Synonyms: imitation game, AI indistinguishability test, human-likeness test, bot-detection test, conversational realism test
Do not index
Definition
The Turing test asks whether a person can tell a machine apart from a human in conversation (Alan Turing, 1950). The UX version flips the question: in a real product, the goal usually is to make an AI they know is AI feel trustworthy and useful. Passing as human is easy to fake and expensive in lost trust.
Use cases
Chase indistinguishability and you build something that wins a lab benchmark and loses user trust the moment the illusion cracks. The cost is broken trust, legal exposure, and creepy experiences.
The undisclosed support bot. A "human" agent turns out to be AI, and the user feels deceived the second they catch it. Disclosing up front trades a little magic for durable trust.
The Google Duplex moment. A bot that books appointments in a flawless human voice impresses on stage and unsettles in real life. People want to know what they're talking to.
The over-human companion. An AI styled as a caring "coach" triggers the ELIZA effect, where users attach to it even knowing it's a machine. That's a design responsibility, not a win.
How it's used in practice
Disclose by default: tell users they're talking to an AI, clearly and early, not buried in a tooltip. Many places now require it (EU AI Act, California's bot law).
Design for "knowingly AI": make the AI pleasant and useful as an AI, with clear capabilities and limits, not a fake persona.
Run the real test: instead of "could this fool someone?", ask "does the user understand what they're dealing with, trust it appropriately, and finish their task?"
Watch the anthropomorphism dial: names, avatars, and warm language raise attachment, so tune them to the stakes.
🪄
Pro-tip: Indistinguishability makes a fine research benchmark and a poor product goal. Models that "pass" usually do it by dodging and pretending, which is the opposite of what a good product does. Aim for transparent and trusted instead.
And remember the ELIZA effect: people over-attribute humanness even after you tell them it's a bot, so disclosure is where the work starts, not where it ends.
Challenges & limitations
Disclosure doesn't undo attachment. People form bonds and trust even when told they're talking to AI, so a label isn't a full safeguard against over-reliance.
"Human-like" can backfire. Too much realism lands in the uncanny valley or feels manipulative; too little feels robotic, and the right setting depends on the stakes.
