Just a quick one to ask:
What is the most important factor to be included in AI to you?
Thanks guys
- GunRunner
ID:152969
 May 18 2005, 1:04 am
|
|
|
May 18 2005, 4:35 am
|
|
digitalmouse
 |
for me, it's learning. nothing makes an AI more realistic than it's ability to learn from past experiences.
|
It depends on the application for me. For strategy or board games, learning would be the top priority. For action or RPG-type games, where your opponent isn't typically around long enough to learn, being able to work with other AI-controlled creatures as an effective team is a big plus.
| |
Igmolicious wrote:
It depends on the application for me. For strategy or board games, learning would be the top priority. For action or RPG-type games, where your opponent isn't typically around long enough to learn, being able to work with other AI-controlled creatures as an effective team is a big plus. Being able to outwit an AI using tactics and thinking is what I would value. If an AI continually moves to your position to shoot you dead, and can't be tricked, I'd say it's more just a procedure than an AI. So in my opinion, giving something the capacity go wrong is a sign of AI. :p | |
for me, it's learning. nothing makes an AI more realistic than it's ability to learn from past experiences. Hah! You'll certainly be waiting for a while. I don't see AI that actually learns being used in games anytime soon. Maybe adaption but not learning but adaptation algortihms require lots and lots of data and time before even small improvements are noticable. That and if you have bad training data its likely to only make the AI worse. | |
I think the most important thing is that it behaves naturally enough to be unpredictable even to the developer. I don't mean anything like learning, ..., just more implementing some normal algorythms. Thinking about several strategies your AI can select from, based on the specific situation, would be impressive enough. Having enough different strategies will give it a touch of unpredictability.
AI like defined in around 1950 (forgot who defined it) is still not available at the moment, so why trying to do something unreachable (for a normal game developer) instead of creating some nice algos which give the impression of AI. Would be still impressive enough for me at all. | |
I think you're referring to the Turing test. I'm not sure when Mr. Turing came up with it though.
The idea of the Turing test is as follows: If a human cannot distinguish between the responses provided by a human, and those provided by a computer, you have AI. | |
I dont think Turing was around when computers were about.
The Turing test has mearly been updated to add in the idea of computers. The Turing test was originaly based on diguising a woman as a bloke, or a bloke as a woman (basicaly: to fake your gender to appear as someone your not). AI uses the same sort of thing, its trying to disguise itself as a person, and this is where the turing test comes in. | |
Acebloke, if you did any effort at research, you would know Turning was heavily involved between 1945-47 in development of a computing engine the National Physical Laboratory in the U.K. "...In 1949 he became deputy director of the computing laboratory at the University of Manchester, and worked on software for one of the earliest true computers — the Manchester Mark I. During this time he continued to do more abstract work, and in "Computing machinery and intelligence" (Mind, October 1950), Turing tackled the problem of artificial intelligence, and proposed an experiment now known as the Turing test, an attempt to define a standard for a machine to be called "sentient"..."
The test itself is pretty straighforward, and does not involve switching of men and women to confuse the tester. "...a human judge engages in a natural language conversation with two other parties, one a human and the other a machine; if the judge cannot reliably tell which is which, then the machine is said to pass the test. It is assumed that both the human and the machine try to appear human. In order to keep the test setting simple and universal (to explicitly test the linguistic capability of some machine), the conversation is usually limited to a text-only channel such as a teletype machine as Turing suggested." -compliments of wikipedia! | |
It depends on the function of the AI, if you are using AI in a game, than yes, it should be capable of erring. But if you're purpose is to beat the world's best chess player, or to be used in military equipment, etc, you would want it to be a fool-proof as possible.
It's an interesting thing to think about in the case of games, because, people aren't super intelligent. In fact - IQ is on a bell curve. And inevitably, people, and therefore their intelligence, is bound to be flawed. Besides, who wants to try to play a game where AI is so advanced the game is un-beatable. | |
Rockinawsome wrote:
It's an interesting thing to think about in the case of games, because, people aren't super intelligent. In fact - IQ is on a bell curve. And inevitably, people, and therefore their intelligence, is bound to be flawed. Besides, who wants to try to play a game where AI is so advanced the game is un-beatable. Rainbow Six- my main gripe with it (the first game, mind you), is that the AI computer baddies could shoot you instantly and 100% accurately (head shots) if you could see only their elbow. So yeah, you're correct- AI in a military defense system is entirely different to a convincing AI in a computer game. | |
For games, it's the ability to mess up. If it's a pretty good AI, it'll know that it's an AI and is limited. The problem with that is that there's so many ways to mess that you can't possibly program them all, so you have to use a rand() function for calculating mistakes. You can factor in some basic categories of hinderances, but not really a decent percentage. Taking the fact that AI in a military defense system has to be perfect, and that AI is imitation of humanity, which is imperfect, then military defense isn't really AI, it's tracking procedures and communication with a real person(usually).
| |
I assume action/adventure/RPG games. I've never seen it in a game so far, so here's mine:
I want the enemy AI to behave realistically when noticing a player. Let's say you're in an enemy base, and you get noticed. What happens in 9 of 10 cases? The enemy attacks with a pistol against your BFG900000, or you slip behind the corner fast, and the enemy waits for a while and then goes back to doing nothing, enabling you to make a perfect headshot in the back. Shouldn't the enemy scream for help, sound the alarm, run somewhere, wait for backup, make everyone on the alert and so on? It might make the game difficult, but then put in some counters like stealing enemy uniforms, turn off the alarm and such. That's what I miss in game AI today. /Gazoot | |
I think an AI like that was in an N64 James Bond game, but I can't really remember the title. It had a blue cartridge, I think.
| |
Gazoot wrote:
Shouldn't the enemy scream for help, sound the alarm, run somewhere, wait for backup, make everyone on the alert and so on? It might make the game difficult, but then put in some counters like stealing enemy uniforms, turn off the alarm and such. That's what I miss in game AI today. Play any of the Hitman games. You'll love them! | |
Elation wrote:
Play any of the Hitman games. You'll love them! Or Thief. Unless of course you want there to be guns. =) | |
For a human to ask it a question, and having the answer be a human answer.
| |