MIT Teaches Computer to Read, Conquer the Planet
- Thread starter Andy Chalk
- Start date
Here more than half the human population barely or doesn't know how to read, and we are worried about teaching a computer how to read? This scares me more than anything resembling Skynet.
But on the other hand, I'd be curious if the 79% constituted just conquering the world or was a mash of different win scenarios between world conquer and reaching Alpha Centauri, and I think the UN is the third, but it has been so long since I played the original. If it was a mashup, which of the three dominated?
Details!
(I'm not the only one curious, there is a related comment on the MIT site as well)
But on the other hand, I'd be curious if the 79% constituted just conquering the world or was a mash of different win scenarios between world conquer and reaching Alpha Centauri, and I think the UN is the third, but it has been so long since I played the original. If it was a mashup, which of the three dominated?
Details!
(I'm not the only one curious, there is a related comment on the MIT site as well)
That sort of AI requires more memory than your dinky little PC can hold. Look up what the kids at MIT get to play with, and then give your PC counseling for the inferiority complex it suddenly develops.Blaster395 said:
No one's asking the important question:
HOW did it win it's Civ games? If it leaned towards cultural or diplomatic victories, what are we fussing about?
Domination, or science, on the other hand? Kill it! Kill it with FIrE!!!
HOW did it win it's Civ games? If it leaned towards cultural or diplomatic victories, what are we fussing about?
Domination, or science, on the other hand? Kill it! Kill it with FIrE!!!
Uh, all that matters here is the fact the computer 'learnt' (to an extent) by itself.
The fact that it applied its ability to this particular video-game means nothing. It could have done the same with any other scenario involving winning and losing.
The fact that it applied its ability to this particular video-game means nothing. It could have done the same with any other scenario involving winning and losing.
It doesn't matter.GrizzlerBorno said:
The computer doesn't understand what science, diplomacy, culture and violence mean in real world terms. It understands them as methods to achieve a goal. It will pick the most optimal solution.
You could rename all of them and it wouldn't change anything.
What we should be focusing on is the fact the computer learnt, rather than what it learnt about.
It is actually pretty cool that they are able to give the computer some form of written strategy and it can interpret it.
I, for one would like to see better AI in games, maybe developers would stop trying to put multiplayer and uninteresting gimmicks into single player games - and even multiplayer focused games would benefit from better AI.
And you can easily limit the AI with difficulty settings and don't have to be afraid that it gets too good if it has the ability to learn and adapt.
I, for one would like to see better AI in games, maybe developers would stop trying to put multiplayer and uninteresting gimmicks into single player games - and even multiplayer focused games would benefit from better AI.
And you can easily limit the AI with difficulty settings and don't have to be afraid that it gets too good if it has the ability to learn and adapt.
True enough. But then again, it could be argued that it doesn't understand the concepts of Art, diplomacy.etc. yet, because it isn't technically sentient?thick doona said:
Maybe it will when it reaches sentience. Who knows?
Regardless, Civilization could work as a pretty effective acid test I think....
Well dont we already have adapting AI opponents in games? i guess this could be a step forward, but its still a long wy till skynet. Computers dont have feelings so they reply on logic. logic dictates that humanity is a plague to earth. thus taking over the world would be LOGICAL decision for computer.
This is entirely true. The end states of the game where the "player" wins are all equal in the strictest terms. When people play such a game, they determine a strategy based upon some combination of what they think will work, what they have at their disposal, and what they prefer. A computer on the other hand can make that decision based entirely on relatively pragmatic concerns of which victory condition is it best able to move towards. In terms of grand strategy, this is literally nothing new. The only interesting bit was that the AI was required to learn all of this rather than relying upon various deterministic routines. The learning is the key, not that it won but rather that it initially had no idea how to play the game beyond the most basic mechanical notions.thick doona said:
if it can learn to play correctly it would be unbeatable, and eventually it would-Dragmire- said:
imagine a computer BUILDING MORE PYLONS err.. i mean micromanaging troops range so your guys are never facing and in range, ever.. it could be done and that's just sickening
if a computer figured out how to micromanage shields and tactical units on several fronts ( that the player cant even see on one screen ) it would stomp them pro or not.
honestly i think this kind of research should be banned for it's potential abuse as a botting program in competitive gaming
it's clever, but it endangers all the fair-play we've come to expect from on-line play when it is stolen and leaked to a torrent site
i guess people would sus your 7000 apm pretty quickly though.
Well if it can only react to situations we'll be ok, it's when it finally decides to work out every possibility and what decisions led to them and what might occur from the results of decisions and can therefore work out what your doing from your opening move that we are so very very f****d.
Great, now there is an AI out there that is more aggresive than Ganhdi.
Very impressive that a computer can read a manual and then apply what it reads in the game, will it be doing online games any time soon?
Very impressive that a computer can read a manual and then apply what it reads in the game, will it be doing online games any time soon?
As long it doesn't do that by 2050, humanity can still win via a time victory we have built more wonders, which will help us get a better scoreNot G. Ivingname said:
Hawk of Battle said:
GrizzlerBorno said:
From the thread on the MIT News post (http://web.mit.edu/newsoffice/2011/language-from-games-0712.html) one of the researchers responded to that question with this:
"Our algorithm is inherently biased towards finding the fastest way to win the game. Against the built-in AI, this happens to be world domination via an "early rush" strategy - so that is what the method ends up learning."
It would be awesome if they could put this AI up against other versions of itself, and go through a process of "natural selection" to create the ultimate AI to play against the best human players.
