Source Summary: Turing 1950
Full title: Computing Machinery and Intelligence
Authors: A. M. Turing
Affiliation: University of Manchester
Venue: Mind, New Series, Vol. 59, No. 236 (Oct. 1950), pp. 433–460
PDF: papers/turing-1950-computing-machinery-and-intelligence.pdf
Note: PDF is a scanned JSTOR copy with no extractable text; content drawn from the paper itself.
Contributions
- The Imitation Game (Turing Test) — operational definition of machine intelligence: a machine is intelligent if an interrogator cannot distinguish it from a human through text-only questioning.
- Digital Computers as universal machines — argued that a digital computer is effectively a universal discrete-state machine and is the right substrate for investigating machine intelligence.
- Learning machines — proposed that rather than programming adult-level intelligence directly, the right approach is to simulate a child’s mind and then educate it.
- Systematic rebuttal of objections — addressed nine major objections to machine intelligence (theological, mathematical, consciousness, etc.) with detailed counterarguments.
The Imitation Game
Setup
Three participants in separate rooms: a man (A), a woman (B), and an interrogator (C). The interrogator communicates only through written text. The goal: interrogator must determine which of A, B is the man and which is the woman. A tries to mislead; B tries to help.
The question Turing actually asks: What happens if we replace A (the man) with a machine? Can the machine take the man’s role successfully — can it deceive the interrogator into thinking it is human?
What this operationalizes
Rather than asking “Can machines think?” (which Turing considers meaninglessly vague), the imitation game provides a behavioral criterion: if a machine can play the imitation game as well as a human, it is intelligent in any sense that matters operationally.
The prediction
Turing predicted (in 1950) that within 50 years, a machine would be able to play the imitation game well enough that an average interrogator would have less than 70% chance of correctly identifying it after 5 minutes of questioning.
Digital Computers
Turing describes digital computers as having three parts:
- Store: memory holding numbers and instructions (equivalent to the tape)
- Executive unit: arithmetic and logic operations
- Control: determines which instruction to execute next
He argues this is a universal discrete-state machine — equivalent to a universal Turing machine — and therefore capable, in principle, of computing anything computable.
The “Child Machine” proposal
The most practically prescient part of the paper. Rather than specifying an adult-intelligent machine (which requires knowing what intelligence is), Turing proposes:
- Build a machine simulating a child’s mind
- Subject it to an education process
- Let it learn
This is functionally identical to the modern paradigm: train a neural network on data rather than explicitly programming knowledge.
Turing also discusses the need for reward and punishment signals to guide learning — anticipating reinforcement learning.
Nine objections addressed
| Objection | Turing’s response |
|---|---|
| Theological: God gave souls to humans, not machines | Limits God’s power unnecessarily |
| Mathematical: Gödel’s theorem shows machines have limitations | Humans have the same limitations |
| Lady Lovelace: machines only do what they’re programmed to | Surprise is possible even in deterministic machines |
| Consciousness: machines can’t have feelings | We cannot verify consciousness in other humans either |
| Continuity of nervous system: neurons are continuous, computers discrete | Discrete approximation can be made arbitrarily fine |
| Informality of behavior: rules cannot capture all human behavior | No evidence that humans are rule-free |
| Extrasensory perception | Properly controlled tests show no effect |
| (Others) | Similar rebuttals |
The Gödel objection is the most technically precise: Gödel’s incompleteness theorems show there are statements a formal system cannot prove. Turing argues that (1) humans cannot necessarily prove those statements either, and (2) the objection only shows machines have limits, not that those limits make them non-intelligent.
Key References
- Turing 1936 — Turing machines (the computational basis)
- Lovelace 1843 — Ada Lovelace’s notes on the Analytical Engine (the objection being rebutted)
- Gödel 1931 — Incompleteness theorems (the mathematical objection)
- Shannon 1950 — Chess-playing machines (related work on game AI)