AI and 1980s computer games

[Disembodied]

http://www.bbc.co.uk/news/technology-31620759

Researchers working for Google in London say they have developed an artificial intelligence system that has taught itself how to win 1980s computer games.

OK, so it's only mastered Breakout so far - but can an elite AI Oolite player be that far away? And then what price online games like Elite: Dangerous? There are saddos who "play" chess online by cranking up a program and playing under its name ... what happens when half the griefers in an online game are all AIs?

[Cody]

There are saddos who "play" chess online by cranking up a program and playing under its name ...

I'd kinda gathered that, which is why I only play chess against my own computer.
As for AIs in online games - it's probably already happening, in one form or another.

[cim]

but can an elite AI Oolite player be that far away?

Depends what you mean by "AI". One that can teach itself Oolite just given an input and output device I suspect is still some way off. One that can provide a reasonable facsimile of an experienced player and only that ... most of the bits are already lying around in various OXPs or the game core.

[Disembodied]

but can an elite AI Oolite player be that far away?

Depends what you mean by "AI". One that can teach itself Oolite just given an input and output device I suspect is still some way off. One that can provide a reasonable facsimile of an experienced player and only that ... most of the bits are already lying around in various OXPs or the game core.

We've definitely got the workings of a good, human-like ship-combat AI in the game - but you're right, it would be a lot harder for an AI to work out an optimal strategy for an open-ended sandbox game. I wasn't being entirely serious, but it does raise an interesting issue for online PvP gaming: how will you know if an opposing player isn't an AI? You couldn't even Turing-test it: the AI's owner might sit in as its comms officer ...

[Wildeblood]

but can an elite AI Oolite player be that far away?

Depends what you mean by "AI". One that can teach itself Oolite just given an input and output device I suspect is still some way off. One that can provide a reasonable facsimile of an experienced player and only that ... most of the bits are already lying around in various OXPs or the game core.

We've definitely got the workings of a good, human-like ship-combat AI in the game - but you're right, it would be a lot harder for an AI to work out an optimal strategy for an open-ended sandbox game.

I disagree. I've already automated trading. Writing a script to automatically buy equipment upgrades would be comparatively easy. Automating navigation from system to system to explore, and take in likely trade opportunities as you did, would also be comparatively easy. A fully automatic Play The Game For Me 'Bot OXP would not be difficult.

[Disembodied]

I disagree. I've already automated trading. Writing a script to automatically buy equipment upgrades would be comparatively easy. Automating navigation from system to system to explore, and take in likely trade opportunities as you did, would also be comparatively easy. A fully automatic Play The Game For Me 'Bot OXP would not be difficult.

Yes, you could do that, if you gave the AI the advance objective of "make as much money as you can" and told it how to do it. It might be trickier though to tell it to "become Elite as quickly as you can".

Here, though, we're talking about an expert system, or specific AI: we tell it what to do and how to do it, in advance. The key factor about the game-playing AI referenced in the link above is that all it was given was the objective of getting as high a score as possible: nobody told it what to do to get this. It had to watch the screen and figure out what inputs produced what outputs. This is general AI, or at least a step towards it.

[Disembodied]

There's an update on the article here:

http://www.bbc.co.uk/news/science-environment-31623427

It's tangential (at best) to the issue at hand, but I have to say I was surprised to learn that there was an Atari game called "Montezuma's Revenge" - I'd always thought that was just another name for the holiday squits. Every day's a school day ...

[Cody]

I'd always thought that was just another name for the holiday squits.

Curiously, I happened to be looking at this just the other day, and the Wiki states:

Montezuma's Revenge is a colloquialism for traveler's diarrhea in visitors to Mexico. The urban legend states that Montezuma II initiated the onslaught of diarrhea on "gringo" travelers to Mexico in retribution for the slaughter and subsequent enslavement of the Aztec people by Hernán Cortés on Aug 13, 1521.

In India, it used to be called Delhi Belly.

[Astrobe]

There are saddos who "play" chess online by cranking up a program and playing under its name ...

I'd kinda gathered that, which is why I only play chess against my own computer.
As for AIs in online games - it's probably already happening, in one form or another.

Yep, and online poker sites are taking the issue VERY seriously.

[Cody]

Yep, and online poker sites are taking the issue VERY seriously.

<nods> I used to play a bit of poker, back in the day - but even if I was 100% certain that the game was kosher, I'd still never play online! If there's serious money on the table, you gotta see the other players' eyes/faces! Not only that, but they play some strange poker variant - cobblers to all that!

[ClymAngus]

You want to see an AI explode? Give the SOAB manic miner, if it wins that one give it Jet Set Willy. 80's game winner indeed. Those games put steel in the spine of older players. A trait I'm seeing sadly lacking these days.

[mossfoot]

There are saddos who "play" chess online by cranking up a program and playing under its name ...

I'd kinda gathered that, which is why I only play chess against my own computer.

At least then you can do this: