Wednesday, November 20, 2019

Will Artificial Intelligence End the Austrian School of Economics?

Peter Thiel

By Robert Wenzel

Last week Wednesday, Silicon Valley billionaire Peter Thiel delivered the 2019 Wriston Lecture at the Manhattan Institute. During the Q&A following the lecture, the topic of artificial intelligence was raised and Thiel discussed it within the context of Austrian economics.

He said (my highlight):

AI is sort of the buzz word of the day. It could mean the next generations of computers, the last generation of computers, anything in between. It can mean the Terminator movie where it's a robot that kills you, it can mean sort of all these sort of creepy social credit scoring things in China. But in practice, the main AI applications that people seem to talk about are using large data to sort of monitor people, know more about people than they know about themselves. And in the limit case, maybe it can solve a lot of the sort of Austrian Economics type problems where you can know enough about people that you know more about them than they know about themselves, and you can sort of enable communism to work, maybe not so much as an economic theory, but at least as a political theory. So it is definitely a Leninist thing. And then, it is literally communist because China loves AI; it hates crypto. And so that, I think, tells you something. And then I think there's a commonsense level on which people are creeped out about it and this is why. And we should label it accurately.
Thiel, by singling out Austrian economic theory, seems to suggest that AI will solve some problems that Austrian economists raise.

He is really chomping at the epicenter of Austrian school economics. If he is correct it would end most of what is known as Austrian economics.

But, for the record, both Austrian school economists Ludwig von Mises and Friedrich Hayek discussed the supercomputer situation decades ago, long before there were even smartphones or personal computers or much in the way of computers.

For example, the essay, The Equations of Mathematical Economics and the Problem of Economic Calculation in a Socialist State, was first written by Mises in German in 1920 as "Die Wirtschaftsrechnung im sozialistischen Gemeinwesen."

In it he wrote:
[E]ven if we know the present conditions, we are unable to say anything of a quantitative nature, on the basis of this knowledge, about the pattern of future values. This is the big mistake that has been made by all those who have wanted to substitute “quantitative” economics for “qualitative” economics. A quantitative treatment of economic problems can only be economic history: it can never be economic theory. And there is no economic history of the future. The equations which describe the state of economic equilibrium include consumers’ preferences. These are the preferences which will prevail at the moment when the equilibrium is established on the market. They are different from today’s preferences as we know them from the way in which they are expressed on today’s market. Today we know nothing about these future preferences and cannot predict what they will be. Thus, though we may know the present-day condition of the market and all the data determining the configuration of today’s market position, including consumers’ preferences as they are expressed in that market position, we still do not know the future preferences of consumers.

We may be justified in assuming that they change. This assumption does not help however. For the economic system is not in equilibrium today, and we want to know the consumers’ preferences for the point of time when it will be in equilibrium and when, in consequence, other conditions will prevail. The progressive approach of things towards an equilibrium situation which we have in mind, and which forms the subject of our inquiry, means the progressive transformation of the conditions determining the preferences and therefore also of the preferences themselves. The problem is not only that, in order to make use of the equations, we need to know the scale of preferences that will prevail at a future point of time and which are not known to us today. Even today’s preferences are only known to us in so far as they are reflected in the system of prices ruling on today’s market. That is to say we know roughly how great is the demand for a certain article by the price prevailing for it on the market today. But we know nothing of what the demand would be if another price prevailed. We do not even know the shape of the supply and demand curves; we only know the position of one point at which the two curves cut or, more precisely, have cut today. Experience tells us so much and no more. It can provide us with no information about the data which we require for solving our equations.

Finally there is still a third point which needs mention: The state of equilibrium which our equations describe is a purely imaginary state of equilibrium. It is merely a hypothetical, though indispensable, tool of analysis which has no counterpart in reality. Thus it is not only a future state which differs from the state of the moment that has just passed and with which we are acquainted: It is merely an imaginary theoretical construction which will never become reality. Hayek (1935, p. 211) has also pointed out that the possibility of using the equations describing the state of equilibrium for purposes of economic calculation presupposes a knowledge of the future scales of preferences of consumers. But here he has in mind only a complication of the practical task of applying the equations, and not a fundamental and insuperable obstacle to their use for any such process of calculation.
And Hayek makes the important point:
We should not expect equilibrium to exist unless all external change has ceased.
If there is external change, Hayek's implication is that some of these changes will not be known in advance and thus these changes in the environment will change individual valuations in unknown ways that even artificial intelligence can't anticipate.

No, artificial intelligence will not "solve" problems of changing valuations.

On a further point, Thiel in mid-sentence during his answer, after saying it is about Austrian economic questions, oddly says it may not be so much about economic theory but political theory.

I have no idea what Thiel means when he says that communism might work from a political perspective because of AI but not from an economic perspective. Communism is about the nature of the economic structure. It can not be planned if unknown changes take place.

Robert Wenzel is Editor & Publisher of EconomicPolicyJournal.comand Target Liberty. He also writes EPJ Daily Alert and is author of The Fed Flunks: My Speech at the New York Federal Reserve Bankand most recently Foundations of Private Property Society Theory: Anarchism for the Civilized Person Follow him on twitter:@wenzeleconomics and on LinkedIn. His youtube series is here: Robert Wenzel Talks Economics. More about Wenzel here.


  1. Thiel'scomment is sort of incoherent. Usually expect more thought out commentary from him.

  2. Today's tech is "AI" compared to 100 or 200 years ago. Scarcity, choice, and imperfect knowledge have not gone away.

  3. We are a long way from omnipotence. Isn’t that what central planning needs to work as it is endorsed to the masses? Even the Judea Christian God gave man free will. I doubt this God could solve the issues Austrian Economics brings up. Can man build an intelligence (or even an intelligence that builds an intelligence) that accurately predicts the future of all existence?

    I’m not holding my breath. It also sounds like a very boring world.

  4. Artificial intelligence is not a cure for natural stupidity.

  5. The Austrian argumentation WRT economic calculation problem relies on the fact that economic actor's utility functions are hidden and not observable through anything other than voluntary trades. This makes the free market the only economic system which is provably weakly Pareto-optimal, beating anything else.

    What Thiel said is that hypothetical super-human AI could in principle learn so much about every economic actor in existence that it could reconstruct the good approximation of the hidden human utility functions. It principle this does resolve the economic calculation problem as information needed for economic calculation is no longer destroyed by the act of managing the trade, which does open possibility of achieving stronger form of optimality.

    In reality, of course, no such AI exists or projected to exist any time soon. Additionally, such AIs would themselves be economic actors, opening the question of super-super-AI capable of learning approximations of utility functions of these lower-level AIs, ad infinitum, so it just recreates the economic calculation problem at a higher level.

    Note that knowledge of human utility function is equivalent to complete description of the inner structure of the mind: i.e. reconstructing it creates human-level strong AI perfectly mimicking the specific human. In other words: not going to happen any time soon, if ever.