A game that computers cannot win over humans.

[kthejung]

http://www.wired.com/2014/05/the-world-of-computer-go/

The human race still reigns supreme over machines in the game of Go.  Aside from originality, I didn't think that humans could beat machines in a thinking game until I read this article.

[kuroman]

there are many things that computers cannot wins in, anything unrelated to or cannot have a simple decomposition into millions of straight calculation, for instance a computer cannot deal with sarcasm or cynism, or any in between the line meaning, while our brains can easly deal with that, there are new alghorithms and computer technologies as proven by Watson or with the newest google implementation but we are still far far away from having a computer than can match the human brain in such areas, but yes when it comes to brute calculations our brain are unmatched, but again our brains has the ability to find solutions to make those calculation much easier and quicker each time something a computer cannot do and each time such solution are implemented to computers

[Malin Keshar]

turing test game(computer wins if he convince human he is a human)

guessing game(like the one between frodo and golum on The Hobbit)


turn games(not real time games) related with long term strategy, in equal conditions.  I see that AI has hard time with long term strategy

[bluefirecorp]

Given enough time, a computer will eventually win. Consider quantum computing for example -- in one calculation, it'll be able to calculate ALL possible moves. Just a single calculation. Of course, once we hit quantum computing, so much will happen so fast.

[r34tr783tr78]

I wasn't expecting that a computer would win jeopardy. It's a very hard competition, so everything is possible:
http://en.wikipedia.org/wiki/Watson_(computer)

[Hazir]

I still think that Deep Blue win against Kasparov was not entirely legit. They were reprogramming it on the fly during chess matches and they gave it database with ALL known chess parties played before on competitive level. It was not just a calculation game.

[r34tr783tr78]

Perhaps, but now no one can beat the best chess player computers. Deep Blue is history. Present supercomputers can beat any human at chess.
Probably, because chess is a simple game, with simple rules.
But not even Watson can compare with science fiction computers, like HAL of 2001. We are still very far from self-aware computers.

[Malin Keshar]

AI of most turn based games sucks and is easily defeat-able by humans.
Same with most bots from FPS



is it because they invest about no time making their AI?

[Benjig]

And i tought that my mobile that has an android on it could think

[Acidyo]

Well as time goes by and technology develops, we won't be able to defeat the machines.

SKYNet is now active.

[Phinnaeus Gage]

It appears that computers are better than humans at rock, paper, scissors:

http://www.essentially.net/rsp/index.jsp

The computer has an overall win/loss ratio of 1.49.

But, Sheldon is undefeated when Lizard and Spock was added to the lot.

[Kluge]

AI of most turn based games sucks and is easily defeat-able by humans.
Same with most bots from FPS



is it because they invest about no time making their AI?

FPS AI is mostly about impairing the AI after coding, not trying to have the AI kick ass (it's made to mimic a brain-dead player). Most games have you competing against 5-20 AI units at a time, so you'd lose the badass feeling devs strive for if AI were competent. If FPS AI was difficult, multiplayer FPS services wouldn't spend so much in resources trying to get aimbots off their servers.

RTS AI is waaaay more complicated because it has to think about the future of its actions, what actions it can take mid-term, then long-term, and see what he knows about the human's strategy to counter-act his plays. Those lead to the "branching computing crawl" - for every action it considers, there are probably 2,000-20,000 moves it can make after that, then another 2,000-20,000 after that, and after, and after, and after, until it has its "best final build strategy" all figured out. Because of this, it usually only has a few preconfigured strategies it randomly selects and then makes minor adjustments based on what it sees the human and other AI doing. A dev company with loads of resources, however, may take the time to introduce a bunch of off-the-wall sub-strategies to AI's arsenal. Humans are very much the same, as far as how they think, but learn to beat the AI by having a larger set of "preconfigured strategies" -- basically, humans are much more random. AFAIK, there aren't really self-learning games, where it'll do something like have the AI watch a match it lost, then try a bunch of different strategies and sub-strategies to win, then save that strategy if it sees the player later trying to pull the same shit. I'll say that custom AI for Starcraft 2 released soon after its release was pretty difficult to beat.

[kthejung]

AI of most turn based games sucks and is easily defeat-able by humans.
Same with most bots from FPS



is it because they invest about no time making their AI?

FPS AI is mostly about impairing the AI after coding, not trying to have the AI kick ass (it's made to mimic a brain-dead player). Most games have you competing against 5-20 AI units at a time, so you'd lose the badass feeling devs strive for if AI were competent. If FPS AI was difficult, multiplayer FPS services wouldn't spend so much in resources trying to get aimbots off their servers.

RTS AI is waaaay more complicated because it has to think about the future of its actions, what actions it can take mid-term, then long-term, and see what he knows about the human's strategy to counter-act his plays. Those lead to the "branching computing crawl" - for every action it considers, there are probably 2,000-20,000 moves it can make after that, then another 2,000-20,000 after that, and after, and after, and after, until it has its "best final build strategy" all figured out. Because of this, it usually only has a few preconfigured strategies it randomly selects and then makes minor adjustments based on what it sees the human and other AI doing. A dev company with loads of resources, however, may take the time to introduce a bunch of off-the-wall sub-strategies to AI's arsenal. Humans are very much the same, as far as how they think, but learn to beat the AI by having a larger set of "preconfigured strategies" -- basically, humans are much more random. AFAIK, there aren't really self-learning games, where it'll do something like have the AI watch a match it lost, then try a bunch of different strategies and sub-strategies to win, then save that strategy if it sees the player later trying to pull the same shit. I'll say that custom AI for Starcraft 2 released soon after its release was pretty difficult to beat.

I'm surprised that a lot of people do not seem to know that AI in games are intentionally dumb down to not be impossible, or even hard, to beat.  I'm sure that even the best human FPS player would lose to a computer that was designed to win.  

[Foxpup]

I'm surprised that a lot of people do not seem to know that AI in games are intentionally dumb down to not be impossible, or even hard, to beat.  

Of course, the purpose of AI in games is usually not to beat the player, but to simulate a human opponent, making it a sort of Turing test. If the human players don't realise the AI isn't playing to win, it passes.