Does AI Solve the Knowledge Problem?

Discussing if modern AI fills the knowledge gap for modern Socialists

Big data, data analytics, and artificial intelligence do not solve the knowledge problem inherent to the Socialist economic calculation. That large firms like Amazon utilize consumer data to predict purchase decisions is not indicative of consumer data’s ability to predict economy-wide planning decisions. These truths were outlined concisely in Jesús Fernández-Villaverde's article for The Journal of the Witherspoon Institute. Fernández-Villaverde suggests that consumer data is associative rather than causal, that data samples may not be large enough regardless, that historic consumption may not be predictive of current consumption, and an application of the Lucas Critique to economic planning. I would like to take his criticism further. In a report authored by Evgeny Morozov for the New Left Review, Morozov proposes that big data can, in fact, solve the knowledge problem in a Socialist system. However, the classical critiques offered by Hayek apply strongly to many of the claims made – despite the author’s clear awareness and even understanding of the Hayekian position.

Morozov, and many big data Socialists, advocate for some buzzwords in their quest for a digital utopia. Purportedly, big data allows for a decentralized plan built from the ground up on consumer preferences because big data can predict our buying habits. The system Morozov advocates for is structured as follows. Firms will be replaced with workers’ councils that will list their product or service offerings in the public domain prior to the beginning of the production cycle. Consumers will submit their needs and preferences to the workers’ councils, who will then construct a production plan utilizing this information. The workers’ councils will then submit a request for the inputs required to produce up to their production plan. Morozov advocates for incentives to be put into place to discourage consumers from purchasing more goods after the initial distribution. There are a few obvious problems with this system. Beginning with the fact that this is not a decentralized structure. Despite data inputs coming from the consumer level, the distribution of inputs still flows from a central planning power whose decision-making must be constrained by scarcity rather than the wishes of workers’ councils. That is simply a centralized system with terminological window dressings, and as Hayek would suggest, most people would find the lack of consumer sovereignty to be repellant.

More deeply, data cannot solve the knowledge problem. As Mises outlines, without private ownership of the means of production, there will be no market for the means of production. This means there will be no prices established for the means of production, and thus, no information from which the planners might distribute the means of production. Fernández-Villaverde effectively outlines the Mises critique in a few sentences.

“The objections to central planning are not that solving the associated optimization problem is extremely complex, although it is. If that were the only problem, AI and ML could perhaps help us solve the problem. The objections to central planning are that the information planners need is dispersed and, in the absence of a market system, agents will never have the right incentives to reveal it.”

Put in simple terms, the problem with planning is not that it is complex but that the information required to plan is either unobtainable or not incentivized to be revealed absent the price system. Fernández-Villaverde goes on to suggest that machine learning finds associations, not causal relationships. So even the data discovered and aggregated by advanced machine processes cannot predict causal relationships on the individual level. This is because the consumer data companies like Amazon track do not predict the individual consumer specifically, but compares the purchase and search history of a consumer as compared to a large number of others who have purchased and searched for similar items. When Amazon suggests a product to purchase next, it is not learning your preferences but applying a set of potential preferences based upon a probability model of many other consumers. Big data is not individualized, it is generalized. Fernández-Villaverde notes this by shifting focus to problems of data in general. He considers the problem of external validity, where it is possible one study may not apply outside of its sample. As well, that the world is constantly changing, calling into question how applicable data from years ago is to now. Fernández-Villaverde compares the machine to the economist, they are both constrained by the current context in which they reside. Yet the machine is a poor economist because the machine lacks a decision-making theory and an understanding of incentives. AI could in no way offer the competent planning Morozov advocates for because of these constraints.

“Digital” is simply a new coat of paint. The Socialist plan remains subject to the very same Hayekian problems that it was in the 1920s. The first and most direct critique is that even centralized systems require prices to inform decisions. Morozov’s preferred system relies on central control of inputs — the means of production. The central power must distribute those inputs under the constraint of scarcity, but without prices to convey information about which destinations need these inputs the most, the planners are operating in the dark. Further, in the absence of profit, the incentives for workers’ councils to operate optimally are poor. A point that Morozov himself highlights,

“… as long as the quest for profitability remains the overarching objective of the entire system, everyone knows what to expect. Of course, if this condition does not apply, the price system immediately loses its coordinating magic.”

Without the coordination of prices to the ever-changing economic system, planners will not be able to dynamically distribute inputs to workers’ councils in an informed and reasoned manner. That Morozov acknowledges this yet still advocates a system outside of price is befuddling. As was the case during the original Socialist Calculation Debate, Digital Socialism misconstrues the role of information and knowledge. Hayek distinguishes knowledge to be that which is decentralized and unevenly distributed and that which is tacitly known at the local level. As such, the decisions of consumers and the process of discovering prices by firms are done through what Hayek termed Tacit Knowledge. This knowledge is obtained through the active search of it. Rather than the innate knowledge which is simply known by some. When applied to a Socialist calculation, this knowledge is either unobtainable by the central planner or inefficient due largely to the absence of price signals. The wave of Digital Socialism mistakes information for knowledge, committing a far greater error than even the original Socialist supporters. Morozov describes the Hayekian understanding of prices conveying information as that of an aerial snapshot of a battlefield. Though apt, prices convey information that informs knowledge as consumers or firms actively search for answers. Morozov seems to suggest that information itself is knowledge. In doing so, he advocates a new system that maximizes information absent the price system. However, information is neither knowledge nor is it magic. Morozov asserts that consumer data compiled in large enough numbers creates a wealth of information that can accomplish anything short of wingless flight. Yet he fails to outline precisely what information would be needed to construct such a system, how that information might be accurately gathered, and how the requisite knowledge to reach optimal outcomes will be discovered through self-reported consumer preference data. That consumer data can aggregate associative preferences does not indicate that consumer data can then discover the requisite information required to seek knowledge and coordinate through it. Even in the current state of the market, big data is only information. It is the knowledge that Amazon has which interprets that information and applies it to maximize profit, not simply the data itself.

              As Fernández-Villaverde notes, the most foreboding question today must be answered the same way it was in the early 1900s. For all the flaws and negative outcomes within a free market system, the alternative is far worse. This is true both politically and in terms of market outcomes. His most poignant insight arrives at the conclusion of his article. Recalling Hayek, Fernández-Villaverde asserts that the rules inherent to the market system were not crafted by red tape or the result of a grand thought experiment from a legislature. The rules of the free-market system came about naturally through human interaction – they’re evolutionary. Hayek terms this thought spontaneous social order. The idea that social systems are the product of human interaction but are never designed. As such, it would be impossible to construct a more rational system. To do so would be to ignore incentives and overestimate our own intellect. Morozov accuses the Hayekians of not knowing how to run the system they’ve created, this vastly misses the point. Hayek sought not to run an economic system, he sought to ensure the continuation of an economic system which runs itself. Now, as it was in Hayek’s day, the debate still rests on knowledge and misguided hubris. Big data does not solve the knowledge problem, just as trial and error did not solve the knowledge problem. Digital Socialism will be doomed to failure as the USSR before it. The argument against Digital Socialism, or any form of Socialism, is far deeper than to say Capitalism is flawed but Socialism is worse. It will behoove modern economists to remember this as these debates progress.