Andres Hohendahl - Apr 10, 2012:

¿why not implement an AI child-bot?

A bot that resembles a learning child, with no knowledge at all, it shold have a limited vocabulary but a immense curiosity and some good learning capabilities (even on vocabulary), and the voting should be as the smarter one and the rated human-equivalent age it implements!

This is a cool idea. It shifts the testing focus from an interrogator exposing a bot’s lack of common knowledge and intuition, to instead a teacher testing a bot’s ability to absorb new knowledge and use it. I’d be curious to see how “human” people would rate a bot impersonating a 5-8 year old.

C R Hunt - Apr 10, 2012:

Andres Hohendahl - Apr 10, 2012:

¿why not implement an AI child-bot?

A bot that resembles a learning child, with no knowledge at all, it shold have a limited vocabulary but a immense curiosity and some good learning capabilities (even on vocabulary), and the voting should be as the smarter one and the rated human-equivalent age it implements!

This is a cool idea. It shifts the testing focus from an interrogator exposing a bot’s lack of common knowledge and intuition, to instead a teacher testing a bot’s ability to absorb new knowledge and use it. I’d be curious to see how “human” people would rate a bot impersonating a 5-8 year old.

The argument against self-learning bots has always been that you’d end up with a poor spelling bot that frequently employed profanity.

I think that if you cared about your bot, it would require a lot of monitoring and correcting, which sort of defeats the whole purpose.  I’m usually trying to find ways of producing a single reply that answers several questions.  Having a bot that would require a lot of oversight would be like having a job without any pay or benefits.

Wow! Step away for a day, and I have tons to comment on!

@Andres:

I’ve certainly read and considered your comments, but what you propose is several levels of sophistication above the average chatbot in use today, and would therefor not be suitable for a near-future contest. I think that what you’ve outlined would make for a great overall goal for chatbots of a few years down the road, but for today, the expectations are just plain too high right now.

@CR:

I like the idea of creating a “child-like” chatbot, as well, but with everything else on my plate, there’s no way I could even collaborate on such a project, let alone create one. I certainly hope that someone can build one, though.

@Thunder:

If you let the general public loose on a self-learning chatbot, I agree that it’s overall output quality would greatly suffer. However, if you only allowed certain trusted members of a community of dedicated chatbot/AI enthusiasts to perform the task of teaching said chatbot, the output quality of such a chatbot would be potentially unsurpassed. I’ve never been much of a fan of a wide scale “wiki” approach to teaching a bot, but I do like the idea of assigning “teachers” to do the job. Just as a child might go to school to learn various subjects, so might a chatbot benefit from the same regimen.

Also, just a thought here… Having a Human child requires a lot of “oversight”, too. And, strangely enough, also can feel like a full time job, with no pay and/or benefits. Ok, maybe not “pay”, per se, but certainly there are benefits.

Dave Morton - Apr 11, 2012:

If you let the general public loose on a self-learning chatbot, I agree that it’s overall output quality would greatly suffer. However, if you only allowed certain trusted members of a community of dedicated chatbot/AI enthusiasts to perform the task of teaching said chatbot, the output quality of such a chatbot would be potentially unsurpassed. I’ve never been much of a fan of a wide scale “wiki” approach to teaching a bot, but I do like the idea of assigning “teachers” to do the job. Just as a child might go to school to learn various subjects, so might a chatbot benefit from the same regimen.

I’d agree in principle.

However… We once had a state legislator who thought the mandatory seat-belt law was such a good idea that he’d extend it to include motorcycles.  But, most can imagine, as your vehicle is headed for the guardrail and down over the hill, the first thing you’d want to do is step off.  It’s one of those ideas that looks good on paper, but impractical when you try to implement it.

I’ve seen efforts like this come and then dissolve into nothing a couple of times.  At the AI Nexus, when a member proposed it, I created a section devoted to his idea, “The Giant Chatbot Collaboration Project”.
http://knytetrypper.proboards.com/index.cgi?board=gd&action=display&thread=2032

Gathering a functioning group of independent botmasters/developers used to working alone on their own projects is a bit like herding cats.

Curiously, I always felt that was what ALICE was about—they already have all of our update.aiml files.  I thought the intention might have been to comb through the files for the best ideas and then apply them to a sort of super-duper bot, but ALICE hasn’t evolved much.