ELIZA: a very basic Rogerian psychotherapist chatbot
TYPE HERE |
I first encountered ELIZA on the Tandy/Radio Shack computers that made up the first computer lab in the junior high school where I taught in the 1970s. By then, ELIZA was a software tween herself.
This early natural language processing program had been written in the mid-1960s at the MIT Artificial Intelligence Laboratory by Joseph Weizenbaum. It supposedly had been created to demonstrate how superficial human to computer communications was at that time. But, when it was put on personal computers, humans found it quite engaging.
WHAT DOES ELIZA DO?
Using "'pattern matching" and substitution methodology, the program gives canned responses that made early users feel they were talking to someone who understood their input. The program was limited by the scripts that were in the program. [ELIZA was originally written in MAD-Slip.] Many variations on
the original scripts were made as amateur coders played around with the fairly simple code.
Perhaps the most well known variation was called DOCTOR. This was made to respond like a Rogerian psychotherapist. In this instance, the therapist "reflects" on questions by turning the questions back at the patient.
ELIZA was one of the first chatterbots [later clipped to chatbot]. It was also an early test case for the Turing Test, a test of a machine's ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human. By today's standards ELIZA fails very quickly if you ask it a few complex questions.
That said, ELIZA delighted my students, and those who were in my little programming club at that time were also delighted to make their own versions by revising and adding to the code and scripts.
One particularly memorable moment was when a student asked "What happens if someone types in a curse word? "
I said, "She will respond based on some word you use, but she won't know those kinds of words."
They tried it out. Someone would tell ELIZA to do something obscene and she might answer "We were discussing you, not me," which is actually pretty clever and funny - by accident.
"Can we fix that?"
they asked.
"Sure. But you'll need to type in every possible curse word so that she knows what to respond to."
"We can do that?" they asked rather incredulously.
"You would have to if you want her to really respond logically, " I concluded. They were delighted and set to work immediately. Luckily, I never received any parent phone calls about it. Our little secret.
The version of ELIZA below is a more recent javascript incarnation to which I have made some cosmetic and scripresponset changes. She is still a baby chatbot, but she has had a 2018 resurgence of interest because she was featured in an episode of the TV show Young Sheldon. [The episode, "A Computer, a Plastic Pony, and a Case of Beer," may still be available at www.cbs.com] Sheldon and his family become quite enamored by ELIZA, though the precocious Sheldon quickly realizes it is a limited program.
Give ELIZA a try. You can sit on your own couch and pretend it is a therapist's couch. And, as with Siri, Alexa and other operating system disembodied voices, feel free to conjure up your own idea of what ELIZA looks like.
This javascript version of ELIZA was originally written by Michael Wallace and enhanced by George Dunlop.
- PDFView PDF
Abstract
While some computational models of intelligence test problems were proposed throughout the second half of the XXth century, in the first years of the XXIst century we have seen an increasing number of computer systems being able to score well on particular intelligence test tasks. However, despite this increasing trend there has been no general account of all these works in terms of how they relate to each other and what their real achievements are. Also, there is poor understanding about what intelligence tests measure in machines, whether they are useful to evaluate AI systems, whether they are really challenging problems, and whether they are useful to understand [human] intelligence. In this paper, we provide some insight on these issues, in the form of nine specific questions, by giving a comprehensive account of about thirty computer models, from the 1960s to nowadays, and their relationships, focussing on the range of intelligence test tasks they address, the purpose of the models, how general or specialised these models are, the AI techniques they use in each case, their comparison with human performance, and their evaluation of item difficulty. As a conclusion, these tests and the computer models attempting them show that AI is still lacking general techniques to deal with a variety of problems at the same time. Nonetheless, a renewed attention on these problems and a more careful understanding of what intelligence tests offer for AI may help build new bridges between psychometrics, cognitive science, and AI; and may motivate new kinds of problem repositories.
Keywords
Intelligence tests
Cognitive models
Artificial intelligence
Intelligence evaluation
Cited by [0]
Copyright © 2015 Elsevier B.V. All rights reserved.