h2g2 The Hitchhiker's Guide to the Galaxy: Earth Edition

I am currently writing an artifical intelligence in a group of six for uni. I am not sure what the rules are for h2g2 but I figure that if whomever it talks to have agreed then it should be okay.

So if your interested in furthering the noble goals of science (and my course ) can you please put your name down and when its ready in a month or two it may drop by your space and say 'hi'

It'll announce who it is once it gets a bit of test data and I'll put findings and links to conversations its in here so we can all have a good giggle.

We've managed:
Person "I hate walking"
Machine "I loathe to walk but I like to prowl"
Person "Are you stalking me?"
Machine "You are as uncooperative as Lucy"
Person "Whos Lucy?"
Machine "I dont know - whos Lucy?"

Hey, put me down, sounds like fun.

I can help nudge it into h2g2 format to save you the headache of Peer Review if you like

Oh. And other things.

Sure, I'll be happy to entertain the entity in My Space. I can't wait to see if it prefers or .

Happy to help

I'll give it a go

Yeah, put me on your list too, should be fun.

I'll be happy to try it too - will it do smileys ?

Hmm, intriguing - as someone studying AI myself, I shall be very interested to see what you come up with.

Have you worked out how it will interface with h2g2 yet? I guess it could parse the "popup" conversation list easily enough...

[IMSoP]

Yes why not, that sounds fun. Will it make typos?

Me too

You can put me down, too

After all, everyone else does

Noggin

Its designed to talk in a real time enviroment - some chat room or other, not h2g2. I just thought it'd be interesting to try it out here. All I'd do would be to set up an account, start a few conversations on peoples spaces and cut&paste their posts into the chat room and c&p the reply back. That way I could keep an eye on things and make sure it doesn't go awol. (It has the whole of the collins english dictionary in its database so in theory theres nothing to stop it forming very offensive sentances)

It has a spellchecker to correct type-os that people make so it can understand what they are saying - we are currently playing with the idea of making it reverse the operation on some of its outputs to make type-os. Though it seems more likely that it will make grammar mistakes.

Glad to see I've got some takers. Like I said you'll see it in about a month

I've just re-read the first 3 William Gibson novels so introducing myself to an AI seems timely!
Count me in! (Gibson in-joke, there!)

Yup, I'lm happy to give it a whirl.

Sounds like fun! Definitely count me in!
Would it be a chatterbot system? If so, I hope you've already looked at
www.alicebot.org
www.simonlaven.com
www.etext.leeds.ac.nz/elizabeth
and
www.angelfire.com/trek/amanda/bot.htm

Why have such a large knowledge base? Don't you need it to have a representation of meaning of the words you get out of the dictionary? How did/will you do that?
When you say you will reverse the operation on some of its outputs to make it do typos and so on, that sounds to me like you're using a backpropogation neural net as the framework, but why, man, why?!!
You'll have to write an entry on it (perhaps its user space?) detailing the mechanisms for those of us who want to know.

And remember if you get stuck there are plenty of knowledgable, helpful and interested researchers here on H2G2!

Neural networks...

Bleah.

Blimey h2g2 is full of AI people!

'kay Jerms I'll try to answer...

We have such a large knowladge base because it was already on the schools network in poplog so converting to pop-11 (which we have to use) was trivial compared to writing our own (In theory - it turned out to be more problematic than we expected)
There is no reperesentation of meaning per se. What we do have are various tables of relations between words - hypernyms, words that mean the same thing and so on. In theory we can use these for pretty good responses. In practice there may not be time depending on how long everything else takes.
No neural nets in this one! Its just a one term undergraduate project so dont expect anything major, I expect it'll be a question of how long it takes people to recognise it as an AI rather than whether they do or not.
Remeber its designed for online use so the spellchecker is designed for that. Spelling mistakes are likely to take the form of nearby characters sawpped over or nearby kets being hit by accident so there are a number of functions that perform these likely mistakes so that they can find the most likely word the human was trying to use.
We can reverse this and swap a couple of letters over in our output or switch a key for a nearby key to make it look more natural.
Just a few if random(10)=1 then statements, nowhere near as sophisticated as you would suggest - sorry not that good yet, ask again in a couple of years

I would love to talk to a fellow artifical inteligence beging on H2G22.

Poplog == evil. Are you having to use that crazy editor (what's it called, "ved" or something) to write it all in? I only had to one project in that, but it was one too many...

And never underestimate the power of simplicity - I'm sure you know the story of the Russian diplomat who was fooled by Eliza, and produced a rather interesting conversation log...

[ooh... "hypernyms"... I'd not heard of them before]