How can we use A.I.M.L so that the ChatBot learns from what users say/type to them?

How can we implement CBR and Reductionism?

Help Appreciated. :D

If you are using Pandorabots, you are welcome to use my learn AIML file to base a learning chatbot on:
http://www.mitsuku.com—free aiml files—- learn.aiml

Well, I think he needs instruction on how to implement learning by bot one his own without any base. I think it would need algorithm that will analyze data and create different categories based on it - something like parser for analyzing human instructions and converting them into AIML. That would be cool, but I think we’re far away to complete something like that.

I assumed as he was referring to AIML, he may be interested in my learn.aiml file which does exactly that.

That would be cool, but I think we’re far away to complete something like that.

Not far away at all. It’s already written.

Umm, I think you didn’t understood me. It’s written, but it can’t be automatically extended like - someone write ‘do you watch Game of Thrones’ and, if bot doesn’t know anything about GoT, algorithm fetch some informations about it on the web, send it through the parser which writes new patterns and creates new topic which would allow better discussion on GoT. I’m not saying anything against AIML, but it’s static.

Actually, I’ll probably try to implement this as I think it will make great additional option for writing bots with AIML. XML parser would be piece of cake to write.

Ah I see. The OP asked that the bot learned from what the user said to the bot and so I based my answer on that. Yes, it would be cool though if a bot could analyse a Wikipedia page for example and generate its own AIML.

Not sure what you mean by AIML being static though. Using the learn tag, the chatbot can create new categories on the fly.

I’m using old specification where learn meant loading other AIML files and I was totally unaware of this new function of the tag. Thanks for notifying, I see it’s used in Pandorabots. However, I think it’s still good idea to create categories with informations from web (if it’s not already available as I’m catching up with AIML 2.0 specification).

Ah yes. It’s been around since AIML 1.0 and is invaluable for responding to questions based on input such as, “Peter is taller than Sue but shorter than Harry”. Trying to create hardcoded categories for things like that would be tricky.

Nikola Stojaković - Jul 7, 2016:

... I think it’s still good idea to create categories with informations from web (if it’s not already available as I’m catching up with AIML 2.0 specification).

Laybia AI does that to some extent, though machine parsing of web scrapes (even xml) into AIML catagories is a tricky proposition and still needs some human “cleaning” of the AIML.  I am always looking for ways to overcome the “static” nature of AIML as it is traditionally used (as a db with or without user specific input learning adding to the existing db).

While it’s possible to come up with a chatbot that can search the web for an answer to a user’s questions, and it’s possible for a chatbot to “learn” from a user’s inputs, merging these two features isn’t something that’s being done yet. Mainly because most AIML chatbot platforms (Pandorabots, pyAIML, etc.) don’t support direct access to web services such as Google, Wikipedia, or Wolfram Alpha. Program O does, via custom AIML tags, and has <google> and <wiki> tags (among others) built in, but Program O doesn’t properly support the <learn> and <eval> tags necessary for the chatbot to “learn” from it’s conversations. At some point, I’ll be addressing the <learn> and <eval> tags in Program O, but it won’t be any time soon, as I’m working on other portions of the code, but I’ll get to it at some point.

Just as a bit of an update here, I’ve decided to move the priority for properly supporting both <learn> and <eval> tags up, and begin working on it this week. After reviewing the “current” state of the code, and comparing it to the desired functionality based on the short paper on badanswer.aiml at alicebot.org, I found out that what I originally had coded worked exactly opposite to the intended functionality. Obviously this needs to change, so I’m working on the code to correct it. I should have something functional in a few days, given my current schedule. I’ll keep you all posted on my progress. Once I get that phase working as intended, I’ll work out some AIML code that utilizes Program O’s built-in <google> and <wiki> tags to be able to learn new information over the web to provide it’s users. That will probably take me a bit longer than a few days, but I’ll see how it goes.

Great news Dave! Inform us what you have done when you finish that.