Last Updated on November 10, 2020 by Sean B
Created by Joseph Weizenbaum at the MIT Computerized Reasoning Laboratory in 1964 – 1966, the ELIZA chatbot was not only the first chatbot, she is still one of the most popular chatbots to have ever been created. ELIZA used a number of different programs or “scripts” to help manage her interactions with users.
The ELIZA Chatbot was named after Eliza Doolittle, a character from a parody called Pygmalion written by George Bernard Shaw in 1912, and the musical adaptation My Fair Lady. In the play, Eliza is a Cockney flower girl who is transformed by Professor Henry Higgins through a variety of lessons to correct the way she speaks and acts. The result is Eliza’s transformation into a proper young lady. Weizenbaum saw his creations ability to be “incrementally improved” with each interaction as similar to the transformation that Eliza Doolittle underwent.
The ELIZA Chatbots Most Famous Script – The Doctor Is In
In the most famous script of the ELIZA chatbots, called DOCTOR, she imitated the famous Rogerian psychotherapist Carl Rogers, who was known for responding to his patient’s questions by simply repeating the question back to them in another way. This was an ingenious way to overcome ELIZA’s shortcomings, as it forced the user to actually direct the conversation.
The ELIZA chatbot used a series of rules in the script to respond to user inputs with open-ended questions. Weizenbaum stated that his goal for ELIZA but using the DOCTOR script was to parody “the responses of a non-directional psychotherapist in an initial psychiatric interview.”
Despite the shortcomings of early artificial intelligence, many users who interacted with ELIZA believed they were interacting with another Human. This included Weizenbaum’s own Secretary, who reportedly asked him to leave the room one time so she and the ELIZA chatbot could have a real conversation.
How the ELIZA Chatbot Worked
Chatbot ELIZA simulated conversation with the methods of pattern matching and substitution, giving those interacting with her the impression that she understood what they were talking about. In reality, the chatbot had no way of understanding the context of the discussion and was simply repeating back the important parts of speech. She was simply pulling out the important parts of the user’s statement and using it to create a response.
ELIZA was written in MAD-Slip, an early natural language processing language created by Weizenbaum that helped her parse the user input and pull out the most important words within it. In creating MAD-Slip, Weizenbaum had five problems that he was attempting to overcome with regard to language processing:
- The identification of critical words
- The discovery of minimal context
- The choice of appropriate transformations
- The generation of appropriate responses
- A way to end the interaction
To solve this, the ELIZA chatbot, using MAD-Slip, first looked for keywords in the text input by the user and ranked them with an algorithm Weizenbaum created. ELIZA then uses the words with the highest rank, or what she feels are the most important keywords, to find a rule to transform the input to create a response.
As an example, I had a conversation where I told ELIZA, “I want to quit my job.” Here’s how it went.
ELIZA: Can you tell me what is going on in you right now ?
YOU: I want to quit my job.
ELIZA: What if you don’t get to quit your job ?
YOU: Why wouldn’t I get to quit my job?
ELIZA: I do not answer questions, sorry. But, what do you think yourself ?
YOU: Maybe my finances won’t permit it.
ELIZA: Is there a way to be sure ?
YOU: I could create a budget and see if I can manage without the job.
ELIZA: That is interesting. Please continue.
You can see one of the major issues with ELIZA in this simple conversation. When I asked a direct question instead of making a statement, she didn’t know how to respond, so she forced me to respond with something more understandable. While the ELIZA chatbot did do a decent job of redirecting me to address the question myself it failed to provide an answer that any person would have easily come up with. The problem for ELIZA was that she had no understanding of what a “job” was or what “quitting” meant.
Some simpler questions have been programmed into ELIZA. When I asked her, “Do you like me?” She responded with, “You like to think I like you – don’t you?” On another occasion, I received the following response to the same question, “We were discussing you, not me.”
The ELIZA Chatbot was the earliest chatbot, developed in 1964 – 66 by Joseph Weizenbaum, during his time at MIT. Weizenbaum also developed the language she was programmed with, MAD-Slip. While she appears simplistic next to the chatbots of today, ELIZA was a huge leap forward in the interactions between humans and machines.
If you’re interested in learning more, read our full article What is the History of Chatbots?