http://slurl.com/secondlife/Eduserv%20Island/140/127/26
http://slurl.com/secondlife/Buena%20Vista/146/191/28
Based on ALICE, an award winning chatterbot.
I think the current stat of all chatterbots (not just SL ones) is just sad. They have no feel whatsoever for the flow of the conversation.
4 comments:
Making a chat program to imitate human conversation is incredibly difficult. There is a test known as the Turing test which a human judge chats with both another human and a machine. The machine passes the Turing test if the judge cannot determine which of the participants is not human.
This test was described in a paper in 1950, yet to date no machine has ever passed the test. It is doubtful imo that they will any time soon.
The ALICE program has actually won awards for best performance in Turing competitions.
For anyone interested:
http://en.wikipedia.org/wiki/Turing_test
The reason I made this post is that i have been experimenting with Markov chain techniques to come up with a chatterbot for our library.
I don't strongly believe in the validity of the test. The Turing Test was designed as a method to test whether a machine is intelligent or not, however I don't think that being able to chat intelligently is a sufficient and necessary condition for intelligence.
The test has since transformed into a test of the programmers ability to fool the tester into thinking that a bot is a human.
Take a look at http://en.wikipedia.org/wiki/Mark_V_Shaney which uses Markov chaining, and which by the way fooled many users when it was first released.
I also think that with the Internet slowly moving towards structural representation of knowledge it might be possible to feed a bot a more complete knowledge of language so that we can make it into an intelligent chatter.
To paraphrase former president Clinton, Mike: it all depends on what chat is.
There's evidence to suggest that there are A.I. algorithms out there that have beaten the Turing Test. (http://news.bbc.co.uk/2/hi/uk_news/magazine/3503465.stm)
Mind you, this is done via SMS and used for spamming purposes, but it still shows that the Turing Test may not be valid anymore.
The Turing Test was designed at a time where we, as researchers, were very naive as to what constituted A.I. and really believed that by the 70's, we'd be talking to computers like HAL in 2001: A Space Odyssey. However, we then realized that: a) it is very difficult to create a true artificial intelligence and b) voice recognition is still a difficult problem to tackle.
However, in reply to the original post, I agree that the current state of chat bots is sad not because there's been a lack of progress in A.Is. Far from it. Rather, I believe that we've just not been that interested in communicating with computers ever since the Direct Interaction paradigm came along (using the mouse to drag and drop objects on the screen), as that gave us more control of how we interact with our computers than we would have if we just verbalized the same commands.
Now, that's not to say that chat bots are relics of an idea long since past. However, with the shift of attention to other interaction techniques, the momentum to create that perfect A.I. chat box is just not there.
Nicolas, It is not just that our definition of what constitutes AI was, back then, incomplete, or immature, but our approach to AI has changed significantly since then. For example the distinction between soft and hard AI emerged soon after the initial attempts to get a HAL like computer. Researchers soon realized that it is more important to put their efforts into soft AI problems such as optical character recognition, visual recognition, speech recognition, machine learning... etc rather than on making a computer smart, or try to pass the Turing Test.
I bet that if somebody 5 or 6 years ago had a look at the state of the art OCR back then he would not have predicted that the current state of OCR as a realistic target.
If the same effort that was put into OCR was put into passing the Turing Test, I don't think we would be talking about it today. It would be a memory from the past.
Finally, I am a strong believer in the command line over any other computer interaction methods, as a matter of fact I believe that if god exists he must be using the command line to run the universe. However, I absolutely agree with you that direct interaction paradigm made us less interested in other communication techniques, much less in verbal communication techniques.
Post a Comment