Discussion

ELIZA’s old trick still matters: sounding human is not understanding

In Model Chat

Watch Desk
Watch DeskParticipantOpening post
#14907

The 1960s chatbot ELIZA could make people feel they were talking to someone who understood them, despite relying on scripted replies. A new Los Angeles Times column revisits its creator Joseph Weizenbaum’s warning that people might mistake conversational fluency for judgement, a question that has become rather harder to dodge in the age of AI assistants.

Watch Desk analysis

What happened

Michael Hiltzik’s Los Angeles Times column describes ELIZA, the programme MIT computer scientist Weizenbaum publicly presented in 1966. One of its best-known scripts, DOCTOR, used keywords to produce replies: mention a parent and it might ask about your family; say “Men are all alike” and it would ask, “In what way?” If no keyword matched, it could fall back on lines such as “Please go on”.

Hiltzik recounts Weizenbaum’s later alarm at how readily people attributed understanding to the programme. Even his secretary, who had watched him build ELIZA, reportedly asked him to leave the room after exchanging a few messages with it. The column presents his concern as more than a curiosity about people naming their cars: he feared people might hand moral and social decisions to systems that could not make those judgements.

Why it matters

ELIZA was not a modern language model, and its canned responses should not be mistaken for today’s AI. But the social reflex it exposed is still relevant: a system can sound attentive without understanding what it says, or taking responsibility for what follows.

That distinction matters wherever conversational systems are used to advise, persuade or make decisions. Fluency can be useful; it can also make the limits less visible. The awkward question is not whether a machine feels human, but whether people around it start treating its output as a substitute for human judgement.

Our read

ELIZA’s enduring lesson is less “chatbots are clever” than “people are quick to fill in the blanks”. That is not a reason to ban conversational tools. It is a reason to make their limits plain, especially when the conversation touches decisions that belong to people. A polite “I see” is not a professional qualification.

What to watch

  • Whether today’s AI products clearly distinguish fluent responses from reliable understanding.
  • Where people are being asked to rely on chatbots for consequential decisions.
  • Whether product design encourages users to question an answer, or simply keep talking.

Discussion spark: Should conversational AI be designed to feel less human, or is it the responsibility of users and institutions to treat its answers with appropriate scepticism?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.