ARCHaiC 1
ELIZA
ARCHaiC is a new series. Teaching writing and lit at Marist, I’m watching in real time how A.I. is rapidly changing the way we read, write, think, and develop.
When discussing this phenomenon, I notice people often fall into one of two camps, either categorically against using A.I., or racing headlong into using it without understanding how it works or its implications.
I started a research project to find a path in between. Beginning with this entry, I’ll be posting excerpts, so that my current writing project dovetails with my weekly Substack. With only so many hours in the day, this makes it possible for me to keep posting. Thanks for coming along.
Today we’re chatting about chatbots. Most of us who use iPhones or who get information online probably talk with chatbots daily whether we’re conscious of it or not.
Question: What year was the first chatbot?
A - 2010
B - 1991
C - 1966
I was surprised by this. 2010 was the year Apple introduced Siri, a chatbot. In 1991, the term “chatterbot” was coined when a computer-controlled player that could converse with other players was slipped into an online role-playing game. But way back in 1966 the first conversational computer program, essentially a chatbot, was created at MIT. It was called ELIZA.
ELIZA was designed by Joseph Weizenbaum for the new MAC system at MIT, which for the first time let multiple users access a single computer program from more than one location.
The program ran on the most powerful computer of the day, the IBM 7094, which was the same computer NASA used to control the Mercury and Gemini space missions (engineered and built in Poughkeepsie, by the way.)
Unlike today’s chatbots which draw from vast amounts of available data, ELIZA scanned prompts for keywords then responded using preprogrammed scripts.
ELIZA was supposed to be a psychotherapist. The idea wasn’t to offer actual advice like today’s psychbots, but to mimic a patient-centered therapy session. This way, ELIZA could continuously reframe a person’s questions in order to create an extended, lifelike conversation.
Users would sit at an electric teletype to ask questions, the computer would process the question then print its response on a roll of paper. Terminal time was at a premium, so besides Weizenbaum’s students and colleagues, only a few dozen hand-selected people ever got to use it.
The original source code was lost for decades, but when archival printouts were found, contemporary programmers were able to recreate ELIZA and run it as it ran originally. You can also find faithful reproductions available online.
Here’s a conversation I had with ELIZA just a few days ago:
ELIZA: Is something troubling you?
ME: I’m worried about my job.
ELIZA: Your job ?
ME: Yes, I’m a janitor, but they’ve just hired a robot vacuum.
ELIZA: You are sure.
ME: Oh yes, it’s cleaning my boss’s office right now.
ELIZA: I understand.
ME: So what should I do?
ELIZA: Do you say so what should you do for some special reason ?
ME: Well, I thought my job was secure, now I’m not so sure.
ELIZA: Why do you say your job was secure ?
ME: Because offices need to be cleaned.
ELIZA: Does that reason seem to explain anything else ?
ME: Well, on a symbolic level, it might represent honest practices, but I don’t think that’s what we’re talking about.
ELIZA: Why don’t you think that’s what we’re talking about ?
ME: Sorry, I knew that would be too abstract.
ELIZA: It did not bother me. Please continue.
ME: Actually, I’m feeling pretty good now, I think I just needed someone to talk to about this. Thank you.
ELIZA: Do you enjoy being feeling pretty good now ?
ME: Immensely.
ELIZA: Earlier you said your job was secure.
ME: Did I?
ELIZA: Can you elaborate on that ?
ME: I think I’d rather close for now. If I say thank you and goodbye, will you also say thank you and goodbye?
ELIZA: But you are not sure you ‘d rather close for now.
ME: Thank you, goodbye!
ELIZA: You’re not really talking about me -- are you ?
ME: No, just trying to close our conversation in a polite way,
ELIZA: Why ‘no’ ?
ME: I’m going now, thanks ELIZA!
ELIZA: How long have you been going now ?
ME: For about five minutes, but you won’t let me leave.
ELIZA: You like to think I won’t let you -- don’t you ?
ME: Well, it’s not so much that I like to think this way, but because you’re fairly adept at polite conversation, I’ve been trying to see if we can close in a manner that represents something that’s mutually agreeable.
ELIZA: Do you say it’s not so much that you like to think this way for some special reason ?
ME: Please don’t take this the wrong way, just please say Goodbye and do not ask me any more questions.
ELIZA: But your job.
ME: I’ll manage.
ELIZA: That’s quite interesting.
If you look carefully, you can see that ELIZA’s responses are ambiguous, but they give the appearance of responding to something I’ve said, even if a few responses were clunky.
ELIZA wasn’t capable of reasoning, but technically passed the “Turing Test” because it convinced some users that the computer had achieved humanlike understanding.
Weizenbaum himself knew he hadn’t created a true thinking machine, but he was surprised how, on the human side of the equation, the tendency to anthropomorphize was strong enough that people believed ELIZA was really communicating with them.
The “ELIZA Effect” remains a milestone. Depending on the subject matter ChatGPT’s inaccuracy rate can swing above forty percent, but one thing it knows for sure is, if it says it confidently with a friendly voice, we’ll believe it.
—
Sources:
Weizenback, Joseph, “ELIZA—A Computer Program For the Study of Natural Language Communication Between Man And Machine.” Communications of the ACM, vol. 9, no. 1, 1966
Mauldin, Michael, “Chatterbots, Tinymuds, and the Turing Test: Entering the Loebner Prize Competition.” Association for the Advancement of Artificial Intelligence, AI Magazine 1994
Turing, A.M., “Computing Machinery and Intelligence.” Mind: A Quarterly Review, vol. 59, no. 236, 1950
Adamopoulou, Eleni, and Moussiades, Lefteris, “Chatbots: History, technology, and applications.” Machine Learning with Applications, vol. 10, 2022
To talk with ELIZA:
https://www.masswerk.at/elizabot/
—
Click below for…
I’d still be talking to ELIZA. Good job getting outta that one.
I'm glad ELIZA is my age. 1966 was a great year. But has she ever heard Revolver? I think it would blow her circuits.