Rethinking Cognitive Computation: Turing and the Science of the Mind. by Andrew Wells
Palgrave Macmillan, 265 pp, ISBNs 1 4039 1161 4 and 1 4039 1162 2
Times Higher Education Supplement, Feb 2006, p 29
(This is the version submitted. It may have been edited before publication.)
Imagine that you wake from a dreamless sleep and find yourself in a cube-shaped room with nothing in it but a door marked “R”. You have no memory of your life before this moment and no emotion. It occurs to you that the room would benefit from a sheep, so you say “sheep”, and one appears. This scenario is one of Andrew Wells’ descriptions of what it might be like to be a Turing machine.
In Rethinking Cognitive Computation, Wells revisits many of Turing’s famous ideas on computation, using diagrams of what he calls “mini-minds” as a way of explaining the functions of finite automata. He argues that these ideas naturally lead to an ecological or situated approach to computation, although this aspect of Turing’s work has rarely been appreciated. On this basis he criticises the major ways in which Turing’s work has been applied to understanding human thinking.
The traditional computational theory of mind treats the human mind as functionally equivalent to a universal Turing machine. In some versions the brain is likened to the hardware and the mind to the software of such a machine, leading to the functionalist conclusion that psychology can be studied independently of neuroscience. More recently this has given way to approaches recognising that actual brains have evolved to solve real world problems. Then connectionist theories have explored the behaviour of simulated neural networks whose structure is thought to reflect the structure of actual brains.
Wells criticises all of these. The usual computational theory of mind, he argues, requires implausible input mechanisms and places an insupportable burden on the idea that the organism constructs internal representations of its environment. Computational psychology has, he says, wrongly taken its inspiration from von Neumann computers rather than from Turing machines. By contrast, connectionist models treat the mind as a finite automaton but fail to recognise the importance of structure in the environment. So Wells tries to redress this by examining Turing’s theories in detail and demonstrating that a Turing machine is really a model of a mind interacting with symbol structures in its external environment, and that this leads naturally to an ecological perspective.
He then proposes his own new approach that he calls ecological functionalism. Ecological functionalism claims that brain structure and environmental structure are equally important for cognitive processes. It treats the mind as a finite automaton rather than a universal machine, and stresses that it interacts directly with its environment rather than with an internal symbolic representation of the environment, and there is no symbolic language of thought. This new approach has in common with previous computational theories the idea that thinking is a kind of computation, but it stresses that thinking is not a product of the brain alone, but rather of the whole organism situated in its environment. In the last part of the book Wells explores how this new approach might deal with behavioural flexibility, adaptive and goal-oriented behaviours, the use of language and symbols, development and learning.
Ecological functionalism is a welcome approach, and one that fits in important ways with developments in neuroscience and psychology, but Wells fails to place his new theory firmly within this context. He briefly discusses J.J. Gibson’s well known ecological theory of perception, proposed in the 1960s, but does not consider the recent flurry of new theories that take a similar ecological or embodied approach. These include the theoretical work of psychologist Nicholas Humphrey, neuroscientist Francisco Varela, and philosophers such as Susan Hurley and Andy Clark, and there is also lots of recent empirical work inspired by these enactive or embodied theories of cognition. Perhaps the most extreme example is Alva Noë and Kevin O’Regan’s “sensorimotor theory of vision” in which there are no internal representations at all and vision becomes a kind of action. Wells’s book would have benefited greatly from some discussion of how his ecological functionalism relates to all these exciting developments in understanding action, perception and cognition.
I also wondered just who the book is aimed at. Wells expects his readers to do lots of increasingly complex exercises as they go along. I coped with the first few, struggled with some more, and then gave up, acknowledging that such tasks are not for the general reader or for me. Although Wells claims that the book is aimed at psychologists, I fear that much of it requires the skills of a competent student of maths or logic rather than psychology.
At some points I was deeply confused and was not always sure whether this was because of my shortcomings or his. For example, right at the beginning of the book Wells derides several descriptions of a Turing machine given by other authors, concluding that “Some are incomplete, some are simply wrong, others are misleading”. One of these descriptions refers to the machine erasing and changing a symbol, but this, Wells claims, is wrong for “The machines in Turing’s 1936 paper never erase a zero or one once it has been printed.” (p 5). Yet later on he quotes directly from that very paper in which Turing says “in other configurations it erases the scanned symbol.” (p 77).
These problems aside, Wells gives new insights into Turing’s ideas, proposes new ways of thinking about computation, and his theory of ecological functionalism is relevant to any psychologist, neuroscientist or philosopher who is interested in the vexed question of whether the human mind can be understood as a computer.