King's Gambit refuted (almost)

[Lazy_Shiftless_Basta]

This seems like an oddly good place for this command stream interception Starcraft AI video. Warning, may cause epileptic seizures!

[AaronSofaer]

@FieryBalrog

They have not, however, come anywhere close to developing an AI that actually plays competitive Starcraft, or done much beyond some basic build order analysis.

You’re joking, right? Read up on the Berkeley Overmind. Even before many, many improvements, it was beating pro-level Korean players. As it stands, it’s easily the best Starcraft I “player” in the world.

Or are you talking just about Starcrat II?

[Wolff]

The ability to simultask instead of mere multitasking I think would give AIs a huge advantage in RTS. Time to execute moves is severely hampered by the interface device for humans. The only games IMO that humans stand a chance are ones that reward creative strategies.

[SuperAdventurer]

So apparently the King’s Gambit lives on. ChessBase got me good:

The ChessBase April Fool's prank

[Robert_Sharp]

I don’t know why anyone reads ANY internet/news stories anywhere NEAR April 1st.

A holiday dedicated to lying is such a crap idea. But maybe that’s the ethics professor in me speaking.

[Ultrazen]

@Miramon

No one claimed it was solved. I was responding to a suggestion that chess might not be winnable by humans in 20 years; IMO it’s not winnable by humans now.

There are 2 ways to beat computers at chess, at least that I’ve found.

1. Sacrifice games. Most computers will always take a piece that has a higher point score. You can use that to manipulate position, which is 95% of chess. Really top end programs won’t fall for this, but things like phone apps etc will.

2. Trade even pieces as often as you can, as long as your position remains in tact. Reduce the game to the least amount of pieces. Computers have a huge advantage when it comes to tabulating multiple scenarios, if you reduce the variables it’s much easier to at least play to a draw.

[GloriousMess]

@Ultrazen

There are 3 ways to beat computers at chess, at least that I’ve found.

1. Sacrifice games. Most computers will always take a piece that has a higher point score. You can use that to manipulate position, which is 95% of chess. Really top end programs won’t fall for this, but things like phone apps etc will.

2. Trade even pieces as often as you can, as long as your position remains in tact. Reduce the game to the least amount of pieces. Computers have a huge advantage when it comes to tabulating multiple scenarios, if you reduce the variables it’s much easier to at least play to a draw.

3. Unplug the computer, re-arrange the pieces into checkmate, plug it back in.

[anymunym]

@Ultrazen

2. Trade even pieces as often as you can, as long as your position remains in tact. Reduce the game to the least amount of pieces. Computers have a huge advantage when it comes to tabulating multiple scenarios, if you reduce the variables it’s much easier to at least play to a draw.

Really? I would have assumed the opposite to be true, i.e. that the computer’s advantage became bigger the fewer pieces there were left on the board as fewer variables would allow it to look further ahead.

[vyshka]

Heh, Chessbase’s April Fool’s joke strikes again.

The ChessBase April Fool's prank

[FieryBalrog]

@AaronSofaer

You’re joking, right? Read up on the Berkeley Overmind. Even before many, many improvements, it was beating pro-level Korean players. As it stands, it’s easily the best Starcraft I “player” in the world.

Or are you talking just about Starcrat II?

Perhaps you can show me. As far as I know this is pure unadulterated bullshit, especially this bolded line.

The program won an AI competition. A competition amongst other AIs. It couldn’t even beat a 3rd rate former Starcraft player at the same time. Its “optimal build order” is based on operating in a vacuum. I have absolutely no idea how you are confidently proclaiming it to be the “best SC1 player in the world”. The best at winning a muta micro competition, uh, OK.

[AaronSofaer]

@FieryBalrog

Perhaps you can show me. As far as I know this is pure unadulterated bullshit, especially this bolded line.

The program won an AI competition. A competition amongst other AIs. It couldn’t even beat a 3rd rate former Starcraft player at the same time. Its “optimal build order” is based on operating in a vacuum. I have absolutely no idea how you are confidently proclaiming it to be the “best SC1 player in the world”. The best at winning a muta micro competition, uh, OK.

Are you talking about the two players the AI did, in fact, beat, one of whom was a respectable ICCup player?

And yes, it won an AI competition. In 2010, with an outdated build that they had already significantly improved upon. It micros, it macros, it maintains map presence, scouts, harasses, does gas-steal shenanigans, it’s far more than a “muta micro” bot.

The version they released mopped the floor with a couple of Korean pros. The versions they didn’t release - because they really didn’t give a shit after the competition, it’s not like they’re in it for the money - were apparently improved in every respect, but they didn’t want to put in the time and effort to QA it.

So in sum, how bout you show me a video of someone beating even the build they released for the AIDE competition?