Saturday, 28 March 2015

The Mismeasure of Minds

In order to tell whether something is fixed, one needs something else that is known to be fixed and can serve as a criterion of judgement. But how can one find that first fixed point? We would like some nails in the wall to hang things from, but there is actually no wall there yet. We would like to lay the foundations of the building, but there is no firm ground to put it in.” —Hasok Chang (2009)
When we perform a skilful action like catching a ball, is it necessarily the case that the brain also performs a series of skilful measurements and calculations? Are our skills of perception and intelligent action evaluative? Many theorists confidently assert that they are. I aim to explain why I think such an assumption is both explanatorily extravagant and a hindrance to enquiry.

The word “calculate” derives from the Latin word "calculus", which once referred to the stones used as counters in abacuses. In standard usage, to calculate something is to determine a value by the use of various mathematical operations applied to quantities represented by symbols. It is possible to perform many calculations without the use of symbols, but the demands of doing so (i.e. the quantities involved), make non-symbolic calculation extremely unwieldy. Certainly we have no evidence of non-symbolic neural calculation even in the brains of simple creatures. So if it is the case that brains perform calculations, then they must be using a symbolic system to encode information. Although there is an enormous quantity of literature on the subject of neural encoding, as yet no code has been identified or unravelled. This alone should make us wary of neurocalculation.

Symbols are highly sophisticated entities. The Roman symbol “IX” bears no resemblance to the quantity it represents. Nor does the equivalent linguistic symbol “nine". Many people in the world who recognise "9" and "IX" do not recognise “nine”. Turn "9" and "IX" upside down and their corresponding quantities change. And if you say “nine” in Germany the meaning is not a number at all. It should be obvious even from these trivial examples that symbolic representation is far from straightforward. Neurosymbolic communication thus carries an extremely heavy explanatory burden. Why, for instance, would brains need symbols if they were already so advanced that they could develop their own symbolic system? As brains evolved, how did the parts that didn't know the code understand the parts that did? Why do we have no evidence of basic symbols amongst simple brains? And why are we denied access to the computational power of our own brains to such an extent that we have to sit through hours of instruction and practice to learn just a tiny fraction of what our brains are allegedly capable? None of these issues refutes neurocalculation but they do help to suggest that we have little reason to be confident about its credibility.

And what of neuromeasurement? 

In 2009 Hasok Chang published a book entitled: “Inventing Temperature: Measurement and Scientific Progress.” In it he provides a detailed history of the many complexities involved in arriving at a system for the measurement of temperature. It seems obvious to us now that water freezes at 0 degrees and boils at 100. But in fact the variables involved, like chemical impurities and atmospheric pressure as well as the fact that thermometers themselves had no standard measure, made the whole process an extremely challenging one, requiring numerous iterative improvements, insights and innovations.

Chang raises a significant obstacle for advocates of neuromeasurement. In order to measure an unknown distance for example, a standard would first be needed. But in order to establish a standard, a unit of measure would also be required. According to Chang: "This circularity is probably the most crippling form of the theory-ladenness of observation" and is the very problem that has caused well documented difficulties in every region of the science of magnitude.

One thing is certain, brains did not evolve their own inner form of science and technology. But if it took science and technology to enable the invention and refinement of our skills of measurement, then it seems extremely unlikely that brains could evolve similar techniques independently. 

Could there be such a thing as a set of evolved biological standards equivalent to those developed by culture? Are such standards actually necessary, or is there a more simple and plausible explanation for the many sophisticated skills we possess?

Imagine a single celled organism that consistently moves towards one sort of thing rather than another similar sort of thing. In that case we would say that the organism is capable of discriminating between these two similar things. Such differential responsiveness is a commonplace amongst organisms and makes up by far the greater proportion of all organismic behaviour. When the iris of the eye contracts in response to bright light as opposed to dim light or when the liver produces bile in response to fatty food as opposed to carbohydrate we do not suppose that any calculation is going on, nor any measurement. The organism is simply behaving in response the presence of one sort of thing rather than another. No choice is being exercised, no calculation, no judgement, no theorisation, no measurement, merely evolved responsiveness.

Certainly the processes involved in more sophisticated forms of responsiveness are of a higher order of complexity. But the fact of differential responsiveness gives us good reason to first explore the potential of this more fundamental form of sensory discrimination in the actions of sophisticated creatures long before we go attributing skills of neurocalculation and neuromeasurement to brains.

When a child discovers by chance that a stick can be balanced across their hand, is sensory discrimination sufficient to explain their capacity to balance the stick? I don't see why not. To be disposed to move the stick in one direction as opposed to the other in order to maintain its horizontal position would seem to be a far more straightforward answer than neurocalculation and neuromeasurement.

Skills like balancing sticks are comparative. They trade both on sensory discrimination and — crucially — its lack. To be disposed to respond to two things (or two parts of one thing, in this case) in the same way because we fail to discriminate between them in one or more respects has significant efficacy. To know how to make a stick balance across your hand is to know how to make one side of the stick behave in the same way as the other side. In essence the skill is based upon an ability to make one side of the stick match the other in respect of the forces acting upon it: to make both sides of the stick indiscriminable from one another in respect of their tendency to fall. From the point of view of artificial intelligence this is an extremely complex skill, but from a haptic point of view it is child's play.

Thanks to Jason Streitfeld for the FB discussion that led to this post.


Post a Comment