Wednesday, 7 January 2015

Blindness - or modeling simplification?

One systematic problem I have noticed with many with social scientists is that they don't always seem to understand the virtue of simplified models. Any model simplification is latched on to as an inaccurate weakness - and becomes the target of criticism.

I've gone into this issue before, in the article: Parsimony in the social sciences - with Jonathan Haight as my foil.

The phenomenon is part of the trigger-happy nature of many social scientists. Historically they have been better at criticism than theory construction. As a result, many social science domains became theoretical wastelands - with researchers systematically failing to grasp even basic theories - such as Darwinism. I've previously covered that phenomenon in the article: Can cultural evolution explain the lack of theory in the social sciences?.

If the virtues of simplified models are not obvious, perhaps read Boyd and Richerson's Simple Models of Complex Phenomena. If you think that article is stating the obvious, that's great. However, many social scientists seem to badly need to read this type of content.

This time, around it is Pascal Boyer who's rubbing me up the wrong way - by being critical of modelling simplification. He objects to lumping the various forms of copying in cultural evolution together - calling it "cognition blindness". Apparently all the different forms of social learning have their own unique cognitive quirks - and a failure to build detailed models of them makes you guilty of "cognition blindness". Here's Pascal:

Cognition-blindness is reponsible for many of the horrors we hear and read in the domain of cultural evolution - e.g. that culture is transmitted because people internalize norms, because they engage in ‘learning’, because they can imitate, because they copy prestigious individuals, or they copy what works, and so forth. If people tried to formulate these common ideas in computationally tractable terms, they would see how problematic they are. But there’s the rub – folk-psychological notions like ‘copying’ or ‘learning’ seem to describe self-evident, straightforward processes, so that many people think using such terms actually describe mechanisms.

This seems to be a pretty ridiculous comment to me. When aspects of cognition involved in copying are ignored, that's generally a modelling simplification. In much the same way, biological modellers frequently assume that the population is infinite, that mutations are random, or that mate choice is random. It isn't that they think that these things are literally true, it's that it is often useful to build models which ignore these elements - in order to better concentrate on other ones.

Of course, there's such a thing as over-simplification. Sometimes simplifications really do eliminate important features or result in inaccurate models or predictions. However, step one in understanding when a simplification is an over-simplification is to appreciate the virtues of simplified models - so that you can perform a cost-benefit analysis. Labeling modeling simplifications as a form of "blindness" probably doesn't help researchers to perform a clear-headed analysis.

No comments:

Post a Comment