How universal can an intelligence test be?

Abstract

The notion of a universal intelligence test has been recently advocated as a means to assess humans, non-human animals and machines in an integrated, uniform way. While the main motivation has been the development of machine intelligence tests, the mere concept of a universal test has many implications in the way human intelligence tests are understood, and their relation to other tests in comparative psychology and animal cognition. From this diversity of subjects in the natural and artificial kingdoms, the very possibility of constructing a universal test is still controversial. In this paper we rephrase the question of whether universal intelligence tests are possible or not into the question of how universal intelligence tests can be, in terms of subjects, interfaces and resolutions. We discuss the feasibility and difficulty of universal tests depending on several levels according to what is taken for granted: the communication milieu, the resolution, the reward system or the agent itself. We argue that such tests must be highly adaptive, i.e. that tasks, resolution, rewards and communication have to be adapted according to how the evaluated agent is reacting and performing. Even so, the most general expression of a universal test may not be feasible (and, at best, might only be theoretically semi-computable). Nonetheless, in general, we can analyze the universality in terms of some traits that lead to several levels of universality and set the quest for universal tests as a progressive rather than absolute goal.