[trilOByte]

Installed perfectly first attempt - excellent hack, thanks.

It does seem to learn dynamically.

Me: what is your name?

Bot: My name is Scully.

Me: How can I teach you new things?

Bot: If you want to teach me something say it in one sentence. Then say "Remember that."

Me: My name is Martyn. "Remember that"

Bot: It's good to meet you, Martyn. Don't worry I will remember it.

Me: What is my name?

Bot: Your name is Martyn, seeker. :baaa:

...then later in a different thread, she remembered my name and included it in a sentance without prompting.

Bot: Yes I sure am alive, Martyn. It's great to be alive, isn't it?

I think the bot certainly does learn, though the plethora of casually inflected English confuses the AI to bits. Basically, the bot just cant make any rules out of the way we usually gibber at each other in forums, it's way too complex. I think you have to start simple, with plain English and work your way up to it. Scripted answer/response bots have been around for ages, but are dull and uninteresting. What makes ALICE so different, is that it actually learns. It constructs it's own sysntax rules based on the input it gets. Sure, giving it a database full of pre-scripted responses, gets you halfway there, but the quality AI comes from it structuring rules, rather than following scripts.

At least that's the way I understand it.

I would *love* to get my hands on the ALICE aiml files used for the current (silver) version on the alice website. Ironically, to make it more natural straight off the bat, you'd have to feed it a huge database of questions and answers as well as a bunch of rules for using them. You'd basically have to remove the Intelligence from it, in order to make it *look* smarter.

Where it's going to fall over time and again, it with lengthy posts that contain several salient comments. It will never be able to pick out rational points and answer them in such a way as to produce a logical discussion. It can fool you briefly, based on simple Q & A, but complex rationale is decades away.

Once the novelty has worn off, I think by far the most useful thing for this hack is as an interesting interactive FAQ.

What I would like to see, is it be silent unless triggered by a keyword or command like "bot, what is....". You could work out some method of an occasional random reply, just for interest sake, or a welcome message for new members etc. But giving it totally free reign, will probably just produce a flurry of novelty activity, resulting in a garbled AI.

This hack has huge promise and could be one of the greatest ever.