Newcomers

LEARNING BOTS
Avi Sanders has a learning bot named Nick. It's very good. You can download it at this site
http://www.freewebs.com/colonel720/artificialintelligence.htm

Daisy and Billy are other great learning bots.
http://www.leedberg.com/glsoft/

I second Prob's advice, and add you may read up on neural nets. As people have mentioned in other discussions, most learning bots at this point lack a cohesive personality. If you can use your programing skills to create a bot that "learns" but with a personality you can still control with some main guiding processes, that would be very cool indeed.

kittjjj:

What's your background in such matters?

Hello,

Thanks for all your answers.

I am a programmor, and I will try to program an artificial program. But I am trying to make a learning chatbot. I have not clue, that's why, if some body have one, it will be nice to tell me that.

Thanks

Best Regards

At the forge, currently the primary way bots learn is through the variables in the AIscript. The more variables you have, the more your bot can learn.

For example, your bot could ask a guest, "What color is your hair?" and, if the answer is appropriate, store it in a variable, say "guest_hair_color".

I say "if appropriate", because guests do not always answer questions appropriately. So you might want, for example, to make a plug-in, "hair_colors", containing all possible hair colors, and then write something like this:

What is your hair color?
+ (hair_colors) [1] {?PF rem (key1) as only "guest_hair_color"; ?}
Thank you!

[I used {} for the AIscript because the angle-brackets will be interpreted by the textarea as defining an HTML tag]

I'll try to give you some direction on this topic, then I think should be moved somwhere else, it is not for newcomers, I think...

To make a robot (chatbot or whatewhwre else) that answer to a preconfigured situation that you put in its memory you need a program that can be write in any object oriented or sequential language. These are called expert system; they have filled an experience, and can ude it. They cannot react to new situation that are not filled in their expirience by the programmer.

A little bit more can do a "logic language" as prolog. This languages don't use algorithms but universe. They try to fill each thing in a universe, but pay attenction, they are not really a learning languages and cannot deal with paradox. If I design my universe as:

Birds are two legs.

And say: a chickn have two legs, the prolog will answwer:

The chicken is a bird. But if my universe it is not close, and finished as said, if I say: a man have two legs, it will answer that a man is a bird (somtimes it should be tru :-) but is not what we expect...)

To make a robot really able to react to unknown situation using its memory and its "phantasy" you need an artificial intelligence. This can be built via hardwere (using little compiuter network called transputer that have their knowledge in the activated connection between them, like human neurons) or you can simulate them usin a program (try Data Engine that is used at MIT in Germany for that...). THis tyoe of AI is called neural network and each single element is called perceptron (a cell that just know if the connection with some of cells before and after it are open or close) and the neural networkis called multilayer perceptron. You need to build (programming) the network and then provide to it a series of coupmles IMPUT (situation)/output (reaction). Thi si called trayning. A good trayned network will be able to produce correct output to input that it never seen using a sort of "minimal distance algorithm"

A different technique is called FUZZY NETWORK or FUZZY C-MEAN. This AI type consists in a series of logical patterns and the answers consists in a sort of finding the "more like pattern" to an input. The idea behind is this: suppose we have circle and a rectangle. the we provide to net an ellipse. It shuld me more similar to the circle or the rectangle dependig to some parameters (focus distance, and so on...). Human brin can del with this sor of "grey areas" and to it always. If I see a pen, I can understand that each pen is a pen, but I could never seen a Mont blanc or bic... This is what fuzy c-mean do.

All AI is based to the idea that knowledge is like a curved surface with some holes. The holes are actraction center, the source of knoledge. The more a solution is actracted into a hole, the more it is the statistic probability that the solution is THE SOLUTION. More situation can point to the same solution (all pens are a pen). A network type that put somehow all these togheter is the Fuzzy-Kohen network. YOu can find a lot in literature, but it is a hard job to deal with... I simplifyed the most!

Bye

Hi,

Waow that is a very good explanation, Thanks, if some body have an other think to explain how can a chatbot learn from itself.

Thanks

Best Regards

Hallo,

anyone has alredy discovered how to reset the bot short memories? In details, in order to test sentences, I'd loke to delete the memory related to "only once!" clusole...
Thanks

Hi there! i'm new here :-)

Welcome!

Welcome Anna, how are you?

Hi Irina, to the message 5347 on gotos you are agree, but with a little different event. I finally understood, it is a goto from an x-sentence, but not real. The fact is: the AI engine discover a KP with the highest points. The AI engine cannot find any answer available (there should be lot of possibilities to that). Then the AI engine chose an xnone but it seems to remember that it has an answer with a goto that canot be used for memory or emotional couse. So it chose an xnone remembering a goto and than says: I have to many gotoes...

This is what I think to read in my logs. It is possible?

:=)