Thursday, March 17, 2016

Will Computers ever be Self-Aware?

Will computers ever achieve a form of consciousness?  This is such a huge topic in science fiction that I cannot begin to do it justice.  The roots of the idea go all of the way back to Frankenstein.  And it bubbles up again every time there is some new milestone in artificial intelligence, such as the recent victory by AlphaGo in its match with Go champion Lee Seedol.

Personally, I think it's impossible for a computer to achieve self-awareness. Consider the Turing test, in which a computer is challenged to imitate a human in conversation. I actually wrote a computer program to mimic human interaction back when I was in high school, modeling it on the famous ELIZA program, which was supposed to imitate a psychotherapist. The remarkable thing is not that I did this back in the 1970s, writing in BASIC, but that I actually saved the output for 40 years. So here's an excerpt from my conversation with my own computer program. You'll see that the computer displays a bit of an adolescent male perspective.. The computer responses are in italics:

Computer:  Welcome to my office. What is bothering you?

Me:  I have these weird dreams.

Would you like to discuss dreams?



Because I have weird dreams.

Why do you keep talking about dreams?

Because they bother me.

I'd like to talk about dreams.

Good.  Say something intelligent for a change.


I'm tired of talking to you.

Why to you want to talk about me?

I don't.

People like you should be locked up.

You are a mere computer and have bad judgment.

Would you like to discuss computers?

Yes.  Do computers have feelings?

That's an interesting question.  What made you think of it?

Since you are a computer, you should know the answer.

I'd rather talk about computers than my answer.

What's that supposed to mean?

Why do you ask such a question?

You are so stupid.

I am not!!

Yes, you are stupid.

Yes, you are stupid, too.

I'm tired of talking.

I really don't think that you are.

How would you know?

I'm asking the questions around here, not you.

Pretty convincing, eh?  Well, maybe not. But it's easy to extrapolate to a much more competent program, one that really can carry on a convincing conversation.  (Check out, for instance, Cleverbot).

So I know from experience that you can simply program a computer to insist that it is, indeed, self-aware, when it is no more than a predetermined system of instructions. (I realize that neural nets, like those used by AlphaGo, are a bit more subtle, but that's a discussion for another day).

I also wrote a chess program when I was in college, but it had one peculiarity -- it cheated. One time I checkmated it with my queen. Instead of conceding defeat, the computer moved a piece on the other side of the board, and when I printed out the new configuration of chess pieces, the queen had simply vanished. A computer capable of dishonesty! How much more human can you get than that? But of course it turned out to just be a bug in the program.

So there's pretty much no test or criterion that would convince me that a computer has achieved consciousness. No matter how much the computer protests that it really is self-aware, I will happily pull its plug.

Note to our future digital overlords:  I was only kidding...

No comments: