A three-component cognitive theory
Arjonilla García, F.J.
MetadataShow full item record
Artificial Intelligence and robotics are developing at a fast pace. However, implementation of general intelligence in computers remains elusive. Currently, there are no technologies able to solve unexpected challenges nor learning algorithms that outperform their initial requirements. This thesis takes an unconventional multidisciplinary perspective on cognitive systems to propose a universal theory of cognition that can be identified in systems exhibiting intelligent behaviour. I present a theory that is based on a formalization of trial and error that is composed of three cognitive components and has the capacity to produce intelligent behaviours. In absence of models, tentative trials to reach a specific goal are inevitable until one succeeds. Each attempt at fulfilling a goal is termed a variant. The cognitive components are: a substrate, a generator of variants and a selector. (1) The substrate holds the information that shapes each variant and it may be either physical or immaterial in nature. It is linked to environmental mechanisms that interpret the instructions conveyed by each variant. (2) A cognitive generator provides the heuristics to produce variants, and (3) the selector chooses amongst the generated variants which one is the most adequate for the pursued goal. Then, I argue that there are three families of intelligent-behaving systems that give evidence to the theory. Firstly, I propose a biologically-based cognition that relies on principles of evolutionary theory. Particularly, I give an alternative interpretation of Darwinism that diverts away from the traditional notion of evolution by chance and credits biological evolution with cognitive capabilities. Secondly, I identify a mapping between the theory and the latest advances in neuroscience and experimental psychology. Specifically, attention drives selection and variants are represented in neural modules. Thirdly, I explore methods in artificial intelligence and I justify their cognitive limitations. I discuss a comparison of these cognitive families where more analogies are drawn, including a description of a putative sequence of cognitive emergences. Finally, I deduce from the theory a novel cognitive rchitecture that does not rely on preconstructed models to interact with the environment.