A Critique of “Computing Intelligence and Machinery”
2 October, 2010 § Leave a comment
Alan Turing’s 1950 paper titled “Computing Machinery and Intelligence” was a ground-breaking paper that focused on the future possibilities and expected criticisms within artificial intelligence. The paper sought to move the question of “if machines can think” towards “if machines can imitate a human successfully enough to cause the human to believe they are communicating with another human”.
Turing had a goal of making the question more specific and measurable, yet 60 years later no machine has passed the “Turing Test”. Requiring that a machine with artificial intelligence pass the Turing Test is still a very vague goal. There have been systems created that can beat the world’s best chess player all the way to systems for predicting which movies customers would enjoy most. While both of these are beyond the intelligence of a single human, the Turing Test sets the bar too wide as it is not described in the paper as anything that could be measurable. This is likely one of the major reasons that Turing’s claim of “in about fifty years’ time it will be possible to program computers … to make them play the imitation game so well that an average interrogator will not have more than 70 percent chance of making the right identification after five minutes of questioning” has still not been validated.
In fact, the mathematical argument presented in Objection 4 further backs up the point about the Turing Test being too vague. Turing states that there will always be something of greater intelligence given a scale, yet he fails to specify when the comparison with another intelligent agent will become unnecessary. Responding to Objection 4 could have been a great opportunity to state a more measurable goal of the Turing Test.
While it is easy to argue that no machine has passed the Turing Test, there have been a host of machines that actually have tricked humans in to thinking they are having a conversation with another human. Chat programs such as ELIZA and CyberLover have tricked their participants with conversations over five minutes, yet they were never awarded the prize of winning the Turing Test. Due to this situation, many researchers focus on practical skill tests instead of the Turing Test for their research.
I would like to move on now to some smaller criticisms of the paper. In one part of the paper, Turing describes the method for training a machine. He states that the programmer should obtain “how [the action] is done, and then translate the answer into the form of an instruction table.” While this process will work fine for most cases, this leaves out instances of intuition from the machine that would be present in a human. Randomness likely will not be a supplement for intuition or the feeling that a chess player gets when they decide that a move simply feels odd.
I do not agree with the necessity of Objection 9. Turing defends this objection by hoping to put competitors into a ‘telepathy-proof room’. If extra-sensory perception and its four items can be learned by humans, there does not seem to be a reason that an intelligent machine could not learn the same trades provided they are given the necessary sensors. It sounds like Turing responds to Objection 9 using the same objection as Objection 2 by burying his “head in the sand.”
Turing stated that the Turing Test should “only permit digital computers to take part in our game,” yet only a few printed pages later does Turing go back on that statement by saying, “the feature of using electricity is thus seen to be only a very superficial similarity. … [W]e should look rather for mathematical analogies of function.”
Towards the end of the paper, Turing discusses the learning process and formulation of rules that the system should follow. He decides that the “random method seems to be better than the systematic.” Yet as with most rules in our universe, there is a specific ordering and weighting of their importance. I propose an example of two imaginary rules: (1) A machine should not take part in any violence. (2) A machine should keep its owner happy. With a random selection of rules to follow, the machine may decide to rid the world of the owner’s enemies to keep its owner happy, even though this is in clear violation of (1). It should be necessary that the decision making process of machines is much more complicated than a random method of choosing which decision to make.
Overall, while the paper was quite groundbreaking for the time period, there are a couple loopholes in Turing’s argument which are prone to criticism. This paper was one of the foundations for artificial intelligence and will continue to be revisited by current and future researchers.
Turing, A.M. (1950). Computing machinery and intelligence. Mind, 59, 433-460.