Model building for fun and profit

By | November 22, 2017

Recently, there was an interesting question posted on Quora, the question-answering site: “Are all sciences derived from physics?” I enjoy participating in these discussions from time to time, and I liked this one since it fit into some ideas I’d been discussing with students recently. So among the 27 answers generated, you’ll find mine tucked in there. I liked it enough to reprint it here:

“The common denominator in all sciences and their hierarchy as discussed in these sconces is modeling – that is, the creation of abstract representations of parts of reality that purport to define the key aspects of causality affecting particular phenomena. The sciences differ considerably in their level of analysis, and therefore the efficiency of their models. The more the more complex the phenomena under discussion, the more abstract the models become, and the less efficient the models are at prediction. General Systems Theory is the overarching framework within which a hierarchy of models can be proposed.

Here’s a simple representation of a model. The “things” are phenomena; the straight arrows represent presumed causal or at least influence connections, while the curved arrows represent correlations that are presumed to be non-causal:Modeling is necessary because in the real world, everything is connected do everything else and therefore everything is to one degree or another the cause of everything else. My mood on any given day is to some degree affected by the position of the planet Uranus. But the degree of connection here is so small and remote that for all practical purposes, I don’t have to pay attention to it. Thus, when I make a model intended to help predict my mood, I can pretty safely disregarded astronomical influences.

A model is defined at a particular level of generality. If I’m working at the level of chemistry, then the “things” might be individual molecules of particular chemicals. If I’m working at the level of an individual human being, then the “things” might be attitudes or behaviors. But the dynamics of modeling are the same.

As I said, the efficiency of modeling is limited by the ability to specify the “things” or the elements of the model in precise detail. The so-called “hard” sciences deal in elements that can be quite precisely defined and replicated. One water molecule is pretty much like every other water molecule. On the other hand, the “soft” sciences such as psychology, economics, and even sociology tend to deal in concepts and constructs that are much less precisely definable, such as “attitudes”, where measurement issues and measurement error become significant problems. Even when there is conceptual agreement on the definition of something like an attitude, there may be multiple ways of measuring it. The less precisely defined a concept is, the more confusion there will be about its appropriate measurement.

A model is defined at least as much by what it excludes from consideration as by what it includes. Implicitly, the specification of a model says that these are the only things that really concern us, and it is safe to effectively disregard all the other influences from everything else in the universe, or at least consign them to what we call “error variance”.

Most of the debate around whether or not things like economics and psychology are “sciences” revolves around whether or not the things being modeled are precisely defined enough to make the very process of modeling useful, and whether the exclusion of other factors makes sense. Some argue that the element of subjectivity in the definition of constructs at the highly complex levels of individuals and groups and societies renders the whole exercise meaningless. Others argue that although the constructs may be defined at relatively high levels of abstraction, they can still provide useful guidance and suggestions about how things are connected to other things. There is no particular empirical way of resolving this disagreement.

In the sense that all “systems” are essentially abstractions out of the one and only real world, each level of systems is built upon a simpler variety of systems. Physical systems tend to be both the most precise and the most specific; chemical systems are somewhat more abstract despite being built upon the foundations of physical systems; biological and ultimately social systems are built upon these lower and more specific systems.

In this sense, physics is indeed the foundation of all sciences, in that if you could completely specify everything physical, you could come close to specifying everything built upon that. The problem is that any physical system that presumed to represent, say, the behavior of a conscious individual would be so complicated and so interconnected that it would not be particularly useful. That’s precisely why we abstract concepts and constructs to represent the behavior of lower-level systems. We can usefully use a concept like “love”, even create empirical measurements for this concept that we can use in models, without necessarily specifying all the physical details that go into its manifestation.

So the simple answer is yes, physics does underlie most of the phenomenon that concern other sciences. However, the fact that they employ more complex and abstract concepts than does physics (in the interests of manageability) does not invalidate them as sciences. It simply means that we need to apply different degrees of expectation about the precision of their predictions. Models are never true or false in any overall sense, since any model is a simplified representation of the real world. Models are judged on their utility rather than their ‘truth’”.

I’ve received all of three upvotes for this, as opposed to the leading answer with some 1,500. Interpret this as you will.