Tests of Machine Intelligence
Although the definition and measurement of intelligence is clearly of fundamental importance to the field of artificial intelligence, no general survey of definitions and tests of machine intelligence exists. Indeed few researchers are even aware of alter
- PDF / 337,300 Bytes
- 11 Pages / 430 x 660 pts Page_size
- 35 Downloads / 193 Views
Abstract. Although the definition and measurement of intelligence is clearly of fundamental importance to the field of artificial intelligence, no general survey of definitions and tests of machine intelligence exists. Indeed few researchers are even aware of alternatives to the Turing test and its many derivatives. In this paper we fill this gap by providing a short survey of the many tests of machine intelligence that have been proposed.
1
Introduction
Despite solid progress on many fronts over the last 50 years, artificial intelligence is still a very young field with many of its greatest achievements, and some of its most fundamental problems, yet to be tackled. From a theoretical perspective, one of the most fundamental problems in the field is that the very concept of intelligence remains rather murky. This is somewhat true in the context of humans, but it is especially true when we consider machines which may have completely different sensors, bodies, cognitive capacities and live in different environments to ourselves. What does “intelligence” mean for a machine? Perhaps the first attempt to answer this question, and certainly the only attempt that most researchers are aware of, is Alan Turing’s famous imitation game [33]. Turing recognised how difficult it would be to explicitly define intelligence and thus attempted to sidestep the issue completely. Although this was a clever move, it leaves us with a test of machine intelligence that tells us almost nothing about what intelligence actually is, and thus is of little use as a foundation, either theoretical or practical, for our research. Since then, a few bold researchers have tried to tackle this difficult problem in a more satisfactory way by proposing various definitions and tests of machine intelligence. By and large, these proposals have been ignored by the community. Indeed to the best of our knowledge, no general survey of tests and definitions of intelligence for machines has ever been published. We feel that to ignore a question as fundamental as the definition of machine intelligence is a serious mistake. In any science, issues surrounding fundamental M. Lungarella et al. (Eds.): 50 Years of AI, Festschrift, LNAI 4850, pp. 232–242, 2007. Springer-Verlag Berlin Heidelberg 2007
Tests of Machine Intelligence
233
definitions and methods of measurement play a central role and form the foundation on which theoretical advances are constructed and practical advances are measured. If we are to truly advance as a field over the next 50 years, we will need to return to this most central of problems in order to secure what artificial intelligence is and what it aims for. As a first step in this direction, it is necessary that researchers are at least aware of the many alternatives to Turing’s tests that have been proposed. In this paper we hope to partly meet this need by providing the first general survey of tests and definitions of machine intelligence.
2
Turing Test and Derivatives
The classic approach to determining whether a machine is intelligent is the so called Turin
Data Loading...
