MIT Teaches Computer to Read, Conquer the Planet

 Pages 1 2 3 NEXT
 

MIT Teaches Computer to Read, Conquer the Planet

image

Researchers have taken humanity one giant leap closer to robotic Armageddon by teaching a computer how to read, understand and very effectively apply the manual to the strategy classic Civilization.

Sure, we all like to joke about the looming machine apocalypse, but when I found out about how researchers at MIT taught a computer to read - and worse, to apply the knowledge it gained from said reading in a simulation about conquering and quite possibly blowing up the entire world - well, let's just say I started to think that maybe it's not all that funny after all.

Regina Barzilay, associate professor of computer science and electrical engineering at MIT, along with her graduate student S.R.K. Branavan and David Silver of University College London, presented a report at this year's meeting of the Association for Computational Linguistics about teaching a computer to "read" through a program in which it learned how to play the PC strategy game Civilization. To play it alarmingly well, in fact.

"Games are used as a test bed for artificial-intelligence techniques simply because of their complexity," said Branavan, who was first author on this paper as well as one from 2009 based on the simpler task of PC software installation. "Every action that you take in the game doesn't have a predetermined outcome, because the game or the opponent can randomly react to what you do. So you need a technique that can handle very complex scenarios that react in potentially random ways."

Game manuals, Barzilay added, are ideal for such experiments because they explain how to play but not how to win. "They just give you very general advice and suggestions, and you have to figure out a lot of other things on your own," she said.

But the truly amazing-slash-frightening part of the whole thing is the fact that the computer began with a very limited amount of information - the actions it could take, like right or left-clicking, information displayed on the screen and a measure of success or failure - and no prior knowledge of the what it's supposed to do, or even what language the manual was written in. Because of that blank-slate beginning, its initial play style was nearly random, but it gained knowledge as it progressed by comparing words on the screen with words in the manual and searching surrounding text for associated words, slowly figuring out what they meant and which actions led to positive results.

The augmented Civ-machine ended up winning 79 percent of the games it played, compared to a winning rate of only 46 percent for a computer that didn't have access to the written instructions. Some members of the ACL audience apparently criticized the report, saying the system performed so well because it was put up against relatively weak computer opponents, but according to Brown University Professor of Computer Science Eugene Charniak, that argument misses the point. "Who cares?" he said. "The important point is that this was able to extract useful information from the manual, and that's what we care about."

It's pretty heady stuff, with a more down-to-Earth benefit for gamers being the promise of far more sophisticated computerized opponents in videogames. Instead of the relatively exploitable preset routines we have today, we could in the relatively near future find ourselves squaring off against computerized opponents with the ability to actually learn, adapt and come at us with ever-evolving tactics and strategies. But the long-term prospects may not be so sunny. If that thing ever figures out how to play Alpha Centauri, we are screwed.

Source: MIT News, via Edge

Permalink

The fact it always beelines to nukes and immediately bombs all other civilizations out of existence is nothing to worry about.

When a Goomba gives me the "come at me bro" I'll know I'm in the future.

Yay! Now I can be beaten by an AI in Civilization!

Wait...

Maybe despairs to strong a word...

Maybe it will make game ai better?

I would like to see what happens if they put two of these computers against each other.

What i want to understand is how a computer with no idea how to play civilization can STILL win 46% of the time.

I know chieftan is easy, but what the hell

Thomas Guy:
I would like to see what happens if they put two of these computers against each other.

isnt it obvious? "a curious game the only winning move is not to play"

Thomas Guy:
I would like to see what happens if they put two of these computers against each other.

They'd have to make one Russia and the other United States...
Professor Oak: So what are your observations from this game session

Dr. Johnson: well both of them took over 50% of the world, then Epsilon created the Manhattan Project where then both of them created massive stockpiles of nuclear weapons, then relayed a message to each other saying, 'I wish these were real' and 'Rodger Rodger' before destroying each other and making every tile in the world un-inhabitable.

Professor Oak: Dear God what have we done!? WE'RE GOING TO WIN THE NOBEL PRIZE!

OT: well better get back to building that rocket ship to the closest habitable planet...

People don't realise that computer won't have desires, to put it in a more approachable way, computer will only want to take over the world if the programmers tell them to, but people are much less expensive and much easier to produce.

Going up against AI opponents in strategy games that can actually learn and adapt sounds pretty awesome. Just please don't put that kind of AI on to any kind of defense system.

Next test: To see how well it does against competent human opponents.

They came, they saw, they conquered.

In other news, researchers at MIT have recently announced that, "We are not robots. We are feeble humans."

Oh good, now let's let them read an anatomy book to know where our vital organs are.

Now, once someone teaches it how to play Dwarf Fortress, the real fun will begin...
Kudos for the awesome title, btw!

I, for one, welcome our new computer overlords.

...Or it's something like that.

Anyone who's played Sid Meier's Alpha Centauri will know where this is eventually headed to.

We better research Mind-Machine Interface

I for one celebrate the coming of our new robotic overlords. Praise be to Ghandi the Conqueror, the dread mechanical ruler of Earth.

edit:

CkretAznMan:
I, for one, welcome our new computer overlords.

...Or it's something like that.

Damn! ninja'd

CkretAznMan:
I, for one, welcome our new computer overlords.

...Or it's something like that.

Man, I knew someone would say that line.

LET IT DIE, I HATE THAT LINE. RAGE.

SCREW YOU KENT, YOU MEME STARTING BASTARD.

image

Is MIT somehow affiliated with Skynet..massive potential for future developments

Did the computer look anything like this?

image

A neat application of very well-known associative learning mechanisms in AI and CL. It's cool and makes for a good news story, but isn't really particularly groundbreaking.

Andy Chalk:
But the long-term prospects may not be so sunny. If that thing ever figures out how to play Alpha Centauri, we are screwed.

Funny. But no computer could ever figure out how to play... OH SHIT!

I think it's disturbingly impressive this thing was able to start from a blank slate and learn at all, much less to read. But do we really want AI opponents who regularly become better than we are? That sounds like a fast train to frustration.

Thomas Guy:
I would like to see what happens if they put two of these computers against each other.

screw that! Teach it Starcraft 2 and have it go against top tier Koreans, that'll be a match to remember!

Wait, it started on a blank slate with no information to go on, but had read the game manual........ok? A Darwinian program being used through trial and error to create a thinking ai? Color me not shocked as I've been trying to say this forever and I'm glad to see it finally being implemented. I'm more curious what level AI it was playing against.

It wouldn't make any sense for a computer to want to dominate humanity, it isn't at all logical.
Now, if you somehow offended it and it could activate the sprinklers... well, you won't be feeling warm and dry for the rest of the day.

-Dragmire-:

Thomas Guy:
I would like to see what happens if they put two of these computers against each other.

screw that! Teach it Starcraft 2 and have it go against top tier Koreans, that'll be a match to remember!

I would pay very large amounts of money to have that set up, but don't tell the human opponent. Watch as they grow more frustrated as the computer adapts and counters every strategy. Then tell them they just loss to an AI. It would crush them I thhink.

-Dragmire-:

Thomas Guy:
I would like to see what happens if they put two of these computers against each other.

screw that! Teach it Starcraft 2 and have it go against top tier Koreans, that'll be a match to remember!

I was thinking exactly the same thing. That would actually be pretty awesome. I think MIT should connect it to the Internet and let people play civ and Starcraft with it.

EDIT: I've just envisioned a world where you come home to find your computer playing GTA or reading a magazine. Imagine it's little robot arms coming out of the monitor flipping the pages!

That's kind of cool actually. Maybe you could connect it with other devices (Phone, console, etc) and it could be your best friend and had a personality and stuff like that. That would be awesome. *looks back at what was just written* ...I shouldn't be posting at 20 to midnight.

ryo02:

Thomas Guy:
I would like to see what happens if they put two of these computers against each other.

isnt it obvious? "a curious game the only winning move is not to play"

Actually in Civ4, it's to build Nukes as fast as possible, and then blow everyone else to kingdome come before global warming sets in. In Civ5, it's the same, except there's no global warming to worry about.

Yes, the day a machine is able to win a match of Starcraft 2 against the worlds best player... That will be like when IBM beat that chess player. It'll be marvelous. My god.

Cpt Corallis:

-Dragmire-:

Thomas Guy:
I would like to see what happens if they put two of these computers against each other.

screw that! Teach it Starcraft 2 and have it go against top tier Koreans, that'll be a match to remember!

I would pay very large amounts of money to have that set up, but don't tell the human opponent. Watch as they grow more frustrated as the computer adapts and counters every strategy. Then tell them they just loss to an AI. It would crush them I think.

heh, reading that gave me the image of the former champ slumped in a chair staring blankly at his hands as a single tear rolls slowly down his cheek.

hmmm... that could be a wonderfully terrible movie premise where he had to work hard to reclaim his title as the SC2 champ. Naturally, it would have an 80's style montage in it.

EDIT:

Secret world leader (shhh):
*looks back at what was just written* ...I shouldn't be posting at 20 to midnight.

I've done several of those myself. Sleep deprivation and forum posting usually don't mix very well.

Well, I think it is a fantastic way to advance robot...I mean human kind. *BZZZT*. Ahem.

Reminds me of the developers comments of stars! 3 (a 4x space strategy game that regrettably never saw the light of day, so there's no real way to know if it was true or just hyperbole)

Supposedly they created AI for it that could learn how to play by observing human players...

They ran into a problem with it though, because one thing it observed and copied, was the tendency human players have of going 'save', and then when something bad happens... 'reload'.

Yes, this AI had figured out how to reload an earlier save-game when something went wrong. XD

Needless to say they had to rewrite the AI routines so it couldn't actually access the save/load mechanisms, but it shows what an AI can learn if you let it...

Thomas Guy:
I would like to see what happens if they put two of these computers against each other.

Haven't you seen Wargames?

 Pages 1 2 3 NEXT

Reply to Thread

Log in or Register to Comment
Have an account? Login below:
With Facebook:Login With Facebook
or
Username:  
Password:  
  
Not registered? To sign up for an account with The Escapist:
Register With Facebook
Register With Facebook
or
Register for a free account here