5 replies to this topic

[taza0]

I have recently become interested in AI research. I have only come across Tom Ray's virtual lifeforms and the MIT wolf group simulation. I would appreciate any links or book references relating to AI research. I am specifically interested in theory of group minds, genetic implementation, present research developments, and relevant programming resources.

(Tom Ray's Tierra: http://www.isd.atr.co.jp/~ray/tierra/)

Thanks

[Bruce Klein]

Welcome to the forums Taza...

Could you be a little more specific about your informational request.. there's and enormous amount of AI information out there. Are you looking at an end goal of some kind? And are you familar with the Singularity Institute's quest for AI? http://www.singinst.org?

BJK

[SiliconAnimation]

I came across a markup language called AIML (artificial intelligence markup language). I've spoken with the chat bot Alice and it works pretty well. However when I looked into AIML it seemed more like a cause-effect sort of responce system. There doesn't seem to be a whole lot of logic behind the bot's understanding of exactly how to respond.

I have been amusing myself with a small attempt to program an AI chat bot with VB 6.0.

I think chat bots are a good start in the quest to understand AI. I'll throw out my idea since I don't think I will go very far with it.

I want to have the bot seperate a sentence into word fragments. Each word will be stored as a variable and be given a number based on where it was input in the string sequence. Then each word will be compared to a database of nouns, pronouns, verbs, adjectives, etc. Once a match is found, the database can load various attributes about the matched word. Such as good and bad.

Example:
You tell my bot "Kick Jeff."
Kick is found to be an action verb. The bot checks its database and finds that kick must be directed at something. Jeff is number 2 in the string sequence. The bot attributes Jeff to the subject of kick. Jeff is in the database as a pronoun with a value of 100 on a -100 to 100 scale. Kick is -10 on the same scale. Kick is now attempting to subtract from a positive number. The computer will identify this as "bad" and now has its first ethics program. Kicking Jeff is not a good thing.

On the flip side lets say the programmer doesn't like Jeff. Jeff gets a -50 value, now the computer sees a negative plus a negative. Two wrongs do make a right. The computer agrees that Jeff should be kicked.

This is a weak system that I'm working on but, I am only a 2nd semester programming student so, I will have time to develop it. Now I give the idea freely to anyone who wants to elaborate on it.

[AIGuy]

Have you checked out this year's bot contest yet at http://www.chatterboxchallenge.com/

Alice based on AIML is entered along with about 87 other entries.

Currently my bot Project Zandra was running 4th place last times I checked.

Voting by the judges in not complete yet though.

Most bots are based on strict stimulus/response without taking into account that many internal variable states.

My bot has been programmed to try to extract the same information as a human does from an input and then save that information as an internal variable state.

My bot is written in VB.Net

It has about 18,000 patterns a little less than half he 40,000 that Alice has.

But depending upon the expressiveness of your pattern language it is possible to design a language where one pattern is equivelent to many patterns in AIML.

Many times users mispell items in their inputs. Try to design a pattern language that allows misspelled word to be recognized.

Multiple wild cards is also a necessity, as well as several pre and post processing steps to normalize input and output.

You also probably will need to impliment conditional logic, variable assignment, and other functions in your template language if you want to be able to extract and act on contextual information that occurs during the conversation.

Most bot languages allow you to code around 50 patterns in an 8 hour day. So the more expressive your bot language is the less patterns you need match human variations for the same thought.

I have not yet incorporated the Bayesian probablistic or fuzzy logic in my knowledge base as you are describing.

I don't mean to discourage you but plan to devote a considerable chunk of your life to the project if you are attempting to compete or win the Turing test.

Even at 50 ppd (patterns per day) every day it has taken me quite a few years to get as far as I have.

If you could design a system that automate pattern creation from other input sources such as the web or have the bot generate it's own patterns from learning from the user you might be able to jump ahead of the competition in a reasonable period of time.

[Lazarus]

well asanti technology as it sound, and usually that is NEVER my veiw point, I do not think that AI should be allowed as we have already shown jusst through science fiction alone that if we create ai no matter what the intentions on either side we will eventually be destroying mankind as we know it, simply for the fact that a machine can progress and will progress expotentionally as a 100 time more accelerated rate than human ebings ever can. because of thisthey will simply being more advanced lifeforms than in such a small amount of time that we have yet to foresee it. and the problem with this is even should they decided to leave us alone and move off planet and no longer interfere with us again, there will be human out there that will feel inferior to machines and feel threatened by them and because of this fear they will try and destroy AI, and as a logical step to keep human beings from threatening the survival of a race of machines, if we were to continue to persist in attack upon machines then the logical step for machines would be to eliminating the human race thus eleminating the threat to thier race. I am not saying that the machines would be evil, but the fact they they would be greater lifefrom than us would start a conflict we have yet to imagine, its just somehting that we dont need. I think the btter goal it to ourselves become machines conciously, for human being to integrate with machines and then after we are machines and can progrees just as they do and at the same accelerate drates then if we were to created AI they would be no different from what we already were and therefore there would be no conflict. but anyone who has seen the animatrix alone should know that if you make a machine things, just because of the greater stability of a computer over a human being they will grow and evolve faster than we can and become the dominate race and anytime one race takes dominace of a previously dominate race, especially if the race ebing taken over started the chain of event, then there is going to be a serious and fatal conflict between the two and it is likely the a race of thinking machine would win. just my input, I am not against I entirly I think that the technology should be explored but it should not be releaced and it should not be allowed to grow to the point where it think byond what we can because its not healthy to create god.

[Lazarus Long]

Too late...

If it can be imagined someone will try to create it.

If it promises power, some will make their best efforts and let no one stand in their way.

If you want it then I must have it to balance this possibly totalitarian force.

Hence we will race to accomplish this and nothing will stop us; most certainly not any claimed or promised prohibition, which will only have the effect of driving the process into more arcane and less regulatable areas. Building AI out in the open is the best method and where all examine the process is probably the best way but it is most naive to expect "the powers that be" to behave rationally about this.

Ender's Game

A suggested reading for all that are examining current overlapping events is the Xenophobe series by Orson Scott Card

Edited by Lazarus Long, 04 April 2003 - 02:27 PM.