Can a bot to learn?

Clearly, yes, as seen in the thread http://www.chatbots.org/ai_zone/viewthread/411/ and as we know there is a tag that allows us to increase <learn> knowledge, as our needs, example to “learn ” about a topic that we have already loaded, or categories of learning to “learn ” certain concepts that are introduced in a particular way.

But what can we learn from the experience? Can you learn questions and answers entered by the user while talking to him?
In our effort to build a bot is the maximum coverage of user questions and also have a variety of important questions to ask.

What if user feedback those questions for which we do not have an answer and reuse them to another user to get answers? Or what if a different response to which the bot has
included it as a valid response to a question from a user?

It could be from a small group of categories (if any) be asking questions and little by little you learn, we ask those same questions and learn answers. In a few sessions the bot can be prepared to interact with a real user (and if instead of a user were thousands, there would be a variety of questions and answers very large).

I know that there are many disadvantages, such as unwanted things to learn or be inconsistent with the answers, but that may be treated differently.

If the topic you might find interesting debate on how to implement this self-learning

While I’m by no means an expert, I thought this was the approach taken by Rollo Carpenter’s cleverbot ?
http://en.wikipedia.org/wiki/Cleverbot

Cleverbot differs from traditional chatterbots in that the user is not holding a conversation with a bot that directly responds to entered text. Instead, when the user enters text, the algorithm selects previously entered phrases from its database of 20 million conversations. It has been claimed that, “talking to Cleverbot is a little like talking with the collective community of the internet.”

This is much the same way as Google voice does with voice recognition- a data cloud approach (using a human voice pattern database to deconvolute the meaning of any given voice input).  Since human language does not follow any one common template (cultural evolution seems to play a more important role), it may be that simple hard-coding of responses, or even iterative pattern matching (AIML) does have the innate capacity to evolve and thus will rapidly become obsolete without constant human grooming of the call-response code.

Thus there is certainly a need for an iterative input, data driven AI for that AI to have any real temporal staying power as a convincing (and socially relevant) chat agent.

I know that the issue of self-learning is complex, but even without fully and properly analyze the input.
In addition to learning self balance creativity happen over bot programmer to programmer language engine. Although I believe that it is not entirely true because the self-learning may be more with the data of knowledge and creativity in the style of the personality of the bot.

Even so, it is a fascinating subject and complex to implement.

Apparently computers can learn to read manuals

http://web.mit.edu/newsoffice/2011/language-from-games-0712.html

Very cool, Shi! Thanks for the link.