Fable
Not just asking whether it can pass as human in an exam room
Early machine exam rooms only asked and answered questions through a curtain. If the examiner could not tell whether a person or a machine was behind the curtain, everyone applauded. Later, machines became very good at small talk, could tell jokes, and could imitate hesitation. But when faced with mathematics, common-sense reasoning, planning, and interdisciplinary questions, they revealed emptiness. The examiner first added chat questions. The machine seemed more human, and scores rose, but it still could not prove it could solve real tasks. A teacher changed the exam room into many small rooms: a logic room, a physics room, a language room, a social common-sense room, and a creativity room. Each room tested only one ability.
The machine no longer passes by imitating tone. It must reason through unfamiliar problems, combine knowledge, explain its choices, and expose which kinds of problems it fails on.
The first year's results looked ugly, but researchers could finally compare the capability profiles of different complete setups, rather than just hearing whether they seemed human.
Later someone pointed out that once the question bank leaks into practice materials, the exam venue also becomes invalid. So question sources, versions, and contamination checks were written into the rules.
The point of this examination hall is not to crown machines with laurels, but to force everyone to view intelligence through a broader list of capabilities.
The teacher said: talking like a person is only the entry point; what really needs to be tested is whether it can solve problems across tasks.
There was a small house specifically for testing counterintuitive questions: where to patch first when a wooden bucket leaks, how to change the route when a caravan lacks horses, how to judge the loss when a page is missing from the account book. If a machine can only make small talk, it will talk big here but fail to calculate clearly.
The examination hall later no longer sought a single total score that crushed everyone, but instead drew a capability map. Which types of questions were solid, which were hollow, and which collapsed as soon as the wording changed—researchers could all see it. The teacher said that a broad examination hall makes boasting harder and also gives improvement a direction. Later, each time new examinees were brought in, the teacher first asked whether the question bank had been memorized, to avoid the exam falling back into performance. If the questions themselves are memorized, even a broad examination hall loses its sharpness. A true examination hall must leave empty talk nowhere to hide. This is more honest than judging whether it sounds human. And harder to cheat.