A Review of Turing’s Paper

Assignment: https://ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-803-the-human-intelligence-enterprise-spring-2006/assignments/2_decode.pdf

Turing’s Paper was first published in 1950:
Turing, A. M. “Computing Machinery and Intelligence.” Mind 49 (1950): 433-460.

Concerning the question of “is it possible to create an intelligent machine?” Turing proposes that the question is transformed into “Is it possible at least for a machine to fool people into thinking it is a person?” This is worthy because it is at least testable, whereas other definitions of intelligence are subject to the vagaries of metaphysical speculation. The test has become so associated with Turing that it has come to be called the Turing test.

He proposes that if it were possible to create a machine that has the intelligence of an infant, then it would be possible to teach it until it became as intelligent as an adult. This turned out not to be so simple, however, as children are less a “blank notebook” than he thought, but rather have inherited precocities toward understanding social interactions and language.

Turing models an intelligent machine as a state machine, with rules for transition between states. Much of modern machine learning, however, such as neural networks for image or speech recognition, is not done with discrete state machines.

Turing proposes several arguments against the possibilities of machine intelligence. Some of these anti-arguments seem irrelevant today because of developments in computing. Others seem to be logical non-sequiturs. These he parries very well. The strongest arguments he fails to address, but refers people to the Turing test.

(1) The Theological Objection. Thinking is a function of man’s immortal soul. God has given an immortal soul to every man and woman, but not to any other animal or to machines. Hence no animal or machine can think.  COMMENTARY: I’m not sure who actually proposes this. As a creation of man, who is a creation of God, can’t a machine be designed to have rudimentary thought?

(2) The ‘Heads in the Sand’ Objection. “The consequences of machines thinking would be too dreadful. Let us hope and believe that they cannot do so.” COMMENTARY: Who is actually proposing this?

(3) The Mathematical Objection. COMMENTARY: Now, this is interesting. From Gödels theorem, we know that any symbolically rigorous system can generate statements that are not provable in the system. Such as “This sentence is false.” This is the strongest objection to Turing’s modeling of intelligence as a discrete state machine. A state machine might be prone to getting stuck when encountering an indefinite state. Turing doesn’t address this question directly, but suggests instead the Turing test – testing to see whether it can full a human interlocutor.

(4) The Argument from Consciousness. COMMENTARY: This seems rather dated. What is consciousness but a symbolic system with a symbol that represents itself?

(5) Arguments from Various Disabilities. examples: A computer can never be friendly, have a sense of humor, enjoy strawberries and cream. COMMENTARY: These are pretty good objections. I don’t even know if I have a sense of humor, or if it were possible for me to learn to have one. Turing doesn’t do a good job of handling these questions, generally saying they are frivolous. He does, however have a creative aside in this section, where he proposes that a computer can learn by altering its behavior and modeling what the result is. This anticipates modern machine learning.

(6) Lady Lovelace’s Objection, that a computer can’t do anything original, only what we tell it, but that humans can do original things. COMMENTARY: Turing proposes that computer behavior can be unpredictable, as well, and proposes that this is as good as original. I might say – are humans completely original? We are prone to getting stuck in largely deterministic programs, as when I used to get in my car to go to buy groceries, but end up at work, or when my grandfather repeats the same story that he has told before.

(7) Argument from Continuity in the Nervous System: essentially – a nervous system is analog, while a computer is digital. COMMENTARY: this seems not so important now that we have machines that can emulate analog machines by calculating to a lot of decimal points.

(8) The Argument from Informality of Behaviour – that computers are programmed to behave according to rules, but that it is not possible to come up with a set of rules for every circumstance. COMMENTARY: This seems irrelevant, too, since it is possible for a computer to play chess, without accounting for every imaginable circumstance.

(9) The Argument from Extra-Sensory Perception – COMMENTARY: This seems irrelevant. Is there even any scientific evidence for ESP?

Turing’s responses to objections don’t prove that computers can think, but the paper remains important for three propositions:

  1. Rather than asking whether a computer is intelligent, can we have a computer emulate intelligence well enough to fool people.
  2. Can we create a computer, and have it learn from experience?
  3. Can we create a computer, and have it learn by modeling the consequences of its behavior?

These remain the subject of research in artificial intelligence today.