ID:272728
Dec 30 2008, 10:10 am
|
|
I'm developing an AI system, and need tips on how to make it work with BYOND... I don't have a set language for it yet, but I am currently working on it in Java. I have done searches on AI systems for a while, but none of them have shown me anything in regards to actual intelligence. They all required pre-defined responses to inputs. I want mine to actually learn from what happens, and have its own responses... Any help on either the Java development side, and/or the integration side would be MUCH appreciated!
|
In response to Tsfreaks
|
|
Some lectures on youtube were rather interesting, I'll admit, but what I need is more like assistance in developing the actual program. If I could get it to the point where it can learn from what goes on in BYOND games (that I either host, or play), then I could take it where I want to. My biggest problem at this time is making it read from the dream seeker, and remember things. In my own games, I can make the chat output to a file, then have the system read the file, but that wouldn't help me in games that i didn't make.
But all that can't even be used without the programming that makes it process the information. If I could get some help on making an AI that could chat with the players on its own (as in, it doesn't need to compare what is said to it with files, then give predetermined responses), then I might be able to finish it... I wish I had a better example of what it is I am asking for, but the only one I have isn't public. (It was an AI developed into a Ragnarok Online server my friend owned. It was commonly mistaken for an actual player, but was really an AI system.) On the part where you said that the software is static, that I believe is not entirely true. While you must program the initial functions and procedures to have it even know what the information is, you can make it dynamic by using databases (see my theory below). Also, by using databases, the AI could then remember information, and use what it already knows to figure out the details of it (like, if the AI was told "It is cloudy today", then it could take that phrase apart, and reference that the person speaking is talking about the weather in their location on that day, which is cloudy). By using this kind of reasoning, then the AI would be able to "learn" and understand based on what it is told. Granted, we would have to give it some information by default so it would initially have something to compare to... ((And now, The Database Theory)) The AI would have access to multiple databases. Each database would have tables relevant to the database's purpose. For example, one database would be "emotions" and contain a separate table for each basic emotion (happy, sad, etc). Each of those tables would have the extended list of emotions for the basic emotion (ie. "happy" would contain "glad","joyful", etc...). Each row/emotion in the table would have fields that described things about the emotion, such as when it can be used, or how extreme it is in relation to the basic emotion it goes with. So correct me if I am wrong, but using these concepts, it SHOULD be -possible- to make an AI that learns from other things, right? |
In response to The_Zergling
|
|
From what I gather you want you'd be better off having a child. Would probably be less time consuming :-P
The AI for Ragnarok Online you referenced would have had to lookup predetermined responses and be taught what to say in a given situation, AI is basically a finite state machine in fact natural intelligence is more or less a FSM as well (in the scientific sense) , and now I have stepped into the philosophical problem of Personhood. If you want a chat bot then there are probably quite a few tutorials on things like that for chat clients try and search for "ICQ chat bot tutorial" and try to apply it to BYOND. |
In response to The_Zergling
|
|
I'm afraid the problem you're trying to solve hasn't ever been solved before. We can't help you make an AI that does semantic analysis of things it is told and constructs grammatically and semantically accurate responses. Mainly because semantics is tricky.
If you want something like that, you'll have to go with a database of replies to individual phrases, or a system more like ALICE that just works out what grammatical structure someone is using and then rearranges the words being used to produce a response. |
If you haven't checked out PolyWorld, here ya go.
http://www.youtube.com/watch?v=_m97_kL4ox0
I've been wanting to do an AI project as well but the more I learn about AI, the more turned off I get. Everything in software based AI seems so static, problem specific, and well... just unimpressive.
I personally believe (currently) that replicating independent intelligence capable of "true learning" will result from work done in brain cloning and manipulation. If they can unlock and streamline brains (insect, animal, or human), there's no telling what they would think of.
tsfreaks