Thursday, February 2, 2017

Austrian School Economics, Praxeology and Artificial Intelligence

Patrick O'Connell emails:
After reading this article does this affect the debate regarding praxeology and econometrics?  I would love to hear your thoughts and that of your contributors:
 http://www.nextbigfuture.com/2017/01/carnegie-mellon-artificial-intelligence.html

RW response:

Praxeology, as Murray Rothbard put it,  "rests on the fundamental axiom that individual human beings act, that is, on the primordial fact that individuals engage in conscious actions toward chosen goals."

Artificial intelligence did nothing to stop the poker players referenced from acting, that is, choosing goals. All AI did in the referenced example is make quicker and better calculations than the poker players, in a given fixed environment.  This is an example for all practical purposes of a jacked-up hand calculator. Such calculators can quickly multiply out to 9 digits much faster than almost all humans but this has nothing to do with eliminating the need for humans to make choices.

But because all human choices are unknowable in advance and many choices change over time there is no set calculation for AI to calculate when it comes to human action, that is, there is no exactly repeating history.

As Rothbard wrote:
[The economist Ludwig von] Mises indeed held not only that economic theory does not need to be "tested" by historical fact but also that it cannot be so tested. For a fact to be usable for testing theories, it must be a simple fact, homogeneous with other facts in accessible and repeatable classes. In short, the theory that one atom of copper, one atom of sulfur, and four atoms of oxygen will combine to form a recognizable entity called copper sulfate, with known properties, is easily tested in the laboratory. Each of these atoms is homogeneous, and therefore the test is repeatable indefinitely. But each historical event, as Mises pointed out, is not simple and repeatable; each event is a complex resultant of a shifting variety of multiple causes, none of which ever remains in constant relationships with the others. Every historical event, therefore, is heterogeneous, and therefore historical events cannot be used either to test or to construct laws of history, quantitative or otherwise. We can place every atom of copper into a homogeneous class of copper atoms; we cannot do so with the events of human history...
Mises's radically fundamental opposition to econometrics now becomes clear. Econometrics not only attempts to ape the natural sciences by using complex heterogeneous historical facts as if they were repeatable homogeneous laboratory facts; it also squeezes the qualitative complexity of each event into a quantitative number and then compounds the fallacy by acting as if these quantitative relations remain constant in human history. In striking contrast to the physical sciences, which rest on the empirical discovery of quantitative constants, econometrics, as Mises repeatedly emphasized, has failed to discover a single constant in human history. And given the ever-changing conditions of human will, knowledge, and values and the differences among men, it is inconceivable that econometrics can ever do so.

2 comments:

  1. I fail to make the connection between this poker game and human action vs econometrics.

    This poker bot is very interesting and I've been following this game pretty closely. Poker is a very complex game. This is quite an accomplishment for AI, a bigger accomplishment than Deep Blue defeating Gary Kasparov.

    I think if I were one of the poker players I'd have played very fast very early. I assume the computer is going to value bet, fold and bluff with optimal frequency on later streets. The only hope is to win a push/fold game. Reduce the computer's massive processing power to simple binary decisions and hope to cooler the bot or get lucky.

    ReplyDelete
  2. “... AI’s ability to do strategic reasoning with imperfect information has now surpassed that of the best humans,” Sandholm said.
    "But each historical event, as Mises pointed out, is not simple and repeatable; each event is a complex resultant of a shifting variety of multiple causes, none of which ever remains in constant relationships with the others. Every historical event, therefore, is heterogeneous, and therefore historical events cannot be used either to test or to construct laws of history, quantitative or otherwise" Rothbard
    I believe that these two quotes comment upon the same situation. We are dealing with imperfect information and as Rothbard was describing historical events that cannot be used to test or to construct laws of history quantitative or otherwise. This s a definite step up in the world of AI. This is an advance over machine learning. This is not about poker as much as strategy and risk taking.

    ReplyDelete