This post starts with what might seem as a discussion about computing, but is actually a poor man’s discussion about philosophy of science and has nothing to do with computing, it is much more applicable to biology, economics and sociology.
Lets be honest, computer scientists are trained to work for banks and insurance companies, to make web sites, software for cars and things like that. Those domains are actually very simple. A bank might be a gigantic institution, but it is possible to capture, granted with a lot a effort, all its processes inside a computer program. This creates a mental setting: Everything that we need to know is possible to be known: we just decide when to stop.
Now think about those simplistic (this is an understatement) mathematical and computational models for scientific problems (differential equations, Monte Carlo processes, Markov Chains, …). They model the “important parts” of the issue under study. These models are much simpler than the models working in computers to sustain day to day banking chores. Somehow it strikes me as strange that something as mechanic as a bank needs a more complex model than “nature”.
In the context of nature and making mathematical and computational models about it, I have a few things in mind:
First of all, in many problems in the natural world we don’t know what are the important parts to start with. This is very different from the “bank mentality” when you can know everything if you try hard. In my personal case, when I model malarial artesunate resistance, I am modeling something that people speculate how it works, and even if the speculation is correct most of the fundamental parameters are unknown. I am still to read a paper modeling something related to malarial drug use that doesn’t have a phrase like: “the relation between this value is and reality is assumed to be this (no citation - or citing something unpublished - or rationale provided)”.
But the cornerstone of my reasoning is that, in complex processes, the devil is in the details and in the interactions between participating factors (most of which we
are unaware of). Soft sciences are holistic by nature. The property of the whole system comes from the everything and everywhere. The “banking” and “hard science” mentality are no good here, we cannot know everything, what we know is probably not enough, and most simplifications will lose something fundamental.
Does this means that I am suggesting that we should stop modeling and all theoretical work? By no means, but we should refocus:
- This is not hard science, don’t try to mask it as such. Hard rules, sensitivity analysis are mostly artifacts to make things look more “serious” and more “demonstrated”. This is biology (or even “worse”, economy or sociology), you don’t c.q.d. here.
- Think you can forecast the future? You think you can… thaen bring me a always correct forecast of the weather in 2 months and I will listen to you. Most models that exist to forecast the future are there because they are very hard to disprove TODAY: climate (as opposed to weather) models, epidemiology, … . The vast majority of models that can be tested fail (think mathematical finance and the current subprime crisis in the USA, think weather predictions…).
- Theoretical work, although not being able predict the future (or explain the past) might help create a cognitive and linguistic framework for discussion: present the fundamental concepts and narratives underlying the research process, make the discourse clearer, less cloudy, point dangerous imprecisions. This is actually the inverse that what happens now: theoreticians speak in a language that most people struggle to understand.
- Theoretical work can create interesting questions for field scientists to try to answer: It is the precise inversion of what happens now: We don’t want models that are
cheated to lookrealistic. We want reasonable models that fail miserably so that we can ask field scientists: This is failing, why do you think this happens? Have you considered this other hypothesis? What about testing it?
<sarcasm>
The existing modeling culture is quite good in the current scientific setting: Makes theoreticians look intelligent with all those complicated mathematics and computer programs (and associated publications) and excuses “practical” scientists of even trying to use their brains: They just apply the existing theory in a process that is more industrial then creative to their research questions. The biggest example that I know of this is phylogenetic analysis: Get data from the field, compute a mutation model from the premise that a small genetic distance is better, burn CPU cycles, publish - You don’t even need a human for this - a trained monkey is probably enough.
</sarcasm>
In economics things are a bit worse: elaborate game theories and such are presented as a “hard, undisputed” justification for an economic theory serving some nice agenda. Nothing more than a authoritarian argument.
PS - If you work in an hard science like physics or chemistry you might be thinking that I am smoking something very strong. I don’t think that this post applies to hard sciences, that is a different game altogether.
Tiago
Great post. I am a quantum chemist by training, so I guess my background is more in the hard sciences. That said, I do think you are right in many ways.
Most problems that computers are used to solve are, at least in principle, limit bound and can can be addressed completely. One of the challenges, at least in my limited knowledge, for addressing natural systems is the sheer number of variables. The question I have always had is this. Is there a minimum number of variables required to reasonably model a natural system. If so, then is it just a question of compute power. Or is, by definition, the number of variables indeterminate, which pretty much means that the only option is negative control (not necessarily a bad thing).