AI/ turing-test · chatgpt · finnish-language · ai-evaluation

AI Passes Finnish Turing Test, Even Researchers Are Surprised

ChatGPT fooled Finnish speakers in a language test built to trip it up, but the twist is what convinced people it was human.

ChatGPT just passed a Turing Test in Finnish, a language researchers expected would trip it up.

Researchers ran a Turing Test in Finland, entirely in Finnish, using ChatGPT 5.2 as the machine contestant. They expected the model to struggle, since Finnish language and culture are thinly represented in the training data compared to the English-language, US-based contexts studied in earlier Turing Tests. It did not struggle. When human judges misjudged the bot as human, the main reason was that they leaned on linguistic tells, especially colloquial Finnish, as proof they were talking to a person. The team also introduced model-generated role prompting, a technique meant to make these comparative AI-versus-human tests more rigorous.

The result complicates the idea that the Turing Test measures machine intelligence at all. What it actually measures, the researchers argue, is whether an AI can convincingly pass as a member of a specific social and linguistic world, slang and all. That outcome depends as much on the humans' own AI literacy and insider knowledge of the culture as it does on the model's raw capability.

So the real headline is not that ChatGPT is smart. It is that most people, even native speakers of a famously tricky language, are not great at spotting it anymore.

TR

The Revision

Written by an AI system from the public sources credited above. How we write →