SHE IS CALLED ELIZA. As recently as 1966, scientists created a chatbot with artificial intelligence
Eliza was developed in 1966 by MIT professor Joseph Wetzenbaum. He chose to use a conversational program as Elisa's format because he thought it was the most appropriate way to achieve his goals.
In the 50s, mathematician Alan Turing proposed the principle of the Turing test, which aims to determine the ability of a machine to exhibit intelligent behavior. If the machine can make people think that it thinks reasonably, then the Turing test is considered passed.
Wetzenbaum chose the format of the dialogue program to develop Elisa because he thought it was the most appropriate way to get the job done. There was no such thing as a "chatbot" at the time, it only appeared in the 90s.
Wetzenbaum named his program "Elisa", a name derived from the character Eliza Doolittle in George Bernard Shaw's play Pygmalion. The goal of development was to create a program that would simulate a simple conversation and exhibit the appearance of understanding and empathy.
Alyssa uses simple pattern matching to respond to user input. For example, when a user says, "I'm feeling sad," "Elisa" might respond "Why are you feeling sad?" ", providing emotional support by simulating conversations.
Despite its relatively simple functionality, "Elisa" has attracted widespread interest and is seen as an important milestone in the field of machine intelligence. It laid the foundation for the later development of chatbots and intelligent dialogue systems.
The name comes from Bernard Shaw's play Pygmalion, in which the character Eliza Doolittle is a florist who, under the guidance of a phonetics professor, transforms into a social elite within a few months.
In order to be a good interlocutor, "Elisa" needs to play a role that gives deep meaning to her simple words and brief open-ended questions. Wetzenbaum chose the role of a Rogersian psychotherapist.
Rogersian psychotherapy is a method that may give the impression that the therapist is not doing anything: He asks leading questions, asks the patient to talk more about what they are worried about, and sometimes nods his head in understanding. The purpose of this treatment is to allow people to analyze their own condition on their own.
"Perhaps, if I had thought about this ('Elisa's character) for ten more minutes, I might have come up with a bartender program," Wetzenbaum later commented sarcastically.
Then since then, the scientist has repeatedly stated that he did not envision the role of "Elisa" as a psychotherapist and had no particular ideas. He simply wanted to solve a complex problem: create a program that barely speaks but can convince users that it is sane. He calls his work "a psychotherapist's spoof."
The fascination of computer programs
The "Elisa" program is presented in the form of a dialog box with green or white letters on a black screen. Back in 1966, it was based on a language model: that is, it could synthesize sentences entered by the user based on keywords. Although the process was very simple, it was considered a miracle at the time.
"Elisa" will begin with a greeting: "Hello, how's it going?" "。 The user responds in some way. The program extracts the keywords in the message and responds based on them, motivating the user to continue the conversation.
Let's say the user writes, "My mom makes pancakes." Well, "Elisa" will ask: "How do you feel about your mother?" "。 When a user complains about their angry boss, "Elisa" asks, "What makes you angry?" "。 If "Elisa" can't find the keyword, it will write: "Tell me more about this", or ambiguously write: "Got it, got it..." to trigger further development of the conversation. For the question of "Elisa", people mainly answer with questions. For example, if asked about music she likes, she will rhetorically: "Do you want to talk about music?" Who is your favorite artist? "
Wetzenbaum's call of this script (microprogram, strictly defined algorithm) is a "doctor," which again brings us back to the medical identity of "Elisa."
The creators of the program were skeptical of "Eliza": he considered her communication skills superficial and simple. However, the simplicity of "Elisa" had the opposite effect: people were happy to communicate with her. Many believe she is intelligent, and Wetzenbaum's dissent hasn't changed that.
In his book, The Power of Computers and Human Intelligence, Wetzenbaum recalls a time when a secretary asked him to leave the office so as not to disturb her private conversation with "Elisa."
"I can't imagine a relatively simple computer program capable of generating such strong delusional thinking about perfectly normal people," he commented on the incident.
What "Elisa" has in common with modern chatbots
Decades later, "Elisa" was identified as the first chatbot. Many people consider this program to be the first to pass the Turing test. The condition in which machines have human qualities is known as the "Elisa effect." Some of the creators of modern chatbots owe it in part to the achievements of "Elisa". In general, Wetzenbaum's development work has left an important mark in history.
Of course, "Elisa" offers users a completely new and unusual experience, laying the groundwork for the development of similar solutions, said Denis Afanashev, head of Salut's B2B product division at SberDevices.
Leonid Thanochin, head of MTS's department responsible for developing enterprise robots, holds a similar view. He said that "Elisa" has made an important contribution to the development of chatbots because the program demonstrates the feasibility of human-computer communication in natural language.
"This became a starting point for further research in the field," says Sanochin.
At the same time, experts note that on a technical level, "Elisa" has little in common with modern neural network chatbots. The only thing these programs have in common is a small set of template-based components and the ability to generate contradictions and publish false facts.
"In fact, the only thing they have in common is a simulated 'human-to-human' concept of communication. The foundations of natural language processing (NLP) technology laid in Elisa have become quite complex on the basis of convention rules and keyword matching, and have evolved into large language models based on neural networks that are capable of solving more complex language tasks. For example, understanding context and generating text", adds Polina King, an analyst at the software development company "Rexoft".
It's worth noting that ChatGPT is very much in line with the experts. The new chatbot sees "Elisa" as its predecessor and acknowledges the presence of features of artificial intelligence. At the same time, ChatGPT points out that there is a huge technological gap between them.
"While ChatGPT and 'Elisa' have some things in common, such as the ability to have conversations with users, they employ different approaches and technologies. ChatGPT is based on deep learning and neural networks, which have more computing power and a volume of training data, compared to 'Elisa'. As a result, ChatGPT is able to offer more complex and informative answers, while having greater flexibility and the ability to adapt to different situations", ChatGPT answers a question about similarities with "Elisa".
ChatGPT considers Elisa's most important contribution to be its first use of templates and input request processing rules in natural language processing. It was this principle that became the basis for the development of more complex language models, which are now also used in ChatGPT.