Can a computer that answers perfectly in Chinese understand a word of it?

According to the philosopher John Searle, no: getting the answers right is not the same as understanding. In a famous 1980 paper, he asked us to imagine a computer so good at Chinese that nobody talking to it could tell it wasn't a hidden Chinese speaker.

Then he put himself in the computer's place. Searle, who reads no Chinese, sits in a room with the program written out in English, plus pencils, paper and filing cabinets. Chinese characters are slipped under the door. He follows the rules step by step, and they tell him which Chinese characters to slide back out. From outside, the room passes the Turing test. Inside, Searle understands nothing. And since he is doing exactly what the computer does, Searle concluded that the computer understands nothing either. Running a program, he argued, is not enough for a mind.

Note what the argument does not claim: it sets no limit on how intelligently a machine can behave. It targets "strong AI", the idea that the right program would literally have a mind like ours.

Most philosophers and AI researchers who have written about it think Searle is wrong. The best-known reply says the man isn't the one who understands, the whole system is. Searle's answer: let the man memorize all the rules; now he is the whole system, and he still doesn't understand. With chatbots now holding fluent conversations, the room feels less hypothetical than ever.