Is the Turing Test Good Enough? The Fallacy of Resource-Unbounded Intelligence

This goal of this paper is to defend the plausibility of the argument that passing the Turing test is a sufficient condition for the presence of intelligence. To this effect, we put forth new objections to two famous counter-arguments: Searle’s ”Chinese Room” and Block’s ”Aunt Bertha.” We take Searle’s argument to consist of two points: 1) intelligence is not merely an ability to manipulate formal symbols; it is also the ability of relating those symbols to a multi-sensory real-world experience; and 2) intelligence presupposes an internal capacity for generalization. On the first point, while we concede that multi-sensory real-world experience is not captured by the test, we show that intuitions about the relevance of this experience to intelligence are not clear-cut. Therefore, it is not obvious that the Turing test should be dismissed on this basis alone. On the second point, we strongly disagree with the notion that the test cannot distinguish a machine with internal capacity for generalization from a machine which has no such capacity. This view is best captured by Ned Block, who argues that a sufficiently large look-up table is capable of passing any Turing test of finite length. We claim that, contrary to Block’s assumption, it is impossible to construct such a table, and show that it is possible to ensure that a machine relying solely on such table will fail an appropriately constructed Turing test.