ELIZA and artificial intelligence
Submitted by T-Bo on Fri, 02/22/2008 - 11:05.
I found it really interesting how some users of ELIZA actually believed it to be a real person, despite the technological limitations of its programming and the fact that its users were INFORMED of the fact of its artificiality. I remember using ELIZA one time in the Orlando Science Center, and I found it comprehensively limited (and pretty funny as well). I've heard that there are additional, more sophisticated interpreters exist currently, but I haven't tried them myself. In any case, the fact that the creator eventually defected to the cause of warning against artificial intelligence is a very intriguing concept--a concept that brings to mind ideas of old science fiction movies, such as "The Terminator", "A.I.", and "2001: A Space Odyssey". I, however, personally consider it impossible for artificial intelligence to reach the complexities of the human psyche, as it would be impossible to engineer a program to produce such emulation without an adequate understanding of the inner workings of the human brain. Scientists have made large leaps of knowledge into this field, but because the brain is so complex, I find it difficult to believe that humans could understand their own minds well enough to create a completely believable proxy. Science fiction would say otherwise, but science fiction is FICTION, and thus I don't find the prospects of "human" artificial intelligence a believable prospect. Perhaps I'll be proven wrong, however, in due time. Only the future can tell.
- T-Bo's blog
- Login to post comments
Inner workings?
Why do you think that's a criteria for assessing the believability of an interaction? I mean, I don't know the inner workings of the human brain, or even the complexities of the human psyche. That is, I "know" it from having been a human, but I don't think I rely on my knowledge of neurology to decide whether or not I'm communicating with a human as I type this comment to you.
And I don't mean to be glib. My point is just that we take our human-ness for granted in a way that we don't when we interface with a chatterbot. So what happens when we apply that assumption to chatterbots? In other words, what if I showed you ELIZA and explained that it was a real psychiatrist, a colleague of mine, interacting with you and answering your questions?
Interesting point
I suppose it is not so much the structure of the brain as it is the structure of the language systems within the brain that I am talking about.
Given that the human speaking to the chatterbot follows a specific set of guidelines with respect to what s/he inputs, perhaps a simple sentence parser can produce a conversation that seems as though it were between two sentients. I considered the program accounting for all possible statements or questions a criteria for a believable interaction simply because we, as human beings, process language through a complex interpretative system that can process a wide array of possible statements and each of the many multiple forms of expressing the same meaning. I found ELIZA to not be believable because she could not understand some statements that most human English-speakers could, or she responded in a way that simply didn't make sense.
Because we do not fully understand how the language systems of our brains interpret and process incoming and outgoing statements, I consider it very, very difficult to produce a chatterbox program that would respond in a "human" (that is, comprehending anything another human could possibly say) way; not only would the programmer have to understand the enigmatic rules of our own language processing systems (which, as far as I know, is still a mystery to everyone), but the programmer would also have to take careful, measured steps to codify them into programming language as rules of the program itself. The program that would result would be intensely complex and would require a large project team and a sizeable budget and allocation of time to complete.
I do understand your point, however. Simply knowing that the program we're "talking to" is only a program already puts us in a different mindset than the one we would take if we thought it was a person, and I do admire the capabilities the program DOES have to interpret language entered into the text bar. If you were to show me ELIZA and tell me that it was a real psychiatrist talking to me, I would probably consider her more believable simply on that basis, but I would likely consider her to have an interesting collection of grammatical and syntax comprehension problems (for a human being anyway), or that she was not a native English speaker and knew a fair deal of the language, but did not understand some of its nuances.
I hope this makes sense and answers your question. If not, I can try and see if we can't talk about it on Monday.