Artificial Intelligence - Eliza

I mentioned the "Animals" program in my last two postings. It is an example of an Artificial Intelligence (AI) program in that it "learns". It only knows about two animals when it starts but it you can teach it about more until it becomes quite impressive (at least if you feed it good information).

I want to talk about two other AI programs.

In this posting I will deal with Eliza, which was written in 1966 by Joseph Weizenbaumin MIT.

This program responds to sentences entered by the user by identifying keywords and manipulating the input string. It pretends to be a Rogerian psychotherapist. Apparently it has fooled some people into thinking that they were exchanging messages with a real person.

You can try out a version of the program (and see some Java source code) here.

Here is a dialogue I had with this program (my sentences are preceded by ">>"):

>> Hello.
How do you do.  Please state your problem.
>> I am feeling a little down
Is it because you are feeling a little down  that you came to me ?
>> yes I am hoping that you can help me
You seem to be quite positive.
>> are you a psychologist?
Why are you interested in whether I am a psychologist  or not ?
>> I am looking for someone who can help me
How long have you been looking for someone who can help you  ?
>> Do you answer all questions with questions?
We were discussing you -- not me.
>> I am not sure that you are real
Do you believe it is normal to be not sure that I are real  ?
>> Are you a machine or a person?
Do computers worry you ?
>> No I use computers in my work
Why do you mention computers ?
>> you mentioned them first
Oh, I mentioned them first  ?
>> I am not convinced that you are real
Do you enjoy being not convinced that I are real  ?

While most of the responses are pretty convincing, a few are not (such as the last one).

The thing about Eliza (aside from how clever it seems to be) is that by almost fooling people it is almost passing the Turning Test.

In 1950 Alan Turing (an English mathematician, logician, cryptanalyst, and computer scientist) proposed that a machine would have exhibited intelligent behaviour if a judge who was interacting with the machine and a person at the same time via written messages was unable to determine which was which.

While Eliza is interesting in that it simulates human behaviour, it is less interesting than Animals in that it is unable to learn. Its conversation skills do not improve with time and practice.

In my next posting I will talk about another "learning" program.