a group of scientists from the USA and Great Britain was able to to revive the work of the first Eliza digital bout, created back in 1966. The discovery made it possible to return to life a program developed by MIT professor Joseph Waitocenbaum.
In 2021, the MILZ Crowley Archivarius found in a box with the documents of the Weatherbaum of code printouts, which turned out to be an almost complete set of source data Eliza. The program, conceived as” Electronic therapist “, used the simple logic of the request and response to interact with the user.
Despite the fact that the term “chatbot” did not exist yet, Eliza already performed its functions. The user asked the program a question, and she answered and proposed a clarifying question, imitating communication with the therapist. However, the original code written in the Lisp language was considered lost for decades.
After writing, the program was downloaded to the ARPANET network, which allowed to spread to the level of home computers, where Eliza gained popularity as a virtual “friend” for technically savvy users.
To restore code, researchers had to recreate the environment for Eliza, as the program was developed for an outdated operating system. The code also needed to be cleaned and add -up, including writing the functions mentioned in the program, but absent in prints.
In December 2024, the team was able to launch Eliza. Despite the simplicity of logic, the program worked better than expected. Of course, she cannot compete with modern language models, but researchers emphasize that the revival was an exciting process.
It is interesting that the original code contained an error: the program completed the work if the user introduced the number. However, the researchers decided to leave this bug to maintain authenticity. Scientists emphasize the significance of Eliza as an important part of the history of computer technology. This is the first example of an electronic chatbot, which has become a prototype for the further development of artificial intelligence.
In December, scientists conducted an entertaining comparison between living people of the GPT-3.5 and GPT-4 and GPT-4 and GPT-4, as well as the Eliza virtual interlocutor, written in the far of 1966. Eliza showed relatively good results and reached the indicator of 27%, ahead of the GPT-3.5.