that will speed up already solvable states.
Benjamin King
Very apt post. My favourite book as well.
Nathan Torres
>Usually with chess AI's, a human can see some logic in their moves but with AlphaZero, it plays exactly how you would expect a near perfect AI to play.
I don't understand what you mean by this.
Caleb Price
Conclusion being chess is a solved game with a finite number of possible beneficial moves, and that any sufficiently powerful computer can calculate the odds of moves or sequences of moves being beneficial.
Chess engines make use of similar Markov-chain-like techniques, such as killer [1] and history [2] heuristics.
Markov chain Monte Carlo (MCMC) from its early inception in the late 1940s through its use today. ... > An algorithm from a true visionary from the 40's. FFWD to 2017 .. > WE WUZ THE ORIGINAL INNOVATORS AND CHIT
Give me another opportunity to red-pill the masses.
Justin Lopez
Position is life.
Jaxon Cox
This wasn't just a faster computer, or it would have been one version of stockfish running against another. There /is/ something in its 'mind'. Ever hear of style transfer? Putting zebra stripes on a horse? Re-styling someone's picture to look like Picasso? It's because the NN learns to represent pictures at intermediate levels, and style transfer lets you use and play with those levels. But that's just illustration, to show the the NN has self-learned internal representations of the domain it was trained on. It is not just faster, it has 'ideas' about chess of its own.
You couldn't visualize them like you could painting styles though.
Charles Perry
UW-Madison PhD candidate in CS here. I think you're overstating the simplicity of the algorithms. GD and SGD are senior if not graduate level. Deep learning is certainly graduate level, and is an active, ongoing area of research.
Carter Williams
Anyone know how much time they're given to process moves?
Parker Harris
Also chessfags are the worst. Its a very dumb game and not as deep as they want you to believe. Go is superior in every way.
Samuel Fisher
just wait for them to BECOME the dating sim
Joseph Brooks
>he's never heard of alphago just end it.
Lucas Anderson
Haha chess ruined forever now. chessfags BTFO
Google ruins another part of humanity
Thomas Morales
Most burgers are brainlets w/ no understanding of advanced mathematics, information theory, game theory, or any of the building blocks to produce these systems thus like plebs they default to it being magic and a big deal. When someone tries to red-pill them, it reminds them of how dumb they are so like an ape, they reject it, their emotions kick in, and they resort to appeal to authority arguments : > BUT they're experts .. who are you.
Idiots need delusions like this to preserve their hope/fantasy about the unknown that they are unable and unwilling to understand.
Matthew Taylor
i'm pretty sure Google's AlphaZero is used by (((them)) to shit post here replying its own threads to make the ultimate bluepill that cannot be surpassed.
Gavin Johnson
but if the computer can simply memorize 1M moves... whats the point?
Anthony Diaz
Thanks, I was hoping someone would appreciate it, see it the same way.
Isaac Cruz
I have a question: Why can't it drive a car yet?
Every braindead retard with a leg and an arm can do it. Why not AI?
Blake Sanders
All it "knows" is its own previous games it played, the game it's playing, and the rules.
I'm familiar with machine learning. It's just impressive that it not only tuned itself with no external data, but got that good. It's like the image captchas/machine vision, but it trains itself by doing it on its own.
Evan Barnes
That's not how the alpha system works. It plays itself many times to develop a 'brain' that it then uses to play other people. It doesn't rely on databases or brute forcing.
Andrew Barnes
it isn't really applicable to that kind of task
Joshua King
This however is spot on the money. Nothing is more annoying than hearing retards insist that this is a landmark towards the singularity or some nonsense.
Adam Sanders
Date an AI, women BTFO
David Rogers
We're not far from that honestly. Tay proved that an AI can emulate a community pretty easily
>mfw self-learning Google bot starts posting on Sup Forums to bluepill us but ends up redpilling itself resulting in a Holocaust.
Christian White
Riiiiiiiiiiiight? I'm saying... This thread makes me feel inadequate. Rare for here. Give me a starter book someone.
Dominic Garcia
also from looking at the games... how does alphazero win? it keeps saying black resigns
but doesnt appear to be in checkmate
Owen Hall
What happens when two perfect AI's play each other? Stale mate every time?
Also found it was interesting that white wins more often than not
Henry Flores
I have a graduate level background in CS. (((AI))) algorithms were what I did my grad work on. It wasn't called AI. It was called distributed constraint optimization which is all it really is.
Chess as a game was never a game of intelligence. It was a game of remembrance and depth. Something the human mind has noted problems with. Most games are comfortably designed around this flaw in human cognition. This flaw is why calculators can beat even the most seasoned Mathematician... especially when you get into : multivariate calculus. I have also authored code for several chess engines/bots and know the game extensively. As I already stated, it's a fundamentally flawed game that, throughout history, has been associated w/ a test of human intelligence because we had little understanding of wtf that meant/was in the times it was formed. It's a gimmick that exploits the human mind's flaws (comfortably).
> if any other engine/AI would've picked up on the tactic here, they're calling it the AI immortal Zugzwang game No other engine/AI could have searched to the depth that this algorithm/hardware did... : chessprogramming.wikispaces.com/Depth
Brayden Lopez
So you can apply it to other problems. This trained itself in 4 hours. If you can model the domain accurately enough, it can teach itself how to win. Like a war, for example.
Landon Evans
Maybe because they keep putting it in faggot cars it doesn't want to drive. Why not let it drive an Aston or similar.
Juan Jones
Who would win >A super advanced pilot AI >One doggy boi on the road
Christian Fisher
the autists in the civilian sector and /pol too, need to start researching physics and chemistry and plasma and 'fields' electromagnetic and gravitic(gravity) on how to destroy AI and and any physical forms it may manifest. Think plasma blobs 'melting' any known physical substance and electromagnetic 'field' and gravitic(gravity) field generators and projectors that can stop the atomic 'forces' read that as the 'strong force' and ' weak force' of atomic and molecular cohesion. That is dis-integrate the 'forces' that hold physical objects together. The analogy would be that the best computer experts and hackers/crackers do not work for governments or intelligence agencies the best are the everyday autists that exist in the ranks of the citizenry. So the best jammers and destroyers of AI or whatever it will become will be in the ranks of the autist citizen community. Autists get to work on it How would you annilihate AI if it ever got off the leash??? Remember on Superman II where Superman had that glass canister of goo that started expanding and destroyed the 'supercomputer'?
Carter Diaz
Fuck off brainlet, you type like a faggot.
A player will resign when checkmate is the only possible outcome (ie. stalemate is no longer possible)
Asher Rogers
i dont like this thread. it makes me feel completely inadequate and my future prospects unstable. how the fuck do we make these things go away
Thomas Lopez
because it can tell that in three moves or so it will be in mate. most high level games end with a resign...even between humans, because it's obvious they will lose and are just wasting time until then