Latest AI Can Play Games With Or Without You
	found on BBC News written by snwodttam, edited by John (Plastic) [ read unedited ] posted Fri 6 Sep 10:29am
		"The front lines of Artificial Intelligence application is not in the development of Hal-like super computers; it's in your internet search engines and in your games. Making a more reliable and human-like AI for games is one of the major reasons for releasing newer versions of old games," snwodttam writes. "And Majorem, an Israel based game maker, might have made the next step in game based AI. The company has developed an artificial intelligence system that learns a gamer's style of play and can take over and play for them if they have to spend time away from the game. Adi Gaash, co-founder and chief executive of Majorem, explains, "The AI is based on genetic algorithms and can learn a player's style." The algorithm infers and tests basic strategies, gleaned from a history of user strategies, eventually deciding upon a single action. Needless to say, this offers up some interesting possibilities for every type of computer gamer: AIs, such as the one described in the article, that can continue to play the game while you are away; more challenging AIs with which you can match wits; or even more manageable AIs that support your own playing (Whada ya mean you changed production from the Human Genome Wonder to an Infantry Man?!?). More interestingly, though, does this 'genetic algorithm' indicate a step further along the path to 'I'm sorry, Dave, I can't let you do that?'"

[ more plastic... ]    

show almost all: 18 commentsbest: 5 comments by oldest first (ignore threads) newest first highest scores first oldest first newest first (ignore threads)

1.  Old News   by wrestler       at Fri 6 Sep 11:21am score of 3.5 funny     I suspect MAYORBOB uses one of these AIs to post stories and comments on Plastic. No real person could have that much time on their hands. Or maybe its the Anonymous Idiot that uses these AIs, as is often reflected in the quality of his posts. At least, that's what wrestler might have said if he had the time.   [ ...reply just to this | comment on the story... | next new ]

3.  Re: Old News   by wrestler       at Fri 6 Sep 11:46am score of 1.5 informative   in reply to comment 1     In case anyone thought I was kidding...   [ ...reply just to this | comment on the story... | next new ]

2.  The Web and AI   by Philosawyer       at Fri 6 Sep 11:24am score of 1 interesting     Just thought anyone interested in this article would probably find today's story in the Washington Post about Artificial Intelligence developed by web searching programs of interest as well: a www.washingtonpost.com link The marketing division of the Sirius Cybernetics Corporation defines a robot as "Your plastic pal who's fun to be with."   [ ...reply just to this | comment on the story... | next new ]

4.  Further developments:   by Huntred       at Fri 6 Sep 12:22pm score of 3 funny     2005: Your computer can text your cell phone and tell me what a great time it's having while you're stuck at work. Millions of slacker friends put "out of work" in a whole new dimension. 2009: You rarely buy the game. You just wave your PDA containing an aspect of your primary AI over the box and it tells you how much you would have liked it. LCD screens on the game box are linked to back to the developers systems so they can adjust and customize the box display according to your known preferences. (Special "Girlfriend near!" button stops adjusting breast size on Tomb Raider VII box) 2011: Duke Nukem released. 2015: Human game developers stop coding as their in-house AI knows how the developers code games to the point where the AI is capable of developing games on it's own with consuming far less Mountain Dew. Game publishers let your AI talk to their AI to be told how cool the game would be if they actually published them. Your AI pesters you via email, pager, cell phone, PDA and other means until you cough up some money to assist in the development of the game so your AI can play it all day. Children are now rendered obsolete. 4 Aug, 2017: Skynet Funding Bill finally passes. Skynet comes up with cool idea for ultra-massive multiplayer live-action role playing game. The rest will be history. Huntred   [ ...reply just to this | comment on the story... | next new ]

5.  This game already exists   by shadarr       at Fri 6 Sep 1:01pm score of 2 informative     It's called Progress Quest.   [ ...reply just to this | comment on the story... | next new ]

13.  Re: This game already exists   by mrphrtq       at Mon 9 Sep 6:25am score of 1   in reply to comment 5     I thought it was called OGC. Same thing, really.   [ ...reply just to this | comment on the story... | next new ]

6.  a bit extreme   by DEMachina       at Fri 6 Sep 1:20pm score of 1     I have to say, doesn't it sort of defeat the purpose of playing a game to have the game do it for you? It seems to me to be the epitome of laziness; rather than engaging in passive entertainment, we'll have things which can do it for us. Granted I can see it with someting like, say, The Sims , but it still strikes me as rather unnecessary. It could, however, give one the interesting option of playing a game against oneself, which would be really neat. "Thou shalt never make a machine in the image of a human mind." --Orange Catholic Bible, Dune Q: What do you think of Western civilization? Gandhi: I think it would be a good idea.   [ ...reply just to this | comment on the story... | next new ]

14.  Re: a bit extreme   by OSULugan       at Mon 9 Sep 9:28am score of 1   in reply to comment 6     I have to say, doesn't it sort of defeat the purpose of playing a game to have the game do it for you? I think that there are quite a large number of uses for technology like this. First, in many strategy games, as the game progresses, so does the level of required Micro-management, if the game can figure out how you tend to do things, and do it well, then your micromanaging tasks drop out, and allow you to enjoy the game, instead of feel like your at work. Also, it opens up the potential to "share" your AI personality with your friends, or across the Internet. Imagine your buddy calling you up and telling you about how you just trashed his ass in a game of Medeival: Total War, before you even came home from work. Better yet, he sends you a replay of the battle he fought, and you can unequivically agree with the computer's moves, or possibly even see a strategy of your used in a situation you might not have thought about, but learn its effectiveness. Plus, in certain games that have a tactical battle element underneath a campaign battle element, you can incorporate various people's AIs in as your Generals or Admirals, and allow them to do battle for you without blindly trusting whatever AI the computer chooses to use. I've also got to wonder if a tool like this, implemented properly, would be an aid to allow game developers to include a large multitude of AI personalities, by "teaching" the game off of historical battles fought by military geniuses like Caesar, Washington, Patton, Rommel, Zhukov, etc. rather than trying to decipher and code the great AIs themselves. Perhaps the value comes not from letting the computer take over for you, but in enriching a computer's AI by learning from you or your friends, and providing the much more difficult online/multi-player experience in a single-player environment. And God says, "No, that's not right." Yeah. Well. Whatever. You can't teach God anything.   [ ...reply just to this | comment on the story... | next new ]   17.  Re: a bit extreme   by adamrice       at Sat 14 Sep 12:50pm score of 1   in reply to comment 14     Also, it opens up the potential to "share" your AI personality with your friends, or across the Internet. Better yet, sell your AI. People already sell off their Evercrack characters, booty, etc. Why not auction off the opportunity to play just like a winner? if irony were made of strawberries, we'd all be drinking a lot of smoothies right now.   [ ...reply just to this | comment on the story... | next new ]

7.  A bit off-topic   by rantor       at Fri 6 Sep 2:01pm score of 2 informative     It's slightly off-topic, but I think interesting: My brother knew some guys in college (this would have been the era of NES or at most Genesis) who would set up the football game on their console for Computer vs. Computer. Then they would watch it. In fact, sometimes a decent number of guys would watch it. In fact, sometimes they would bet on the outcome. It's sad when you realize that the future is already here and it isn't very promising. At the final shot he won the war, after losing every battle.   [ ...reply just to this | comment on the story... | next new ]

8.  Isn't that just the way   by megamoze       at Fri 6 Sep 3:30pm score of 1     Now all of the geeks who spend their hours in front of the Playstation after rejection from girls and schoolmates, are even being snubbed by their game consoles. If a geek can't rely on the Mario Bros for love and support, then what else is there? This is the most depressing thing I've ever heard.   [ ...reply just to this | comment on the story... | next new ]

9.  Still no "intelligence" here   by Mad Ogger       at Fri 6 Sep 4:18pm score of 2 intriguing     More interestingly, though, does this 'genetic algorithm' indicate a step further along the path to 'I'm sorry, Dave, I can't let you do that?'" Short answer: no. Commenting on AI techniques, some guy named Denker said "Neural networks are the second best way to do anything." Russell and Norvig, authors of an AI textbook, go on to say "Genetic algorithms are the third best way to do anything." The good thing about neural networks and genetic algorithms is that they can be used for a wide range of applications. The bad thing is that their performance is somewhat limited. I believe that neural networks are easier to analyze mathematically and tend to learn faster than genetic algorithms. This kind of genetic algorithm is not going to yield a truly intelligent player. For those not familiar with genetic algorithms, you express a strategy as a string (like a string of DNA). You create a population of strings, the first (or zeroth) generation. Then you test each one according to some procedure and assign a numerical fitness value. You breed the ones with higher fitness to form the next generation. And so on. There are lots of variations you can apply to the breeding procedure, but that's the gist of it. The question is, how do you encode a strategy as a string? You could encode the game right into the string. For example, part of the string could have 10 genes for what to produce. Each gene could be MILITARY_UNIT or CITY_IMPROVEMENT, for example. It would be easy to write a fitness function comparing how well the strategy's behaviour matches the players, and you could expect the strategy to converge to one matching the player's. (But a neural net could probably do it faster.) Of course, it only matches the player's strategy at the coarsest level, and certainly isn't intelligent. If you wanted something really intelligent, the strategies that you test and breed would have to be adaptable. For example, maybe I like to build mostly city improvements, but I build up my military if my opponents are doing that too. Now the strategies get more complicated, with if-then rules, logic statements, and so on. Even if you code it right, and it actually works, and it learns quickly enough to be practical, it's very hard to predict in advance what kind of adaptability you need. So you're still not going to get anything intelligent. If you haven't given up yet, you might try to go for ultimate adaptability by making the strategies completely general, so the strategy can be any computer program. If you believe the Church-Turing Thesis, this could become intelligent. But you'd have to design a programming language that can tolerate being bred and mutated. And the learning rate would be extremely slow. Think of how long it took to go from fish to human. And you can't simulate a fish in real time, so it'd take even longer on a computer. I'm not saying it's impossible to make something intelligent using genetic algorithms. I can't predict the future. But if it is possible, there're some revolutionary discoveries yet to be made. I would guess that they chose genetic algorithms for this application because it was the easiest to design and implement. I remember that while Command & Conquer was under development, Westwood was saying that the opponent AI would use genetic algorithms. I don't know what happened, but it didn't, and it was pretty stupid too. Generally, computer game "AI" is very simple, since most of the CPU time is given to the graphics engine. So, game AI is far behind the state of the art. (Apparently game programming isn't too good either. My game-programmin' friend says game programming practice is 8-10 years behind the rest of the industry.) All that said, it will be cool if it works well enough to be useful. Has anyone played Civ 2 or 3? The micromanagement gets really tedious toward the middle and end of the game. The game does have an AI that will try to make certain decisions for you, but they're usually not very good. So you have to micromanage everything yourself, until you get really sick of it and just accept the AI's stupid decisions.   [ ...reply just to this | comment on the story... | next new ]

10.  Re: Still no "intelligence" here   by Herban Legend       at Sat 7 Sep 1:13pm score of 1.5 informative   in reply to comment 9     If you wanted something really intelligent, the strategies that you test and breed would have to be adaptable. For example, maybe I like to build mostly city improvements, but I build up my military if my opponents are doing that too. Now the strategies get more complicated, with if-then rules, logic statements, and so on. Even if you code it right, and it actually works, and it learns quickly enough to be practical, it's very hard to predict in advance what kind of adaptability you need. So you're still not going to get anything intelligent. The unpredictability of what will work can be addressed, at least in part, by mutation - random changes to the 'genetic code' of some strings. There have been several notable genetic programming experiments where mutant strains became wildly successful precisely because they worked in ways completely unanticipated by the programmers. (Or even because they learned to exploit bugs in the code, to unfair advantage.) As far as "becoming intelligent" goes, a genetic algorithm isn't going to become self-aware anytime soon. A workable definition of what it means to be "intelligent" is still a long way off. Maybe when we understand what the question is, we can start to provide meaningful answers. In the meantime, genetic algorithms and neural nets sometimes do fascinating things. I think an obvious application of this type of programming in games is the computer opponent that learns to beat you and can adapt to your changing strategy. For this to work, the game has to be "fair" - the computer has to be playing by the same rules as the human player, no cheating. Combat simulators are often written this way (at least the ones I have influence on, dammit!); both the computer and player (and network opponents, for that matter) control their entities through the same kind of virtual "joystick", and no distinction is drawn between them when the simulations are run. For another intriguing possibility, consider a game like Tie Fighter (from a few years back, but it's the only example I can think of off the top of my head). After you had created and played with a pilot, you could play as someone else, and use the original pilot as your wingman. It might be cool if your (former player-character wingman) really fought as well as you did, using your tactics. The idea has been bandied about, but I've never known a game to use it - yet.   [ ...reply just to this | comment on the story... | next new ]   11.  Re: Still no "intelligence" here   by ideonode       at Mon 9 Sep 4:42am score of 1   in reply to comment 9     Then you test each one according to some procedure and assign a numerical fitness value. You breed the ones with higher fitness to form the next generation. And so on. There are lots of variations you can apply to the breeding procedure, but that's the gist of it. I have not a criticism of the parent post (which is very good), but simply a question I've been meaning to ask about genetic algorithms: how exactly do you go about 'breeding' algorithms? Isn't an algorithm just a chunk of code? How can two algorithms be bred such that the outcome is also machine-readable and semantically coherent?   [ ...reply just to this | comment on the story... | next new ]   15.  Re: Still no "intelligence" here   by OSULugan       at Mon 9 Sep 9:36am score of 1   in reply to comment 11     Generally you can use boolean flags to denote certain actions being taken at certain times. This is where the chess or Tic-Tac-Toe analogy helps, as it's a perfectly informative system. Theoretically you can analyze the board for every possible move you make, and then every possible response by your opponent, and so on. Most modern Chess games only do this for a certain number of moves, and will attempt to weed out "bad" threads early in the analysis (which is how Kasparov was able to defeat Big Blue). Now, if you assign points to certain positions and flag each piece's possible movement, you can now breed your game-play algorithm by just manipulating a bunch of boolean values and analyzing a bunch of integer values. Of course, for really complex, you can write mutating code, but I've yet to hear of an AI based on writing mutating code. And God says, "No, that's not right." Yeah. Well. Whatever. You can't teach God anything.   [ ...reply just to this | comment on the story... | next new ]   16.  Re: Still no "intelligence" here   by Mad Ogger       at Tue 10 Sep 11:59am score of 1   in reply to comment 11     How exactly do you go about 'breeding' algorithms? Isn't an algorithm just a chunk of code? How can two algorithms be bred such that the outcome is also machine-readable and semantically coherent? Yes, it would be pretty tough to breed C++. I've never actually coded any genetic algorithms, but I can suggest some things. - You could make a big list of if-then rules (e.g., if an enemy wanders within sight of my town, then build 4 more guards). Each gene determines whether the "organism" follows that rule or not. - You could have a standard algorithm with some parts that vary according to genes. For example, the "organism" might use a cost-benefit analysis, and the genes control how it assigns costs to things it senses. - Similar to the previous, you could have a function with some pluggable constants controlled by genes. For example, a simulated ant could have sensors for the position of its legs, and then a function to determine how to move the front left leg based on the positions of all the legs. You would select for "ants" that move forward. This starts getting very similar to neural networks.   [ ...reply just to this | comment on the story... | next new ]

12.  Code, or electromagnetic sentience...   by DarkReazon       at Mon 9 Sep 6:22am score of 1.5 informative     Carnegie Mellon Comp.Sci. Genetic Algorithm FAQ - Part 2. Perhaps this will explain. At the bottom is pseudo-code, convert to your language of choice, chill for 15 minutes, and enjoy. He who fights with monsters might take care, lest he thereby become a monster. - F.W.N.   [ ...reply just to this | comment on the story... | next new ]

18.  E-Z Play   by adamrice       at Mon 16 Sep 6:43pm score of 1     I'm reminded of those E-Z Off ads. A woman is tucking in to bed, or playing tennis, and turns to the camera to brightly exclaim "I'm cleaning my oven!" Just imagine: some guy standing in line for Star Wars Episode III turns to the camera and says "I'm storming the castle!" if irony were made of strawberries, we'd all be drinking a lot of smoothies right now.   [ ...reply just to this | comment on the story... | next new ]