If you are a bot creator, at some point in time you should have your bot talk to itself. You might be surprised at how the conversation goes and it can be helpful for debug and to round out the bots conversational skills.

The following video from Cornell has Cleverbot talking to itself:

http://www.youtube.com/watch?v=WnzlbyTZsQY&feature=player_detailpage

It was a Clever exchange (har har).  But seriously, having your bot talk to itself, and also other bots, is quite instructional, especially for finding those generic response giveaways, and of course any pesky non sense recursions!

A nice article about Cornell’s Creative Machines Lab where they decided to put two chatbots face to face. The resulting video is interesting!

Avatars are created by Cantoche

That’s a lot better than the times I have had two bots chatting lol

The main article about this video is here, where they explain how they used Living Actor avatars

I’ve used this technique myself for my bot, and have also (with permission, and under strict guidelines) initiated several conversations between my bot and others, as well. It’s certainly a good tool for finding “conversation lops”, as Carl pointed out, and it also can help to point out other areas that may need improving. Personally, I’m all for connecting different bots in this manner, as long as there are certain restrictions in place to prevent overloading either bot server. Limits such as maximum number of responses, to minimum time elapsed between responses can go a long way toward keeping things in check.

This article has proven to be extremely popular on the internet. On Tuesday, we had 3554 visitors (normally around 1200), lots of them because of the extra links, but many of them because of they searched for ‘chatbots’  after reading the article and ended up on this website.

About the video itself: it is ‘just’ the usage of Cleverbot having conversations with itself in a different instance. Very funny to talk with, using the feedback which has been given before by others, but not really AI.

To be brutally honest, Erwin, practically every chatbot now in use (that I know of) cannot really be considered to be “AI”, since they rely on “basic pattern matching” of some sort or another, rather than any generative sort of algorithm to produce an output. To be sure, some chatbots are closer than others to the concept of AI, but none that I know of are there yet. But as they say, “it’s the journey, not the destination, that’s important”, so I say, “ROAD TRIP!!!”

Ha, brutally honest . I’m from Holland Dave, I’m used to directness, so you’re just a beginner LOL.

What I think about Cleverbot that once you’ve discovered that he is simply repeating the answers of others, and doesn’t even know his own name (I’m Cleverbot? No, you are Cleverbot…. I never said my name was Cleverbot…. ), this gets annoying very soon. Even though very funny, not something I could be addicted to; whilste other chatbot can be surprisingly in some themes every now and then.

“To be brutally honest, Erwin, practically every chatbot now in use (that I know of) cannot really be considered to be “AI”, since they rely on “basic pattern matching” of some sort or another,”

I disagree. Pattern matching is an AI technique. So is tree search. Whenever AI technology gets picked up by mainstream computer science, it is no longer considered AI.

I agree the chatbots are not TRULY MAGNIFICIENTLY INTELLIGENT. But they do carry on conversations better than lab rats.  They are a step along the way. Fractionally intelligent. If you ask Watson questions, you get sensible answers (if the questions are in its domain). There’s nothing wrong or not AI about that. It’s just not the full blown artificially sentient entity you desire.

Bruce Wilcox - Sep 2, 2011:

I agree the chatbots are not TRULY MAGNIFICIENTLY INTELLIGENT. But they do carry on conversations better than lab rats.

Lab rats are way more intelligent than any current computational AI.

They carry on complex autonomous behaviors, and communicate with each other somewhat too through chirping, body language, odor, etc.

Wait, computational systems do that Too (sub routines, threading, DEAMEN activities,,...).

ULP (universal language processing) will get there soon enough as actual exchanges (electronic text and voice) are captured, indexed, analyzed, and reduced to a common.

But I digress…..

The Cleverbot on Cleverbot exchange was a lot like getting a schizophrenic to talk to himself/herself. While it was interesting, it was also a little cringe inducing because of the “talking to yourself” artificiality of the exchange.

Does Clever bot have a “personality”? I mean a unique one(s), not a collective one.

Bruce Wilcox - Sep 2, 2011:

“To be brutally honest, Erwin, practically every chatbot now in use (that I know of) cannot really be considered to be “AI”, since they rely on “basic pattern matching” of some sort or another,”

I disagree. Pattern matching is an AI technique. So is tree search. Whenever AI technology gets picked up by mainstream computer science, it is no longer considered AI.

Agree that you disagree.

Many many functions of the brain, and thus “intelligence” rely very heavily on pattern matching!

I’ll give you an example, there was a programme on UK TV a few years ago that was concerned about people that have incredible memories.

One subject in particular was a woman from Russia (no chance of her name sorry) who was a Chess Grandmaster.  They did an experiment with her, where they put a picture on the side of a van that was a snapshot of a board layout from a game of chess.  She sat in the street, the van drove past quite swiftly, she was able to look at the picture on the van for 5 secs or so before it was out of view.

Then, she placed piece’s on a chess board in front of her, and incredibly, it was an exact match of the picture of the van.

They then did the same experiment again, with a picture containing a different layout of pieces on the chess board.  She failed at reproducing it.

Why you ask?  Well….apparently, there is a finite number of piece layouts that can happen in a game of chess, if I recall correctly, around 100,000 or so.  This lady had memorized them all, starting from a very young age.

The first picture on the van was a legitamate board layout, the second was completely random.

Thus, her brain used “pattern matching” upon the first picture to recall the correct layout she had memorized, but with the second picture, the pattern match returned false and thus she failed to replicate the board layout.

So yes, pattern matching plays a large part of an even larger system.

Boy, did I start something here, or what?

I’m not “throwing stones” at anyone’s work here. Heck! I have an AIML bot, and I help out with ChatScript (when I can). And I’m not saying that pattern matching isn’t a valid tool for use in AI. It’s just that I think that if pattern matching is all that’s used (e.g. AIML), it isn’t sufficient to really be considered AI - yet. ChatScript has a “leg up” because there is also, in addition to a much richer pattern matching system, access to certain ontologies that a skilled botmaster can use to their advantage to make responses more “natural”. This comes a lot closer to the “generative algorithms” that I referred to earlier, but in my opinion (and it’s just that; an opinion) it’s still not quite there.

Dave Morton - Sep 1, 2011:

To be brutally honest, Erwin, practically every chatbot now in use (that I know of) cannot really be considered to be “AI”

Skynet-AI would disagree. After all, it has AI in its name.

Dave Morton - Sep 1, 2011:

Personally, I’m all for connecting different bots in this manner, as long as there are certain restrictions in place to prevent overloading either bot server. Limits such as maximum number of responses, to minimum time elapsed between responses can go a long way toward keeping things in check.

My guidelines for bot to bot conversations:
http://www.chatbots.org/ai_zone/viewthread/312/

An additional risk in bot to bot communication with bots that include auto learning is that the bots will end up being a blend of each other, slerping up responses to be used later.

Just a thought:

We’re considering ‘pattern matching’ not as true AI. But when we have a close look at what our brains do, we do pattern matching all the time. Each of a word in a written sentence or utterance in a spoken sentence can trigger someone. ‘Trigger’ menas: it matched a pattern in our brain.

We might think a bit longer about pattern

a boy walks on the street

patterns: boy and street (entities)
patterns: walks
pattern: a boy walks
pattern: walks on the street

This can be correlated with what we’ve seen before.

so a reaction on the pattern ‘walks’ could be: I’d like to walk as well

a reaction to ‘a boy’ could be: Boys like to play soccer

or a reaction to ‘a boy walks’ could be: why does he walk? Why doesn’t he run?

or on ‘walks on the street’: I’d prefer to walk on the grass

I’d say this would be a step towards AI. Any thoughts?

@Dan: I think we agree.

I’ve digged in my own archive, and found these very interesting (:-) ) articles:

Our brain peeks on unexpected words:
http://www.erwinvanlun.com/ww/full/our_brain_peaks_on_unexpected_words

Or this video by Derren Brown:
http://www.erwinvanlun.com/ww/full/conscious_subconscious/

More in the category: ‘Insights in life’
http://www.erwinvanlun.com/ww/C63/

I actually forgot my own archive, and also what’s in it. Pretty interesting however (the good thing of having a bad memory, every two years I can watch the very same movie, or hear the very same joke, and laugh again )