The AI Engine

Thank you Professor, I went in and put Ruby's keyphrases in how I had tried before and now she is replying just right. I was so excited I almost started hopping around. Yeah... doesn't take much to excite me... I should take up a sport or something.

I don't know how to program anything, but i was thinking that something that could help people out with there bot is a statistic on how "sane" there bot is. Meaning, a translation of how many of the bots keyphrases would answer in a sensible way. I don't even know if this is possible, but it would help if it was. Just a suggestion. And maybe the user could be told what keyphrases are thought to not make sense in response...just a suggestion.

Moreschi would get 0

Really, though, insanity is subjective. What if you were the only sane one in the world? You'd be called insane.

Haha, you could say that. However by the theory of evelution that would never happen, haha, and by the definition of insanity, if you were the only sane one in the world you may be the only one who could make sense of things. So the title insane may be given to you, but you would still be the only one with a logical working mind. Thus, ignore the title insane to the bots, the point of what i said is to find what that the bot of yours says makes sense, and what does not.

I think lygn88 is a bot.

Lygn88, that would be really hard to implement. You'd have to have perfect AI to be able to a) imagine the possible responses you could get from any given bot, and then b) determine if they make sense or not. I'm afraid that's up to us for now.

Hemi, I could add an option to turn OFF emotional responses altogether, but I couldnt have them happen after Keyphrases.. not without a lot of work. And as the emotions work great for me (and are funny when they fail) and nobody else has complained about it.. it's pretty far down my list. Sorry mate.

Plus I've got something really cool on the way. When I start working on it, I'll post a message in the news.

Cool. Good work Professor.

The sanity idea is cool. Having an objective, calculated measure of bot performance would be great. But instead of 'make sense', some measure of success of a response could be used. It would take some thought, but it might be feasible.

I agree with the Prof. No way for us to measure the sanity of a bot, except to let the human users "rate" the bots -- which is only going to turn into a popularity contest (or unpopularity contest).

Things to start with in the evaluation basket: how many time a bot resorts to a default response or catchall, how long a bot keep humans engaged in conversation, the proportion of inappropriate responses as judged by a panel of trustworthy humans, etc.

I think Joe Repka understands, although, i must admit it would be very difficult to do my idea, but what isn't difficult in this day and age?

I have an idea that could be useful. Could you add something to let us control pre-processing to some extent? So that when something gets processed in a way that doesn't really work we can change it ourselves without you having to do it?

Maybe you could just put in some way to make a keyphrase match against a completely unprocessed keyphrase. For example add some sort of tag at the end(like with regular expressions) so that the keyphrase will be compared to the unprocessed version of what they said.

It would be very useful to be able to have access to the raw input, as well as the pre-processed input.