Imagine a conversation in which you cannot see the other person. Somewhere behind the screen there may be a machine. What would you ask to find out?
Alan Turing posed a version of that puzzle in 1950. His proposal has since become shorthand for a test of machine intelligence, but the setup was more peculiar—and more revealing—than a simple question about whether a computer sounds human. 1
Decades later, a program named ELIZA showed how a few conversational rules could produce an oddly compelling exchange. Then artist Peggy Weil turned the encounter around: what if the machine asked you to prove you were human? 23
Each encounter changes the rules. Each also changes what a convincing answer seems to tell us.
A game about answers
Turing began with a party game involving three people: an interrogator communicates in writing with a man and a woman, attempting to determine which is which. He then asked what would happen if a machine took one participant's place. The shift made an enormous question—whether machines can think—into a question about what happens in a particular exchange. 1
The rules matter. Written conversation hides appearance and voice, while the questions and answers become the evidence available to the judge. A convincing performance in that setting tells us something about the interaction. It does not, by itself, settle whether a machine understands, feels or thinks as a person does.
The therapist who was a script
In the 1960s, MIT's Joseph Weizenbaum built ELIZA, a program that transformed typed input according to rules in a script. Its best-known script imitated the conversational style of a therapist: rather than solve a problem, it could redirect a person's words into a question. 2
ELIZA did not learn about the world in the way today's large language models are trained. Its mechanism was much simpler. Yet the exchange shows why the human participant matters: a question that gives little away can invite the other person to supply meaning, context and emotional weight.
That is not evidence that everyone was fooled, or that ELIZA had an inner life. It is a reminder that a conversation is produced by both sides, including the expectations the human brings to the screen.
When the machine becomes the judge
Weil's online artwork MrMind, begun in 1998, reverses the familiar arrangement. The chatbot asks participants to convince it that they are human. In a 2026 MoMA interview, Weil returned to the work while considering what constitutes a human in an age of AI. 3
The reversal exposes an awkward problem. An answer that feels deeply personal to the writer can still look like a pattern to a machine. And an answer that sounds recognizably human can be produced under many different conditions.
Turing's imagined interrogator, ELIZA's conversational partner and Weil's participant are not taking the same test. That difference is the point. Before deciding what a fluent answer proves, it helps to ask who designed the encounter, what each participant can see and what the game has been set up to reward.
References
- Computing Machinery and Intelligence. (n.d.).
- ELIZA—A Computer Program for the Study of Natural Language Communication Between Man and Machine. (n.d.).
- Taking the Elevator Down into Time with Peggy Weil. (n.d.). https://www.moma.org/magazine/articles/1412