Turing Test

The Turing Test is a behavioral criterion for machine intelligence, proposed by Alan Turing in his 1950 paper “Computing Machinery and Intelligence” (Mind). Turing’s proposal: if a computer can pass for human in an online conversation (the “imitation game”), we should grant that it is intelligent. It is the most famous — and most debated — test in the philosophy of AI, and the primary target of Searle’s chinese-room-argument. ¹

A Simplifying Introduction

Imagine three participants in separate rooms: a human, a machine, and an interrogator. The interrogator chats with both (by text only) and must decide which is the human and which is the machine. The machine’s goal is to fool the interrogator; the human’s goal is to help the interrogator identify correctly. If the machine is indistinguishable from the human in conversation, Turing argued, we should treat it as thinking — because there is no better evidence for human thinking either. The test sidesteps the unanswerable question “what is thinking, really?” and replaces it with an observable behavioral criterion. ¹

The Imitation Game (Original Formulation)

Turing’s own framing: the interrogator knows the two participants only as X and Y, asks questions, and at the end declares which is the person. Turing predicted that “in about fifty years’ time” (i.e., by ~2000) computers with storage capacity ~10⁹ would play the game so well that an average interrogator would have no more than a 70% chance of correct identification after five minutes of questioning. He also predicted that by the end of the century general educated opinion would accept talk of “machines thinking.” Both predictions were largely unmet at the deadline — the Loebner Prize Competition (an annual Turing Test contest) was widely regarded as an embarrassment, with entrants easily detected. ¹

Objections Turing Addressed (1950)

Turing anticipated and replied to nine objections in the original paper:

1. Theological objection — thinking is a matter of an immortal soul; machines can’t have one.
2. “Heads in the sand” — the consequences of machines thinking would be too dreadful.
3. Mathematical objection — Gödel-style limits on formal systems constrain what machines can do.
4. Argument from consciousness — machines could pass behaviorally but lack genuine experience (the “other minds” worry in reverse).
5. Arguments from various disabilities — machines can’t be kind, resourceful, fall in love, etc.
6. Lady Lovelace’s objection — machines can only do what we program them to do; no originality.
7. Continuity of the nervous system — discrete-state machines can’t mirror the continuous brain.
8. Informality of behavior — human behavior is too unrule-governed to be captured by a program.
9. Extra-sensory perception — a curiosity of its time, but Turing took it seriously enough to address.

¹

What Exactly Does Passing the Test Establish?

The SEP distinguishes four interpretations of what the Turing Test provides:

Interpretation	Claim	Status
Logically necessary & sufficient conditions	Pass iff intelligent	Doubtful; few defend this reading (maybe Block 1981)
Logically sufficient conditions	Can’t be both unintelligent and pass	Vulnerable to the “philosophical zombie” objection: a being could behave perfectly yet (allegedly) lack inner life
Criteria (Wittgenstein)	Defeasible outward grounds for ascribing mental states	The most defensible philosophical reading: “an inner process stands in need of outward criteria”
Probabilistic support	Passing makes intelligence likely	Closest to Turing’s own intent; his prediction was explicitly probabilistic

¹

Alternative Tests

• The Total Turing Test (Harnad 1989, 1991) — requires full human-like sensorimotor response to all inputs (robotic embodiment), not just text. Ties to symbol grounding (see chinese-room-argument).
• The Lovelace Test (Bringsjord et al. 2001) — an agent passes iff it produces output that its designer cannot explain by appeal to its architecture and knowledge base — a test of genuine creativity beyond programming.
• The Truly Total Turing Test — demanding neuromolecular indistinguishability from humans.

¹

Is the Turing Test Harmful as a Research Goal?

Hayes and Ford (1995) argued it is: the test sets a null-effect benchmark (no behavioral difference between machine and human), and null experiments can’t measure anything — there’s no stable endpoint because judges might always have asked harder questions. On their view the test is really “a test of the ability of the human species to discriminate its members from human imposters.” ¹

The Chinese Room Connection

The chinese-room-argument is the most famous objection in the vicinity: Searle imagines a symbol-manipulator that passes the Turing Test for Chinese while understanding nothing. The SEP’s reconstruction notes the argument’s modal structure — there is a (remote) possible world in which a digital computer simulates intelligence without possessing it, and actual computers don’t differ in any way that could change that verdict. The Systems, Robot, and Virtual Mind replies all turn on whether the whole system (not the room operator) could be the understander. The Turing Test’s behaviorism is precisely what Searle attacks: passing behavior is not sufficient for understanding. ^{1 2}

Modern Relevance

The Turing Test remains central to debates about LLMs: models that write essays, pass exams, and converse fluently are the first systems to approach the original behavioral criterion at scale. Defenders of the functionalist view (e.g., Agüera y Arcas) hold that sustained successful imitation is the real thing; critics in the Searlean tradition hold that syntax without semantics remains simulation. Whether “understanding” is judged by outward criteria or inner states remains unresolved.

Relationship to This Wiki

• The behavioral criterion that Searle’s Chinese Room is designed to refute
• The functionalist methodology defended in what-is-intelligence-book (“the Turing Test as valid”; the Kidney Turing Test as functionalism extended to biology)
• Its modal structure (possible worlds where machines simulate but lack intelligence) connects to possibilism-actualism and modal-logic

See Also

• chinese-room-argument — The famous challenge: can a symbol-manipulator pass the test without understanding?
• what-is-intelligence-book — Modern defense of the Turing Test as valid criterion for intelligence
• john-searle — The Turing Test’s most famous critic
• modal-logic — Modal structure of the simulation-without-intelligence possibility
• ai-scientist — Modern AI systems approaching the behavioral criterion at scale

References

• Oppy, G. & Dowe, D. (2021). “The Turing Test.” Stanford Encyclopedia of Philosophy. ¹
• Turing, A. (1950). “Computing Machinery and Intelligence.” Mind, 59(236), 433–460.
• Cole, D. (2024). “The Chinese Room Argument.” Stanford Encyclopedia of Philosophy. ²
• Harnad, S. (1989). “Minds, Machines and Searle.” Journal of Experimental & Theoretical AI, 1(1), 5–25.
• Hayes, P. & Ford, K. (1995). “Turing Test Considered Harmful.” Proceedings of IJCAI-95.

Footnotes

1. raw/articles/sep-turing-test.md ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹

2. raw/articles/sep-chinese-room.md ↩ ↩²