Abstract

A plethora of new kinds of systems need to be evaluated and characterised, including AI systems, technology-enhanced humans, biologically-enhanced computers, collectives and other hybrid cognitive systems. The successful populational psychometric approach used for over a century has to be overhauled for this new space, since populations are now arbitrarily created and no longer linked to species or human groups. Still, many ideas from psychometrics (humans), comparative cognition (non-human animals) and AI (non-biological machines) can be brought together upon a new foundation based on algorithmic information theory (AIT). As examples I will show (1) how some common items found in IQ tests can be seen as a paradigmatic example of inductive inference using AIT and (2) how the difficulty of a task can be derived intrinsically, from the resources used by its simplest solution policy, in agreement with Levin’s universal search. This represents a solutional, non-populational approach to the arrangement of the space of abilities, from specific tasks to general intelligence, and a ratiocentric perspective to analyse ethics and risks about the future of intelligence.