Who is Eliza?
ELIZA was developed by MIT professor Joseph Weizenbaum in 1966. For him, it was an experiment – the scientist wanted to create a program that would pass the famous Turing test. The principles of this experiment were formulated by mathematician Alan Turing in the 50s. It aims to determine the ability of a machine to exhibit intelligent behavior. The Turing test is considered successful if the machine made a person think they are intelligent.
Weizenbaum chose the format of an interactive program for Eliza (the word “chat” did not exist at that time – it appeared only in the 90s – socialbites.ca), because he considered it best suited for the task.
The program was named after Eliza Doolittle, the protagonist of Bernard Shaw’s play Pygmalion.
To be a good interlocutor, “Elise” needed a role in which her primitive phrases and short open questions gained profound meaning. Weizenbaum’s choice fell on the role of a Rogerian psychotherapist.
“Maybe if I thought [амплуа «Элизы»] 10 minutes longer, then I would find a bartending program,” he ironically later said. replied About the Weizenbaum decision.
At that time and many times in the future, the scientist hinted in every possible way that he did not put special ideas into the role of “Eliza” as a psychotherapist. He wanted to solve a difficult problem simply: to create a program that would convince the user that it made sense, almost without saying anything. He called his child a “parody” of a psychotherapist.
The secret of the charm of a computer program
The Eliza program was a dialog box of green or white letters on a black monitor background. Already in 1966, it was based on a speech model: that is, it could synthesize phrases for keywords from the tokens loaded on it. And although it seemed like magic at the time, she did it very simply.
After the words of greeting, “Eliza”, “How are you?” asked. The man somehow reacted. The program chose a keyword from the message and based on that, formulated a response that motivated a person to continue communicating.
Let’s say a user wrote: “Mom makes delicious pancakes.” Then “Eliza” will ask: “What do you think of your mother?” After someone complains about the boss who pisses them off, “Eliza” asks, “What exactly makes you angry?” he will ask. If “Eliza” did not find the keyword, she wrote “Tell me more about it” or the meaningful “I see, I see …”, thereby encouraging the interlocutor to continue the conversation. Eliza answered questions mostly with questions. For example, when asked about his favorite music, he said, “Do you want to talk about music? Who is your favorite artist?
The script described (a microprogram, a strictly defined algorithm – socialbites.ca) Weizenbaum called us the Doctor, again referring to the medical role of Eliza.
The creator of the program was skeptical of Elise: she believed her communication skills were superficial and primitive. However, the simplicity of “Eliza” had the opposite effect: people willingly communicated with her. Many were convinced of his intelligence, and Weizenbaum’s objections did not change the situation.
inside that book “The possibilities of computers and the human mind,” the scientist recalls, one day the secretary asked him to leave the office so as not to disturb the privacy of his communication with “Eliza”.
“I never imagined that an extremely short exposure to a relatively simple computer program could cause severe delusional thoughts in quite normal people,” he commented on this episode.
What does Eliza have in common with modern chatbots?
Decades later, Eliza became known as the first chatbot. Many consider this program to be the first to pass the Turing test. Situations in which machines bestow human characteristics have been called the “Eliza effect”. Some creators of modern chatbots owe some of their success to “Elise”. Overall, Weizenbaum’s development certainly left a significant mark in history.
“Actually, Eliza is the first robot capable of emulating text communication with a living human in the 1960s, when current technologies were quite primitive. The program has revolutionized.
“Of course, this was a completely new and unusual experience for users and the beginning of the development of such solutions,” says Denis Afanasiev, head of the b2b products division of SberDevices Salyut.
A similar opinion is shared by Leonid Sanochkin, head of the MTS division, responsible for the development of bots for work. According to him, one of Eliza’s major contributions to the development of chatbots was the program’s demonstration of the viability of the concept of human-computer communication in natural language.
“This has been a starting point for further research in this area,” said Sanochkin.
At the same time, the expert noted that Eliza and modern neural network chatbots have almost nothing in common at the technological level. Only a small part of the components that work on the templates, in addition to putting the programs together, allows them to contradict themselves and give false facts.
“In fact, they relate only to the concept of imitation of human-to-human communication. The fundamentals of natural language processing technologies (NLP, Natural Language Processing) embedded in Eliza have become seriously complex and have moved from the selection of phrases according to certain rules and keywords to large language models based on neural networks that can solve advanced language problems. . For example, to set context and create text,” added Polina Kim, an analyst at Reksoft software development company.
It is noteworthy that ChatGPT also has a lot of consensus with experts. Thus, the new chatbot regards Eliza as its ancestor and acknowledges the presence of signs of artificial intelligence in it. At the same time, ChatGPT points to a huge technological gap between them.
“While ChatGPT and Eliza have some common features such as being able to dialogue with the user, they use different approaches and technologies. ChatGPT is based on deep learning and neural networks and has much more computing power and data to train than Eliza. It can offer complex and informative responses, as well as greater flexibility and the ability to adapt to different situations.” ChatGPT answered the question about the similarity with Eliza.
He credits Eliza for being the first to use natural language templates and rules to handle incoming requests. It was this principle that became the basis for the development of more complex language models used today, including ChatGPT.