Does AI make central planning work?

Those who thought that economic production would become more optimised if centrally planned were given a boost by the development of mainframe computers. It was wishful thinking. I have a friend from Eastern Europe whose job was to collect the data to feed into such computers. He told me his figures were simply binned, as the computer operators knew that solving the problem of production was well beyond their technology.

Each breakthrough in IT brought some of the same hopes, and same dashed expectations. And today there many who believe that the arrival of AI at last give us the tools to succeed at rational economic planning. Well, dream on. 

True, AI is impressively powerful, particularly at classification and pattern forecasting. It can predict probable next events, up to a point. But it has problems regarding causation. And it cannot predict what is in the minds of the human beings whose choices and actions are the economic system that it is being asked to manage.

As F A Hayek realised, the problem in economic planning is not one of computation. It is the fact that the knowledge it requires is widely dispersed over millions of individuals. Only those individuals know their own minds — and they might not even know what they would choose until actually presented with the choice. (How many of us have gone into a shop and come out with something quite different?) Much of the required knowledge is practical — knowledge how to do something rather than knowledge that something is true. And such knowledge cannot be transmitted to a central planner anyway. Through years of experience, a producer might understand their local market. But they are probably quite unable to write down the principles by which they understand it. When they take on apprentices, they in their turn have to acquire that experience. There is no manual. Economic reality is not a set of fixed facts; it has to be discovered and learnt. No entrepreneur knows precisely whether their product will be favoured by customers, or what price they might be willing to pay. Make the wrong prediction, and time, energy and capital have to be written off.

In any case, it is the height of conceit to imagine that one single entity, even an AI-powered one, can possibly handle all the information that exists in the minds of millions of individuals, some of it right, some of it wrong, all of it dependent on context, locality and the motivations of many individuals. There are, as Hayek said, millions of planners already. Individuals and businesses make plans on the basis of the best information available to them. Much of that information cannot be transmitted to some central authority, and much of the information that is transmittable would be out of date before it got there — never mind before it was acted on by the central authorities.

That may sound like a counsel of despair, but it's a conclusion that makes me rather happy. Because dispersed information means dispersed power. As I look at people in the street, all happily interacting through their mobile phones, I rejoice in the thought that no central power could possibly even absorb all the information they are constantly creating and moving between each other. A central authority that could do that would be a dismal threat to human freedom. Central authorities who even try to do that are bad enough.

Eamonn Butler