Many tasks that were once considered difficult for computers to do are now routine. Whether transcribing a credit card number or brewing an espresso, we are served by artificial intelligence every day. While a driverless car giving us a ride across town is the new normal,the sudden emergence of powerful language models that can compose email, write papers, and even pass exams has raised other questions.
What about using AI to plan the economy? Can AI do that? Is it even possible? Some say yes. The World Economic Forum has published a video on “economic degrowth.” As the economic growth is put into reverse gear, AI, according to the video, could decide which industries should be eliminated first. @RokoMijic, a self-described “AI not kill everyone-ist” suggests that AI could plan an economic system better than the market. A commenter on the same Twitter thread thinks that communism might have succeeded if the Bolsheviks had computers.
While advanced AI is new, the idea that computers could do economic central planning is not. This was first proposed almost 100 years ago as a part of the “socialist calculation debate.” This was a historical controversy within the economics field over the possibility of a centrally owned and centrally planned economy.
Austrian economist Ludwig von Mises launched the controversy in a 1920 paper where he argued that a single centralized agency could not determine a rational use for all for productive assets without a market for capital goods. Modern economic systems have a vast accumulation of capital goods. Because these productive assets have many alternative uses, there must be a rational basis for deciding between them. In order to compare, alternatives must be reduced to a single common measure taking into account costs and results.
In the market economy, the common measure of costs and revenues is money prices. Prices reflect the value of alternative uses because multiple private firms independently value each productive asset based on how it contributes to their own business. A competitive bidding process between firms drives prices to reflect the highest and best use of each asset.
Because prices are all in units of money, each alternative can be reduced to a single net amount of money. A net positive amount is a profit, a negative is a loss. Profits are earned by the firms that are able to find the opportunities to do more, with less. In a market economy, entrepreneurs estimate future market prices in order to plan what they will produce.
Mises called this comparative process “economic calculation.” Socialism is an economic system without privately owned capital goods. Productive resources are centrally owned by the state. Without independent private owners bidding against each other, there is no competition and therefore no market prices, no profits, no losses. The choice among alternative uses for productive assets becomes a purely administrative process.
According to Mises, this problem cannot be solved administratively, leaving no solution at all. The single owner of all capital goods would have no rational basis for choosing one alternative over another. He would have no way to know whether the one set of final goods met consumer needs better than another.
Nor would there be any means to ensure that the intermediate steps in the supply chain would provide the proper quantities of parts and raw materials, in the right time and place, for production to proceed. If too many parts were produced, then resources would have been wasted. If too few, then subsequent stages could not proceed due to lack of parts, labor or some other crucial resource.
Computers can do mathematical calculations. That has always been true. A computer, or perhaps several big ones, could solve a very large number of equations in a reasonable time. The economic calculation debate began in 1920 and continued until around 1950. Modern computers did not exist at the start of that time frame, but began to appear near the end. Although not widely used in 1950, their capabilities were evident.
The Polish economist Oskar Lange proposed that central planners allocate resources without a private market. His idea was to use a mathematical model of the price system to simulate the market. Economists at the time had developed a system of equations called general equilibrium theory. These equations express the optimal use of all existing resources at a single point in time, given consumer preferences. If the market could “solve” the equations, why could socialism not solve them as well? Lange even conceived of the market economy as a “crude computer.” If it all worked as Lange had expected, then computers could calculate the prices that would be used in economic calculation.
If a market economy could be reduced to a problem of computation, then yes, a computer could solve it. But to define the problem this way was a reductive move. In so doing, the problem was defined out of existence. What F.A. Hayek calls “the economic problem” is not a computational problem. It is the problem of economizing on scarce means to achieve the most important ends:
The economic problem of society is thus not merely a problem of how to allocate “given” resources—if “given” is taken to mean given to a single mind which deliberately solves the problem set by these “data.” It is rather a problem of how to secure the best use of resources known to any of the members of society, for ends whose relative importance only these individuals know. Or, to put it briefly, it is a problem of the utilization of knowledge which is not given to anyone in its totality.
The purpose of an economic system is production. Socialism is more properly described as a centrally owned system, not a centrally planned system. The question at hand was not whether quantities of inputs could be computed. The question was whether a centrally owned system is capable of producing goods and services without consuming more valuable assets in the process. And that requires prices and calculation.
By focusing on the problem of equilibrium prices, the socialist debate team had narrowed the problem considerably from production to computation. The entire debate is remembered as a debate about planning. Planning was limited in scope to determining the quantities of inputs to be transformed into outputs, assuming known production methods.
Production, like all things that take time, requires planning. The end result, production, requires both planning and execution. The fixation on computing the quantities of inputs to be used in the socialist system ignores the execution step. Nor are the two entirely separate; the boundary between planning and execution is permeable. Some steps fall purely into one or the other, but a lot of what goes on in a business falls somewhere in between. Producers refine their plans as they are executed, and revise them as circumstances change. A plan gives the business enough confidence to start, but it takes more than a plan to finish.
Given the so-called “plan” consisting of quantities of inputs and outputs, the socialist society still would not have the ability to produce anything. As economist F.A. Hayek observed, once the computing the quantities, “would be only the first step in the solution of the main task. Once the material is collected, it would still be necessary to work out the concrete decisions which it implies.”
Economic production is – for the most part – not chemistry where 2Hs and one O are needed to make a water molecule. There exist many possible variations in both a product and the production method used to create it. A modern car contains a quantity of steel, zinc, manganese, nuts, bolts, plastics and other materials and parts. But at one time, cars were made of wood and this continues to be a possibility. Some choices are made at an early stage once a firm has ordered supplies. When a machine has been deployed to the factory floor it would be very costly to change course. At that point the firm might have to take a loss on the machine if the plan is changed.
Many other decisions that impact costs and product quality are made on a daily basis. Many decisions cannot be planned in advance and are left to be addressed during execution. The comparison of alternatives using market prices is ongoing from planning through execution. As production advances, many decisions – big or small – must take into account the competitive price system in the same way as did the earlier versions of the plan.
A construction project knows approximately the amount of materials required to build a house, but the supervisor must organize the crew and direct their labor each day to ensure that the building is built properly. Unusual weather, shortages of drywall or unexpected soil conditions must be taken into account. If a work crew is short-handed, what is the best way to economize the limited supply of labor available on that day? Should the smaller crew proceed with tasks that do not require a large number, or should temporary workers be hired? If the desired construction material is in short supply, should construction halt or a lesser quality substitute be used?
As production proceeds, the remaining cost-to-complete will tend to decrease because some costs are paid along the way, and fewer costs remain. But, If market conditions have moved far enough away from the original assumptions, then abandoning work in progress will result in lesser losses than completing the project. Walking away can be the best thing. In large cities, you may see partially finished office buildings. The initial plan was not completed. Why? The real estate developer may have run out of funds due to underestimating costs. Or due to a decline in office building prices, it no longer made economic sense to complete construction.
Within a firm production is a mix of managed and price-driven. The firm to some extent works on a centralized model, the way that socialists think that socialism should work for the entire system. People are told what to do, resources are sent from the loading dock to the department. Departments of the same firm usually do not bid against each other for the chance to fill an order. But a business plan is only detailed up to a point. Many more decisions must be made along the way. Market prices often are the deciding factor in these choices.
In most jobs, employees need to have a rough idea of the costs of the supplies and equipment they use. The rank and file employee is often the one to decide which supplies may be used more freely – when more would help – and which ones must be used more carefully – when necessary. A barista using an extra coffee filter is a negligible expense, but 100 pounds of prime cuts of steak must be refrigerated to avoid spoilage. In technology startups, the value placed on bringing a new product to market quickly dominates other costs; in those situations, “move fast and break things” is the right decision. When software operates critical equipment such as an airplane or medical tech, extensive (and costly) testing is necessary because the cost of accidents is so high.
The best or most efficient production method is not purely a technical problem. It cannot be entirely solved by computation. Production methods can only be compared with market prices because the costs of alternatives must be valued differently. In many industries best practices have been established. Firms in the same industry learn what works based on the history of what has been tried. Along the way, many things did not work and losses were suffered as a result. Production methods that succeed result in lower costs or improved products, and so contribute to profits of the initial adopters.
Production methods are not simply given to the management of firms. Improvements came about because an entrepreneur has the freedom to to try something different. If the socialist factory manager was provided with a list of inputs and required outputs, they would not be in the same position as capitalistic management in a market economy. They would not have prices to guide them in the choice of production methods and the many decisions about how and what to economize along the way.
The contribution of intelligence, skill, and decision-making in the execution of production is considerable. Some tasks can be delegated to software – AI or otherwise. But there are aspects of human decision-making that can be so easily captured. Hayek pointed out that within a specialized industry, “Most [of what we call knowledge] consists in a technique of thought which enables the individual engineer to find new solutions rapidly as soon as he is confronted with new constellations of circumstances.” Some jobs repairing or operating industrial systems are defined almost exclusively by the ability of the practitioner to resolve unforeseen issues within a reasonable time.
We have established that production involves planning and execution. Can AI help with either? Yes, surely it can. When processes within a firm can be measured and then data used to train AIs, then software can be taught to do some things well, and other things just well enough. Over time, human skill in one area may be augmented, or replaced by a computer.
As increased AI capabilities become readily available, they will be offered on the market, for a price. AI, robots, and computers will replace human labor under the rules of economic calculation. Successful choices will become best practices across an entire industry, in the same way that all businesses now use supply chain automation and payment processors. Once widely adopted, these innovations provide a similar benefit to most firms, and no longer differentiate one competitor from another.
But replacing labor with a machine does not necessarily mean reducing cost. A decision to replace people with AI is subject to the same rules of economic calculation as any other choice between alternatives. Whether a machine lowers costs or increases revenues depends on what it does and how much it costs. It is not free to deploy software. Like all technology, AI has a price tag.
Businesses will adopt AI when it makes sense and in other cases not. I often have a worse experience talking to a voice recognition system than to a person. It costs a credit card issuer as much as $5 per call to hire a person to provide customer service. This cost would have to be passed along to me in some form. Would I be willing to pay $5 more for a better experience? I may prefer a worse experience rather than a higher cost.
Now we are at a point to ask: Do large language models such as ChatGPT, or, other recent advances in AI, rescue the socialist project from its inability to calculate? The last time around the answer was “No.” Today? Not so much. AI can perform specialized tasks. But AI cannot replace entrepreneurs.
Training an LLM is something like statistical averaging over all of the language samples in the input. This is what enables the LLM to produce a coherent response to a prompt. ChatGPT gives a summary of what the average internet writer thinks about a topic. That serves well enough to be useful for many things. If I want to know how to change a setting on my iPhone, then ChatGPT can tell me that because it is widely known.
As Bronze Age Pervert explains:
I think what gets called AI now is good. It’s not really intelligence think of it as a “normie simulator”; a contentless mimicry of language and application of rules already describes normie mind.
Markets are driven by the differential knowledge, skills, viewpoint of the management and leadership of business firms. The market price formation process is a type of consensus. Through the bidding process we find what the prices are. The bidding process also determines which firms shall have control over specific assets. Each buyer has their own specific use for the asset.
Entrepreneurs are not normies. Entrepreneurs succeed or fail by differentiating themselves from competitors. The buyers who succeed in the bidding process for scarce labor and capital goods are willing to pay a bit more for an asset. The high bidder can see why a particular asset is worth more to his business than to other firms who are not willing to offer as much. Oil and gas billionaire and Dallas Cowboys owner Jerry Jones described this as “overpaying” for a high quality asset. But it can just as often mean finding employment for workers or assets that are selling for a bargain price because they are underappreciated. The entrepreneur sees that a warehouse that has not been leased for six months could be repurposed as a yoga studio.
Entrepreneurs bring together in one person the ability to earn profits by directing production. “Using existing assets to produce goods and services” is not a single thing that anyone does, or could do. We have no data on “planning the entire economy” that could train an AI. Firms plan, and individuals plan, but production involves the interplay of all the plans of privately owned firms and all the executions.
Entrepreneurs accept a risk of loss, if they fail at any stage – calculation, planning, or execution. The market economy ties production to the personal accrual of profits or losses. A human person starts a business or invests in one in order to provide for themself, their family or however they envision their future.
The meaningful use of time in a person’s life requires a continuation of their consciousness from the past into the present. Each business has its own time horizon, as required by the time needed to create the product or service, before profits can be captured. The current generation of AIs has no consciousness that spans over time. They spin up some computing power when asked a question and tear it down when the conversation is complete. They have no continuous being or purpose that ties past, present, and future into a single timeline.
AIs can be trained to do specialized things, when there is a demonstrated body of data collected from people doing that thing. For example, exploration geologists already use AI to identify drill targets that may lead to discovering a mineral deposit. Many other specifics in the management of a firm can be partially or entirely automated, or assisted with AI.
What AI cannot do is incorporate into a single entity all of the specialized skills that the entrepreneur has; the abilities to calculate, plan, and execute, the personal acceptance of profit or loss, and the continued span of consciousness over time which makes the pursuit of wealth purposeful.
Published under a Creative Commons Attribution 4.0 International License
For reprints, please set the canonical link back to the original Brownstone Institute Article and Author.