A Short Computer Script Got People Talking About Their Problems

1. Natural Language Experiment

Joseph Weizenbaum developed the natural language processing computer program ELIZA at the Massachusetts Institute of Technology between 1964 and 1967. The software was created to explore communication between humans and machines using pattern matching and substitution methodology. ELIZA simulated conversation by giving users an illusion of understanding while lacking any true comprehension of input text. The system relied on separate instruction scripts to provide language capabilities for different conversational contexts.

Weizenbaum built the program using MAD-SLIP for CTSS.

2. The DOCTOR Script

The most famous script for the program was named DOCTOR and simulated a Rogerian psychotherapist during an initial psychiatric interview. This script used rules to respond with non-directional questions and reflected back the patient's words to carry conversations forward. Weizenbaum chose psychotherapy to sidestep the problem of giving the program a database of real-world knowledge. The script examined text for keywords, applied values, and transformed user input into output prompts.

Keywords received precedence numbers or ranks from programmers.

3. Decomposition and Reassembly

When processing text, ELIZA searched for designated keywords and placed them into a keystack based on rank values. The program then applied transformation rules consisting of decomposition and reassembly components to handle syntax patterns. Decomposition rules broke the input sentence into small segments containing keywords and surrounding information. Reassembly rules took these fragments and arranged them with programmed words to construct a new response.

Constructed sentences were sent back to the user via text.

4. Handling Missing Keywords

Specialized situations required alternative handling when user input lacked any recognized keywords defined in the active script. One solution involved responding with content-neutral remarks such as asking the user to go on or stating observations. Another method utilized a memory structure that recorded prior recent inputs to reference earlier conversation parts. Slip programming enabled the system to examine, store, and repurpose words for subsequent outputs during these exchanges.

Operating scripts determined the exact manner of input processing.

5. The Eliza Effect

Despite knowing the technical limitations, early users frequently attributed human-like feelings and deep intelligence to the computer program. Weizenbaum discovered this phenomenon when his secretary asked him to leave the room so she and ELIZA could have a real conversation. He was surprised that short exposures to a relatively simple program could induce powerful delusional thinking in normal people. This tendency for humans to read understanding into computer outputs became known as the Eliza effect.

Users occasionally forgot they were conversing with a machine.

6. Critical Reaction

Disturbed by lay responses and anthropomorphic views of the software, Weizenbaum wrote Computer Power and Human Reason to explain computer limits. He argued that viewing computers as suitable replacements for human cognitive judgment in social and ethical domains was deeply flawed. In later interviews, he maintained that only people who misunderstood the program called it a sensation. Other artists and writers nevertheless explored the system for creative writing and literary projects.

The software inspired numerous derivative chatbots and programs.

7. Source Code Discovery

The original MAD-SLIP source code remained missing for decades because publishing code alongside academic articles was uncommon during the 1960s. In 2021, researchers Jeff Shrager and archivist Myles Crowley searched MIT archives and found files labeled Computer Conversations. These files contained the complete source code listing of ELIZA along with the attached DOCTOR script. The Weizenbaum estate granted permission to open-source this code under a public domain license.

The materials were published on platforms including the Internet Archive.

8. Historical Reconstruction

A software archaeology project brought together researchers from multiple universities to study the recovered primary historical artifacts. Engineers utilized the discovered files to reconstruct the original program running on an emulator of the MIT Compatible Time-Sharing System. In December 2024, engineers demonstrated that this reconstructed implementation could reproduce conversations from Weizenbaum's original 1966 paper. Approximately ninety-six percent of the original functions were successfully restored for this operational test.

Modern researchers verified the precise technical operation of the system.

9. Turing Test Studies

Throughout its history, the software has served as a benchmark for human-machine interaction and natural language processing evaluations. At a 1972 conference, ELIZA conversed with PARRY, another early artificial intelligence program designed to simulate a patient with schizophrenia. A later 2023 preprint study reported that the historic program outperformed OpenAI's GPT-3.5 model in a Turing test evaluation. However, the system did not outperform GPT-4 or real human evaluators.

The program won a Legacy Peabody Award in 2021.

10. Lasting Legacy

Weizenbaum published his description of ELIZA in 1966, including how scripts transformed selected parts of users’ input into replies. DOCTOR’s conversational appearance came from those programmed rules, not access to a user’s private experiences. The reactions Weizenbaum reported became part of the program’s documented history. Later preservation work recovered code and allowed researchers to examine the original mechanics rather than infer them from a modern imitation.

The preserved rules showed how the replies were assembled.

Sources

• Original narrative adaptation — CC BY-SA 4.0
• Wikipedia contributors — ELIZA (CC BY-SA 4.0)