An Overview of Approaches Evaluating Intelligence of Artificial Systems

Abstract

Artificial General Intelligence seeks to create an artificial system capable of solving many different and possibly unforeseen tasks thus being comparable in its intelligence to that of a human. Such an endeavour, however, requires suitable methods that can evaluate whether an artificial system is intelligent, and to what extent. This review paper searches for such evaluation methods. Therefore, an extensive literature overview is conducted that covers both philosophical and cognitive presumptions of intelligence as well as formal definitions and practical tests of intelligence grounded in Algorithmic Information Theory. Based on a comparison of the introduced approaches, the paper identifies two distinct groups based on fundamentally different presumptions. The one group of approaches, such as Turing test, is based on the presumption that success in a complex task is a sufficient condition for intelligence evaluation, while the other group of approaches, such as Algorithmic Intelligence Quotient test, also require explicit verification of success in simple tasks. This paper, therefore, concludes that the Algorithmic Intelligence Quotient test, derived from Universal Intelligence definition, is currently the most suitable candidate for a practical intelligence evaluation method of artificial systems. Although the test has several known limitations.

Keywords

• Artificial General Intelligence
• Universal Intelligence Definition
• Anytime Intelligence Test
• Algorithmic Intelligence Quotient Test
• Evaluating Intelligence of Artificial Systems