What is the Turing test?
In "Computing Machinery and Intelligence," published in the philosophy journal Mind in 1950, the British mathematician Alan Turing argued that "Can machines think?" was too vague to answer and proposed the imitation game instead. A judge holds text-only conversations with two hidden parties and tries to say which one is human. If a machine fools the judge about as often as a person would, Turing suggested, we have good reason to call it intelligent. This is what is now called the Turing test.
Turing predicted that in about fifty years an average interrogator would have no more than a 70 percent chance of identifying the machine after five minutes of questioning. He also answered a series of expected objections, including the theological objection, the mathematical objection and the argument from consciousness.
- Criterion
- Indistinguishable from a human in conversation
- Strength
- Replaces a vague question with an observable test
- Weakness
- Fooling people is not the same as understanding
Defining artificial intelligence
The term "artificial intelligence" first appeared in the 1955 proposal for the Dartmouth Summer Research Project, written by John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon; the 1956 Dartmouth workshop is often treated as the founding event of the field. Russell and Norvig's textbook Artificial Intelligence: A Modern Approach sorts definitions into four groups: thinking humanly, acting humanly, thinking rationally and acting rationally. The Turing test is the classic example of the "acting humanly" approach.
The Chinese Room objection
In 1980 the philosopher John Searle offered the Chinese Room thought experiment. A person who knows no Chinese sits in a room with a rulebook and returns correct Chinese answers to Chinese questions. To outsiders the room seems to understand Chinese, yet no one inside does. Searle concluded that manipulating symbols by rules is not enough for understanding or meaning. Because it judges only outward behavior, the Turing test is also described as a behaviorist criterion.
The Turing test today
Competitions such as the Loebner Prize, held from 1991, tried to run the test in practice, but short conversations and inconsistent judging limited their value as a measure of intelligence. Now that large language models produce human-like text, researchers increasingly evaluate specific abilities such as reasoning, factual accuracy and safety rather than whether a system can pass as human.
References
- Turing, A. M. (1950). Computing machinery and intelligence. Mind, 59(236), 433–460.
- McCarthy, J., Minsky, M. L., Rochester, N., & Shannon, C. E. (1955). A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence.
- Searle, J. R. (1980). Minds, brains, and programs. Behavioral and Brain Sciences, 3(3), 417–424.
- Russell, S., & Norvig, P. (2021). Artificial Intelligence: A Modern Approach (4th ed.). Hoboken, NJ: Pearson.
- Oppy, G., & Dowe, D. "The Turing Test." Stanford Encyclopedia of Philosophy.