[TerraOmnia]

Behavior Based AI's basic idea is to take sensor input and "dumbly" just react to it. Several levels of these systems reacting in different ways to their own constraints inhibit or enhance other levels, leading to emergent "smart" systems. Usually this technique is used only in robotics.

Has anyone ever heard of this technique being used for language? If there is a chatbot which is implemented this way, I would love to know about it. If not, then why not? I am aware of some stigma about using symbolic representations in the technique, but surely that cannot be the only reason.

[Art]

Several years ago there was a bot that seemed to respond quite favorably to positive reinforcement or phrases when chatting with it.

I asked its author and he said that it was a feature that allowed the bot to "Accept" and file away the reply in a positive and acceptible manner. Useful in future chats in case the same or similar subject / topic was discussed.

Most chatbots seem strictly based on pattern matching and some can alter the responses based on previous convos while others are strictly "canned" replies.

My late friend, Knytetrypper, was able to identify which type of bot was chatting based on the responses it gave after only a couple of sentences.  If I were to construct a bot there would have to be some degree of randomness or self-generated reply mechanism.

Check out Daisy by Greg Leedberg. It starts out only knowing a few basic rules and over time learns how to create and structure its own sentences. It's no einstein but in time it will amuse and amaze you.


...Later...

[TerraOmnia]

Thanks, Art. I looked into it. Daisy certainly did amuse me, pert girl.

It sounds like your friend spent a lot of time with chatbots, that's a pretty cool talent they developed.

I looked at Daisy's source and although I learned a little, I don't know Pascal so my understanding was limited. Although it wasn't certain it was behavior based AI, I tried coming up with an example implementation of what I was talking about and found difficulty. Finally, there was some inspiration when I learned that apparently the heart has a mini-brain. The interactions between heart and brain to inhibit or enhance certain behaviors (such as amorous actions around a loved one) would indeed count (in a sort of remote, detached way) as this sort of thing, if I understand it right.

A dual neural net system which mimics this poetic interaction might not exactly be a chatbot, but might be able to be set upon the task. I'm still fleshing out the idea, but I wanted to write it down/share my thoughts before I forgot it in the midst of finals. I'm aware of some obvious obstacles, but in the sketch in my head it seems that the heartbeat itself is key, as well as "damage" to the network for "creativity". I wasn't remarkably impressed with the neural network Nick (though I admit I only played around for half an hour with the sample file), so I hope it's not a completely wild idea. Please tell me what you think.

[Art]

I would suggest you whet your whistle a bit by watching a couple of short videos.
The first is from a company who is promoting a conversational agent technology if you will. The program's personna is Cassandra and she (it) is able to carry on a conversation and provide answers to the users questions while chatting. Quite lifelike and she even answers based on previous alusions to words like "it". Watch the vids, Hamlet and Space. Even though the program uses a structured XML format it is quite impressive in its design and execution! Mr. Emmett Coin seems to be a very nice man as he has answered several of my inquiries in the past. http://ejtalk.com/v1/demo.html

Next would be a most interesting chatbot (again conversational entity). called Silvia. You can find her at www.cognitivecode.com. Look at the 2006 proof of concept video. Leslie Spring (her creator) assures us that their conversation is not scripted in any way and that her program can be used in a variety of applications.
Again, very impressive.

The problem with both of these remarkable programs is that we, the users, can't really get our hands on the programs to experiment or play. If the day does come where we can, the programs will most likely be financially out of the reach for the average, student or working person!

It's still nice to see that some are trying to push AI in more positive directions and with good results!

Thoughts??

[TerraOmnia]

Ah, finals are over! Sorry about dropping off the face of the earth for a while.

These two are... very impressive. That Silvia asks several times for a repeat makes it easier to believe it isn't scripted.
Cassandra is done in xml?? Goodness. I'd seen her before, but I didn't know what was lying behind it.
Thanks for sending me the links.

It is a shame that there's no demo which we can play with ourselves. Is there any word on Silvia being deployed in applications yet? I read on their crunchbase account that they were signed on to Northrop Grumman (apparently a security company), but I didn't see any products that were obviously Silvia based.

Since I have to actually make something to satisfy the project requirements, it's possible that I could attempt a simplified (I only have a year to work with) reconstruction of one of these things just based on what I can see, but... that might be a bit difficult since most of it is inferred and I'd probably get marked for lack of creativity anyway.

It really looks like they're doing it, although through different strategies. It's both depressing and inspiring, haha. I'm not sure where I can go from here.

[Art]

TerraOmnia,

Glad you liked those bots.

Do check out a couple others like Cameron Jone's Jeeney - She searches the web in an effort to help obtain appripriate answers yet her response time is quite rapid. www.jeeney.com

AND

Bruce Wilcox's Suzette using CHAT-L  programming and is featured on the Simulation / game Blue Mars. (I assume...www.bluemars.com).?

Also give a long look at our hosted Chatterbot Collection on this site. Look for it....

Luck on your assignment!!