On Potential Cognitive Abilities in the Machine Kingdom

Abstract

Animals, including humans, are usually judged on what they could become, rather than what they are. Many physical and cognitive abilities in the 'animal kingdom' are only acquired (to a given degree) when the subject reaches a certain stage of development, which can be accelerated or spoilt depending on how the environment, training or education is. The term 'potential ability' usually refers to how quick and likely the process of attaining the ability is. In principle, things should not be different for the 'machine kingdom'. While machines can be characterised by a set of cognitive abilities, and measuring them is already a big challenge, known as 'universal psychometrics', a more informative, and yet more challenging, goal would be to also determine the potential cognitive abilities of a machine. In this paper we investigate the notion of potential cognitive ability for machines, focussing especially on universality and intelligence. We consider several machine characterisations (non-interactive and interactive) and give definitions for each case, considering permanent and temporal potentials. From these definitions, we analyse the relation between some potential abilities, we bring out the dependency on the environment distribution and we suggest some ideas about how potential abilities can be measured. Finally, we also analyse the potential of environments at different levels and briefly discuss whether machines should be designed to be intelligent or potentially intelligent.

How universal can an intelligence test be?

“…This is a well-known problem in animal cognition (and also happens with other places where intelligence is sought, such as plants, bacteria, SETI, etc.). Moreover, this under-estimation also happens in human psychometrics, where the term ‘potential intelligence’ is applied to ‘test potential’ (Mahrer, 1958; Thorp & Mahrer, 1959; Little & Bailey, 1972), which is not to be confused with the term ‘potential intelligence’ applied to the ability of becoming intelligent (Hernández-Orallo & Dowe, 2013).…”

“…During a very long exploration for optimal configurations we may (inadvertently) train the agent we want to evaluate and make it more intelligent than it was. In fact, there are training sequences (Solomonoff, 1962, 1984, 2010, 1967) for any universal Turing machine such that the machine becomes intelligent (in fact, as intelligent as we want, as it can simulate any behaviour, as shown and fully discussed by Hernández-Orallo & Dowe, 2013). So there is, in principle, a problem in making intelligence tests too long, as the agents can be domesticated (trained) to become intelligent, unless the agents can be reinitialized for each iteration of the adaptive procedure.…”

“…So there is, in principle, a problem in making intelligence tests too long, as the agents can be domesticated (trained) to become intelligent, unless the agents can be reinitialized for each iteration of the adaptive procedure. If the agent is trained (or just domesticated), the test mixes actual and potential intelligence (in the sense of Hernández-Orallo & Dowe, 2013), and the reliability and validity of the test are highly compromised.…”

Towards a Universal Test of Social Intelligence

“…This set should incorporate many different types of agents: random agents, agents with some predetermined policies, agents that are able to learn, human agents, agents with low social intelligence, agents with high social intelligence, etc. The set of all possible agents (either artificial or biological) is known as machine kingdom in [45,46] and raises many questions about the feasibility of any test considering this astronomically large set. Also, there are doubts about what the weight for this universal set should be when forming line-up patterns (i.e.…”

Universal psychometrics: Measuring cognitive abilities in the machine kingdom

“…If this is finally the right track in AI, then universal psychometrics will be absolutely necessary, and inductive inference abilities will play a central role in the analysis of potential versus actual intelligence. For a more extensive account of the notion of potential intelligence for machines, we refer to (Hernández-Orallo & Dowe, 2013).…”