I write flight simulation software for a living, and thus the machine at my desk has to have a relatively recent graphics processing unit (GPU). Maybe not the most bleeding edge GPU, but something at least relatively recent. Thus one (small) requirement of my job is to maintain an acquaintance with the capabilities of GPUs as they evolve over time.
One capability that I have been watching with growing interest is the use of GPUs in biotechnology applications, specifically protein folding. The research into a complete understanding of protein folding is exciting for many reasons: it has the potential to cure devastating illnesses, it requires immense computational resources, and it gives insights into how non-mechanical machines work. In the last few years, the folding at home project supported by a biotech lab at Stanford has written a couple versions of their protein-folding software to take advantage of the linear algebra capabilities of modern GPUs. For instance, the lab has just released the second version of their GPU software in their effort to keep up with the state of the art in GPU hardware.
Of course, the reason that efforts like folding at home are interested in wringing the last cycle of performance out of modern GPUs is that their computations are of enormous complexity – to achieve accurate results, the proteins being studied must be simulated at the atomic level. Since each protein molecule contains hundreds if not thousands of atoms, the software that simulates the physical interactions between all the atoms must perform billions of linear algebra calculations.
So what does protein folding have to do with Mises? Protein folding requires a focus on each, individual atom. Protein folding algorithms must calculate the actions of each individual atom to achieve results worth more than a bucket of warm spit. Each atom is vitally important. Any “statistical” efforts to understand the protein “as a whole” are worse than worthless. You have to do the math, per atom. If you have to use statistics, you really don’t know what you’re talking about.
And, as Mises showed again and again, the same principle applies to economics. To achieve any results worth more than a bucket of warm spit, you have to consider the individual actors. Period. Anything else is mere guesswork. And in the age before supercomputers, Mises had the courage to say so. Which takes a lot of courage – Mises had the integrity not to spew a lot of fake “mathematics” to cover his (and everyone else’s) necessary ignorance of economics “in the large.” Would that more economists followed his example.
I say “the age before supercomputers” in the hope against hope that immense computational resources might help us to achieve results deeper than what Mises could achieve – results based on the computation of the interaction between individual actors. But I think Mises would probably deny that even supercomputers would help economics. They certainly help for very complex physics problems: weather prediction, protein folding, atom smashing, and so forth, but of course human reasoning does not follow physical laws.
Being the curious person that I am, I’m always interested to see if some economist can achieve deeper, more fundamental results than Mises did. Unfortunately, without an accurate model of human action based on linear algebra, I’m afraid that supercomputers aren’t going to help all that much.