A public browser version of ELIZA now gives users a way to prod one of computing’s best-known chatbots while watching the machinery underneath do its rather limited work.
The implementation is credited to Ant and Max Hay in 2023 and is released under CC0 1.0 public domain terms, according to the project page. It runs Joseph Weizenbaum’s 1966 DOCTOR script, the simulated psychiatric interview program published with his paper, “ELIZA: A Computer Program for the Study of Natural Language Communication Between Man and Machine.” The DOCTOR script is listed as copyright 1966 by the Association for Computing Machinery.
The page presents ELIZA in its familiar clinical mode: “HOW DO YOU DO. PLEASE TELL ME YOUR PROBLEM.” Users are told to type about feelings and thoughts in the style of an initial psychiatric interview, then press Enter to receive a response. The maintainers say they believe the replies match what the original 1966 ELIZA/DOCTOR would have produced.
That claim is the interesting bit. Modern chatbot demos usually hide the stack behind a friendly box and a pile of branding. This version includes inspection commands. Typing an asterisk after an exchange can show a trace of the most recent reply. A double asterisk shows the transformation rules used in that reply. Other commands reveal all keywords in the current script or the rule attached to a selected keyword.
In plain terms, the program is not “understanding” a user in the way current AI marketing likes to imply. The interface points users toward the rule machinery: keywords, scripted transformations, and logs showing how a response was assembled. The page says the trace is best read alongside Weizenbaum’s 1966 CACM paper.
Runs locally, with caveats
The maintainers say conversations are private to the user’s machine and that nothing typed into the page is transmitted elsewhere. They also add the necessary warning: ELIZA is not a real psychiatrist.
The extra controls go beyond the original ELIZA. Commands can clear or reset the session, save the conversation, save the conversation with its trace, and export Weizenbaum’s 1966 DOCTOR script to a text file. Users can also load a Weizenbaum-format script from disk, change display colors, toggle full-screen mode, and adjust font size.
One command replays the conversation from Weizenbaum’s 1966 CACM paper. Another, *maxtran in the interface text, sets a maximum number of transformations, with the page noting that a value of zero removes the limit and that Escape can interrupt Turing machine operation. The implementation also includes trace modes for showing every exchange or just the input sentence and active keyword.
The result is less a therapist bot than a small archaeology exhibit with working parts. It lets people talk to ELIZA, then inspect the exact rule-driven path that produced the answer, which is more honesty than many newer chatbot interfaces offer.
This story draws on original reporting from Daring Fireball.
