The Chinese Room is John Searle's argument against what he called "strong AI": the thesis that a computer with the right program does not simulate a mind but literally has one, and that the program explains human cognition (Searle, 1980, p. 417). Against "weak AI," which uses the computer as a tool for studying the mind, Searle has no objection.
The argument
Searle, who knows no Chinese, is locked in a room. He receives batches of Chinese writing that are squiggles to him, and rules in English for correlating symbols with other symbols by their shapes alone. Unknown to him, the people handing him papers call one batch "a script," another "a story" and another "questions"; what he hands back they call "answers," and the rules, "the program." With practice his answers become indistinguishable from a native speaker's; even so, he says, "I still understand nothing" (1980, p. 418). The immediate target was Roger Schank's programs that answered questions about restaurant stories, but Searle notes that the argument applies equally to ELIZA, to SHRDLU and to any Turing machine simulation (1980, p. 417).
The conclusion: manipulating formal symbols is not sufficient for understanding, because programs have "only a syntax but no semantics" (1980, p. 422). Confusing simulation with duplication is the same mistake as believing that a simulated fire will burn the neighborhood down (1980, p. 423). Intentionality, he holds, is a biological phenomenon, as dependent on its biochemistry as lactation or photosynthesis (1980, p. 424).
The replies
Searle anticipated several replies and labeled them by their "geographic origins" (1980, p. 418). The systems reply (Berkeley) says the man does not understand but the whole system does; Searle answers that the man could memorize the rules and do everything in his head, and still not understand (1980, p. 419). The robot reply (Yale) proposes adding cameras and motors; Searle replies that the man inside would just receive more symbols, a homunculus who does not know what is going on (1980, p. 420). The brain simulator reply (Berkeley and MIT) proposes simulating the synapses of a Chinese speaker; Searle then imagines a system of water pipes and valves, and concludes that copying the formal structure does not copy the causal powers (1980, p. 421).
Behavioral and Brain Sciences published the article with more than twenty commentaries. Dennett accused it of sophistry and described the case as an "intuition pump" (1980, pp. 428–429); Hofstadter called it a "religious diatribe" (1980, p. 433). In 1990 Searle adapted the example against connectionism with a "Chinese gym" full of men acting as neurons (Searle, 1990), and Paul and Patricia Churchland (1990) replied, in the same issue of Scientific American, with the analogy of the "luminous room."
The room and LLMs
The p(doom) song goes "Trapped in the Chinese room, with a bag of shrooms" (osmarks, n.d.). An LLM changes the set: nobody wrote its rulebook; it consists of billions of parameters tuned to predict the next token. But the argument is indifferent to that change, because it applies to whatever formal principles are put into the computer (1980, p. 418). LLMs neither refute nor confirm it; what they change is the dialectical terrain. The systems reply becomes more interesting when Shanahan (2024) insists on separating the "bare-bones" model from the system in which it is embedded (2024, p. 72); the robot reply returns with multimodal models and agents; and the symbol grounding problem problem of Harnad (1990), a direct heir of Searle's, is being reformulated for vectors (Millière & Buckner, 2024a, sec. 3.3). Bender and Koller's octopus explicitly starts from this room (Bender & Koller, 2020, p. 5188).