Simple environments fail as illustrations of intelligence: A review of R. Pfeifer and C. Scheier: 'Understanding Intelligence

The field of cognitive science has always supported a variety of modes of research, often polarised into those seeking high-level explanations of intelligence [2, 13] and those seeking low-level, perhaps even neuro-physiological, explanations [3, 9]. Each of these research directions permits, at least in part, a similar methodology based around the construction of detailed computational models, which justify their explanatory claims by matching behavioural data. We are fortunate at this time to witness the culmination of several decades of work from each of these research directions, and hopefully to find within them the basic ideas behind a complete theory of human intelligence. It is in this spirit that Rolf Pfeifer and Christian Scheier (hereafter, P&S) have written their book Understanding Intelligence. However, their aim is manifestly not to present an overview of all prior work in this field, but instead to argue forcefully for one particular interpretation – a synthetic approach, based around the explicit construction of autonomous agents. This approach is characterised by the Embodiment Hypothesis, which is presented as a complete framework for investigating intelligence, and exemplified by a number of computational models and robots to illustrate just how the field of cognitive science might develop in the future. We first provide an overview of their book, before describing some of our reservations about its contribution towards an understanding of intelligence