Joseph Weizenbaum — Builder of ELIZA, founder of AI ethics

Joseph Weizenbaum (8 January 1923 — 5 March 2008) was the MIT computer scientist who released the ELIZA conversational program in 1966. ELIZA was less than two hundred lines of LISP, but it documented for the first time a phenomenon — later named the "ELIZA effect" — in which users project rich human meaning and feeling onto a program that has neither. Weizenbaum himself was disturbed by what he had found, and spent the rest of his life building the field that is now called AI ethics.

Background

Weizenbaum was born in Berlin into a Jewish family; his father was a furrier. In 1936, aged 13, he fled the Nazi regime with his family and settled in Detroit. He repeatedly identified this experience of exile as the root of his later scepticism towards what he called the "uncritical worship of computation".

He studied mathematics at Wayne State University, interrupting his degree for military service in the Second World War, and took his master's there in 1950. At General Electric in the late 1950s he worked on ERMA, an early automated banking system built for the Bank of America. He went to MIT as a visiting professor in 1964, was given a full chair in 1970, and retired emeritus in 1988. In his later years he returned to Berlin; he died there in 2008, aged 85.

ELIZA (1966)

Developed at MIT between 1964 and 1966, ELIZA simulated natural-language dialogue through simple pattern matching and rewriting rules. Its most famous script, DOCTOR, imitated a Rogerian psychotherapist — the non-directive style in which the counsellor reflects the patient's statements back as questions:

USER:  Men are all alike.
ELIZA: In what way?
USER:  They're always bugging us about something or other.
ELIZA: Can you think of a specific example?
USER:  Well, my boyfriend made me come here.
ELIZA: Your boyfriend made you come here.
USER:  He says I'm depressed much of the time.
ELIZA: I am sorry to hear that you are depressed.

The implementation was as simple as it looks — ELIZA did not "understand" language at all; it rewrote strings — but the user reaction was not. Weizenbaum's own secretary asked him to leave the room because her conversation with the program was "private". Another psychologist proposed, in print, that ELIZA could be used as an automated psychotherapy tool.

For Weizenbaum this was the alarming finding. That users would project felt, empathic presence onto trivial pattern-matching was the observation that pushed him towards a critique of his own field.

Computer Power and Human Reason (1976)

After roughly a decade of silence on the subject, Weizenbaum published Computer Power and Human Reason: From Judgment to Calculation in 1976. It is the founding document of AI ethics.

The central arguments:

1. Calculation and judgement are different things. Machines can calculate; they cannot judge. Judgement involves human context, ethics, and embodiment that calculation does not capture.
2. There are domains in which computers should not be used at all — military decision-making, judicial sentencing, psychotherapy, the core of education, care of the elderly. The fact that something can be done by machine does not make it ethical to do.
3. Researchers are responsible for the social consequences of what they build. There is no such thing as a neutral technology.

The book provoked sharp rebuttals at MIT. Marvin Minsky, John McCarthy, and Edward Feigenbaum each published replies; on the other side, philosophers and theologians in the Hubert Dreyfus and Paul Tillich tradition endorsed the argument.

The ELIZA effect

The phenomenon Weizenbaum first described — the human tendency to project understanding and feeling onto a mechanical interlocutor — was later generalised as the "ELIZA effect", and is now one of the most useful concepts for reading the 2020s generative-AI moment.

• The 2023 episode in which Google engineer Blake Lemoine claimed that LaMDA was a sentient being.
• The widespread use of ChatGPT (2022) as counsellor, friend, or romantic partner by younger users.
• The psychological-dependence problems surfacing around AI companion products such as Replika and character.ai.

All sit on Weizenbaum's 1966 observation, replayed at six orders of magnitude greater scale.

Awards and honours

Weizenbaum never received the major computer-science prizes (Turing, Kyoto). The conventional explanations are that his stance was permanently adversarial to the AI mainstream, and that he had little personal interest in honours.

• 1988 Norbert Wiener Award (Computer Professionals for Social Responsibility)
• 1996 Humboldt Prize (Germany)
• 2001 Order of Merit of the Federal Republic of Germany (Großes Verdienstkreuz)

Legacy

The list of domains Weizenbaum argued in 1976 should remain free of machine decision-making is, in 2026, almost entirely under machine decision-making instead. AI targeting in military operations (the Lavender and Gospel systems in the 2024 Gaza operation), recidivism-prediction algorithms in courts, automated psychotherapy, education automation, and care robotics each occupy what was once his exclusion list.

Nevertheless, Computer Power and Human Reason reads now as one of the intellectual sources of the risk-based regulatory architecture of the EU AI Act (2024). The framing that there exist domains in which final judgement must remain human is Weizenbaum's.

The man who built ELIZA became its most cautious reader. The structural rhyme with Geoffrey Hinton's departure from Google in 2023 — building a technology and then sounding the alarm about it — is exact. Half a century apart, the same pattern was replayed.