To avoid defining thinking, which is a challenging task, Alan Turing proposes a new question—can machines do what humans (thinking entities) do?—to answer the original question. He believes humans decide whether others think by judging others’ behaviors. He ignores cognition and focuses only on behaviors to test whether machines can think. If machines demonstrate thinking entities’ behaviors, machines think.
The Turing Test (the Imitation Game) is constructed to determine if machines can imitate thinking entities. The Imitation Game contains two players, a computer and a human, who are placed in two separate rooms. Communicating through only typewritten papers, the human acts as the interrogator asks questions and demands responses from the computer. If the human cannot tell that the responder is a computer, the computer passes the Turing Test. A modern example, PARRY, the responder, is a computer program that plays an imitation game with human psychiatrists, the interrogator, by attempting to pass as a paranoid schizophrenic patient. In the game, PARRY was mixed up with a group of real schizophrenics, and psychiatrists had to tell them apart. Only 52% of the psychiatrists during a study could identify PARRY (Wikipedia).
The Chinese room experiment poses the question of whether passing the Turing Test is enough to prove that a computer thinks. Searle conducts a thought experiment: A native Chinese speaker, the interrogator, sends questions in Chinese to a responder who has no understanding of Chinese in the Chinese room. The responder will follow instructions that will help him produce responses in Chinese and deceive the Chinese speaker into thinking that the responder is Chinese.
Searle’s experiment calls attention to cognition, which Turing ignores, and attacks the worst-case scenario of the Turing Test: The computer passes the test but understands nothing. The responder here can produce perfect answers with perfect instructions but has no understanding of Chinese. The responder is a metaphor for a computer that can pass the Turing Test. This computer does demonstrate cognitive behaviors to an observer. However, inside the computer, its only cognitive process is following instructions without understanding anything. Searle claims if it cannot understand, it is not thinking. Therefore, the Turing Test is not an adequate test to show that computers can think.
To judge whether Searle successfully undermined Turing’s method, first I have to consider a definitional question: Do following instructions without understanding the instructions count as thinking? Searle attempts to incorporate understanding as part of the concept of thinking and tries to manipulate Turing’s argument from “if one behaves like a thinking thing, one thinks” to “even if one behaves like a thinking thing, one must understand to think;” he is simply suggesting a new premise—thinking must include understanding. I believe a person in the vegetative state has no understanding of anything still “thinks” in the most basic sense, since its brain is still regulating basic body functions like respiration, unconsciously following basic instructions for survival, like a computer. To refute what I just proposed, Searle must take an extra step to prove his new premise. Otherwise, his argument cannot proceed with a controversial premise. Also, another problem with Searle’s argument is that it’s begging the question. It assumes that machines don’t understand, so he concludes that it cannot think. If we accept Searle’s premise that the machine who passes the test can only simulate but cannot understand, the conclusion must be that machines cannot understand. Furthermore, I believe machines must be able to understand to pass the Turing Test 100% of the time without any assistance, so Searle’s question-begging premise becomes irrelevant. The interrogator will say a new expression at some point that all humans understand while machines cannot process with the available instructions; the machine will fail the test. Machines have to understand the human language to comprehend the new expressions and always pass the test.
You can also read: