Slashdot Mirror
Keep Playing With AI
An anonymous reader writes "The BBC reports
how a newly developed AI system 'learns' your playing behavior and can even play for you when its time to take out the garbage or do other non-essential things around the house. My only question is if it could even learn to bs for me on those laggy starcraft 3v3 games."
Something like this?
The simple truth is that interstellar distances will not fit into the human imagination
- Douglas Adams
I would prefer that the AI took out the garbage so that *I* can continue playing my game ;-)
Besides which, who wants to give up their game for "someone" else to play. I mean it would be bad enough coming back from running an errand and finding that your sibling/gf/friend has died and put you back to the start let alone your friend. Or even that they've managed to get you past the point you've been banging your head against for ages so that now you feel cheated at not having achieved the goal yourself.
Nope, I think the "pause" button is not going to be replaced by an AI any time soon.
A little planning goes a long way...
Then I want it to play the best strategy, not do the same stupid stuff I do wrong...
Imagine all the lonely AI cyber-sex sessions that will happen in this world...
Seriously, though, there has to be a line drawn here. Sure, it'll be good for parents to get the kids off the machine for dinner, but won't it eventually lead to being an all-AI game? Isn't the point of big games, like MMORPGs to be that the people with no life and play 800 hours a week to have better characters than the casual gamer? With this system, you teach the AI to practice blacksmithing, let it run day and night for a few days, and come back with a master blacksmith. Just seems like you are taking out the challenge of the game...
For the record, I don't play MMORPGs.
Good quote, too many chars. Seriously, the slashdot 120 char limit sucks!
...when I need to get off my lazy game-playing ass and hit the gym?
I don't think you want spend all that gold on that that Palladin shield. Maybe you should relax a bit and think it over.
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.
So If i'm not very good at a game the ai wont be either? Even so this could be exploitable and used to be better at a game than a friend, we all remeber zbot from quake2.
He said many players of online games become frustrated because their lifestyle limited their interaction with a game world.
but in a stragagy game you can run when nature calls and be mostly ok
Typically they involve creating lots of slightly different solutions to a problem, testing to see which perform best and then taking and randomly mutating these to produce a new batch that are again tested, mutated and so on.
They should focus this advanced AI on the computer players of the game not into an autopilot mode.
"The secret of success is to know something nobody else knows." -Aristotle Onassis
Ok, I'm no hard-core gamer but personally, I can't think of anything worse than AI making guesses about what my strategy is and what I'm planning and thinking of doing.
So the question is, what's the point? If "real life" intruides on my gaming, I simply hit pause and come back to it later.
It just seems to me like one of those things that'll make people go "wow!" for the first couple of minutes and then never use again.
In other words, a bit pointless, especially if you could have been spending that development time doing something more worthwhile (like adding depth to a game, improving other AI, adding extra levels, better documentation etc. etc.)
Avantslash - View Slashdot cleanly on your mobile phone.
One big problem that is immediatly obvious. If the games fancy genetic algorithm is actually good, you could probably hack the game with a state file made by someone who was really GOOD at the game.
However, as anyone who's played this type of game before would attest : sometimes you just HAVE to go afk RIGHT NOW or you lose your connection to the server due to technical reasons.
It would be very nice if the game would take at least basic action to prevent death (such as casting teleportation spells, healing, running away...SOMETHING other than standing there) if you get attacked and you are afk or linkdead.
I wonder if you could get the computer to absolutely demolish your opponents with rushes if you simply have it watch you build a few buildings, set up a super-fast rush attack, repeat, and then just turned it on and watched it go. If you did it fast enough, and the AI's sample size of your behavior small enough, you in theory just watch it pump out Zergling-equivalents after just 3, 4 minutes of "teaching it." Let the AI 0wNz0r your opponent... unless he's doing the same thing. I can sort of see this becoming something of a fad, who can "train" their AI's to be more vicious/effective.
My only question is if it could even learn to bs for me on those laggy starcraft 3v3 games.
I don't know about bsing but maybe if you hooked up a mechanical system to your serial port and ethernet cable, it would learn how to pull the cable out of the wall just before the end of your starcraft games.
Outdoor digital photography, mostly in New Engl
Virtual Fighter 4 for Sony Playstation 2 (PS2) allows you to train fighters to fight in your style. It takes quite a bit of repetition to get it to learn, and then it can fight for you and like you do. It is very reminiscent of the portable Gigapets and their ilk.
Imagine, I can now eat hot buttered popcorn with both hands as the game plays for me! Is their no limit to my weight gain?
Rather, there won't be much of a need to teach it to play FPSes while using aimbots or RTSes while using de-fog of war'rs; once these bots are released it'll only be a matter of time before the proxy bots themselves are hacked and that will present another, potentially impossible obstacle to overcome. If it can out-micromanage and out-aim a human and it can't be detected as it's sitting as an external process, but some may use it as its intended for (why? can you not turn the game off for 15 minutes?) then there's going to be a lot of hand-wringing and a lot of irate people on either side of the fence.
Easy does it!
This comment has been submitted already, 276865 hours , 59 minutes ago. No need to try again.
Actually, at one point there was a game based in part on the ability to be disconnected and return called NetStorm. I actually liked it quite a bit (was a beta tester and bought it when it came out) but it ended up selling a very small number of copies and all the players on the server were using hacked clients by day two of the actual release.
Anyway, the game would fight on while you were gone, which was possible because the pieces were stationary cannons and the like, so when you came back you probably were a bit behind, but not wiped out. I won a few times after a reconnect, so the idea worked.
Sig under construction since 1998.
I agree the server should be watching the player. How better to take notes on how to improve the gaming experience? But don't use it to play the game FOR the player when the player's bored with the game. FIX THE GAME.
If you've designed your game with lots of boring repetitious stuff which is well-suited for a machine, then you've gone the wrong direction.
If your idea of making certain events rare is a spawn-rate measured in hours or days, then you've gone the wrong direction.
If you think of your paying customers as gerbils who will do anything, especially hitting the spacebar or attack key every ten seconds, for eight hours at a stretch, then you've gone the wrong direction.
Instead, if you want to keep your player's interests, offer more entertainment that works within their available time. Make the player's time in the game more valuable. Make it possible to play a little over lunch, a little on Thursday evening, and still feel accomplishment.
For starters, employ adaptive spawning instead of location-based spawning. If the server notices a party of adventurers who haven't fought anything in a while, decide approximately how tough an encounter should be, then let it descend upon them. Vary the toughness, vary the approaches, vary the circumstances which trigger a spawn. Don't count server time to the next spawn, count character time to their next adventure opportunity. If the game isn't focused on hunting and leveling to the exclusion of all else (hah, yeah, like THAT will ever happen in THIS industry), then watch the players' behavior to decide what kinds of quests the player likes. Ration those out at a rate that keeps them interested, in character-time, not server-time. If the player plays twice a week, give them the stuff they like each time they log in. If the player really does enjoy slashing for hours on end, then give them a little surprise every now and then.
Massive multiplayer games should take advantage of the massive multiplayer-ness. Like, duh. The statistical analysis which could be done on player behavior in MMORPGs is staggering. The fact that game designers just don't bother doing it or using it, is mind boggling beyond the extreme. The fact that today's MMORPGs are essentially single-player games with thousands of human-powered NPCs just makes me wonder whether anyone really gets it.
[
Ever play UO? It's THRILLING to sit for hours and practice magic, or hiding.
I had a little metal ball that sat on my Macro Key - so I didn't have to sit for hours before I could go out in the woods and be PK'd.
"I can't give you a brain, so I'll give you a diploma" - The Great Oz (blatently stolen sig)
Nah, aimbots point you at the enemy. This copies your playing style.
So in my case, it'll consistently shoot about 5 pixels away from the enemy ^_^
Maran
I'm waiting for the day I can deploy my MMORPG character using an AI, head off to work, and once and a while repond to his thoughtful emails when he encounters trouble in the game.
I'll send him back a response telling him what to do, and if he screws it up, and doesn't get iced by some goblins or whatever, you can be sure it will come up at his next performance review.
I've heard/seen of macroing in MMORPG's for years now.
In this system, the AI learns your playing style and writes the macros itself.
Will I retire or break 10K?
"The Electric Monk was a labor-saving device, like a dishwasher or a video recorder. Dishwashers washed tedious dishes for you, thus saving you the bother of washing them yourself, video recorders watched tedious television for you, thus saving you the bother of looking at it yourself; Electric Monks believed things for you, thus saving you what was becoming an increasingly onerous task, that of believing all the things the world expected you to believe. "
An now we have AI's to play tedious computer games for us!
Strange game, the only way to win is not to play... how about a nice game of chess?
Lame I know, but with the WarDriving reference just 2 stories away on the main page I just had to.
I thought this would be a great way for children to practice the game. Seemed very "Diamond Agey" to me.
www.HearMySoulSpeak.com
I have developed an AI that will make your Slashdot posts for you. It just pastes big quotes from the article and throws in a few off-topic references to the DMCA.
- Have a picture
My only question is if it could even learn to bs for me on those laggy starcraft 3v3 games.
That's not funny at all... I always longed to meet one of those assholes in real life so I could slap them silly. Or at least scream at them for a little while. People like that are one of the many many reasons it takes twice as much time to find a decent starcraft game as it takes to play the damn game.
There are no trails. There are no trees out here.
Back in the day, when I was hacking netrek, I had a damn good go at writing a learning robot client, using a genetic algorithm.
I failed, along with every other developer that tried it. I failed because while the game is composed of simple concepts - speed, turn, weapons, tractors, transporters - the emergent strategic complexity is way beyond an artificial player.
The robot could win dogfights, but while it won the battle, the opposing humans were winning the war. It could never figure out or negotiate strategies. Even if I had got it to play a good strategy, the human opponents would have just found a better one, as they have done again and again when playing each other.
That emergent complexity and strategic depth is what makes netrek such a great game, even today. As a simple rule, if you can write an AI that can beat a human, then you've got a game that's strategically limited, like chess, rather than one where strategy must be a flexible concept, like go.
If you were blocking sigs, you wouldn't have to read this.
lately, the way ive been playing on battle.net, the AI would watch my strategy closely, evaluating diligently, predicting the style and outcome of my strategy,...
and at that crucial moment i get a phone call and need its help for a few minutes, i click it on and...
every character would slit its own throat, saving the gold and lumber.
There's nothing Intelligent about Intelligent Design.
No, no, there's no time machine. You remember the whole Atari ad campaign about "gaming at the speed of light"? See, when you do that, time slows down, so you can game longer. Get it?
Virg
The article appears to have been slashdotted, but even so, given the state of AI research right now as I understand it (and I'm probably behind), there are some significant limitations on how good a system like this could be.
1. If this system truly learns from players, it will require a significant set of training examples in order to 'learn' what the best decision is at any point in the game. In this case, I would think 'best' does not mean 'most likely to win the game' but 'most like the player'. With the number of variables involved in playing a game, it would either take a lot of saved data or a long time to learn any sort of useful evaluation function.
2. How would a learning system like this decide on which variables to examine when making its decision? Games vary widely. Usually, the less specific the variables--specific meaning the more evenly the value of a variable seperates information into groups--the less accurate the result of the decision will be. This is called 'Information Gain'.
Even if the system takes the easier route of applying its own evaluation function instead of trying the learn from the player, there are still a lot of difficulties to overcome. For some games--like starcraft--evaluating the state of the player's game is somewhat easy. Using starcraft as an example, one could attatch values to all of the units and then try to move to the state with the highest unit value. However, for other games--like, say, MMORPGs--this would be a nightmare.
As an example, one of the research projects I've worked on involved training a decision tree algorithm to evaluate link texts to web pages based on user evaluation of those labels. The object was to create a system that would take a page, create an anchor text, and then use a user-data trained evaluation function to choose the best label to pass to the page generator. Even with good data (albiet not enough) and good differentiation between page evaluations, our system was right about half the time.
I don't know if I buy a realtime learning gameplaying system that's good enough on current hardware. Especially one that works out of the box on all games. There are tricks one can that help, but the real thing isn't quite here yet.
- - - - - - - -
Don't worry, being eaten by a crocodile is just like going to sleep in a giant blender.
I'm wondering if anyone has attempted to link the knowledge stores of an AI chatbot like Alicebot.
I enjoy the fact that she can learn, but it seems she would learn at a much faster pace if she could link with other Alicebots via Jabber to syncronize her data stores.
Intelligent Life on Earth
I would want it to do the taunts for me.
nothing like have somebody (the computer, in this case) talk smack to your opponents when your hands can still be occupied kicking his / her ass.
speaking of which... AI to find "my style of pr0n" while hands are occupied with other stuff might be useful too. hmmmmm...
My life in the land of the rising sun.
I am far too busy with my daily routine and work -I want to play games, but my schedule just does not allow it.
This is so great, now I can have this thing play games for me all the time.
I feel like I am having more fun already.
There are some odd things afoot now, in the Villa Straylight.
I've wondered if Tivo qualifies for the electric
monk from the Dirk Gently Holistic Detective book
I can't count the number of shows that the tivo
decided I should record which I've never really
watched before the space was reused for another
show.
granted, the Tivo doesn't ride a horse
Apparently the Mods have Packard Bell mice AND AIWA Mega-Bass 3D LCD Super Casino Boombox/Stereos. Sour Grapes. :P
- Right now, even good games have their boring periods where the user must do many repetitive actions dictated by some simple rules. These actions would easily be learned by a good AI engine. Players could relegate such tasks to AIs and focus on the fun stuff instead. The trick would be to make sure the AI knows in what situation to begin the repetitive task -- maybe a little tougher, but should be manageable. It's similar to what medical diagnosis AIs have been doing for years.
- The ability of players to automate repetitive tasks will encourage game developers to focus more on the aspects of play that are really fun. It will make boring games more obviously boring -- if I can spend an hour playing and then turn it over to the AI and have it play just as well, then it's not really a very fun game is it? So, we raise the bar for innovative gameplay.
- Whole new types of games could take advantage of this, along with the always-on connectivity of broadband connections. I'd love to see persistent strategy games, where you can check in on your kingdom for a few minutes or hours each day, and the rest of the time your trusty AI "advisor" runs things for you. Sort of remniscent of the old BBS door games, in a way. With MMOGs expanding into things like The Sims Online, it's clear that the medium is ready for more than just RPGs. A learning AI of this sort could be a massively effective enabling factor for innovation (chokes on buzzwords, hehe).
I want to see it in action. I've got to wonder about their choice to use genetic algorithms -- they're notoriously "the third best solution to any problem" (neural nets are the second). Seems like a more traditional decision-tree-based approach would be more suitable. But it just depends on whether they've been able to tune the selection criteria and crossover function in a suitable way to address the problem at hand. Hopefully it works out well, because I'd like to see this approach become quite common.My deviantArt site