I am trying to think of an idea of an autonomous bot to implement my ideas of a human-like agent. I just came up with the following idea:

Create a bot that you can ask questions. This bot will then e-mail some people he knows and wait for their replies. When they come up with some ideas he will report this to you.

People who are sent the e-mail may either give the answer immediately, or return the e-mail address of someone who they think does know. The bot will process these relations.

Questions are related to subjects (tags). People are clustered around these subjects. People that have given correct answers on certain subjects should be tried first by the bot.

The bot may send the e-mail to several people at once. If several answers agree, they are more likely to contain the truth.

Once the bot has given you the answer, he will want to know from you if this answer is correct. If it is obviously wrong, the bot will decrease the believability of the one that gave the answer, and will hesitate to ask him again.