Can computers plan a socialist economy? The idea is not new; it first appeared in the debate over economic calculation, which began in 1920 with Ludwig von Mises’s first article on the topic and continued until 1949. This was a time when computers had recently emerged. Computers were not widespread, but their possibilities were evident. Oskar Lange considered the market economy to be a “crude and soon to be obsolete computing machine.” More recently, rapid advances in artificial intelligence (AI) may have tilted the balance of power toward computers.
It was accepted by all sides that, in a complex economy, there is a near infinity of possibilities for both what to produce and what production methods to use. It was also accepted that the aim of production is to produce more of the goods that consumers prefer at the cost of fewer goods of lesser importance. Mises explained:
Hence the economic problem arises: to employ [capital goods] in such a way that only those goods should be produced which are fit to satisfy the most urgent demands of the consumers. No good should remain unproduced on account of the fact that the factors required for its production were used—wasted—for the production of another good for which the demand of the public is less intense.
The possibilities of producing more of the same goods already in stores as well as new products that are not currently available must be considered. The variability among goods along aspects of product quality creates even more choices. Production methods are constrained by the scarcity of existing capital goods and labor. Other variables such as the location of existing plants and workers, the distribution of these goods, and the necessary production time require even more decisions.
Mises’s major insight was the necessity of calculation using money prices to compare alternatives. Decisions must be reduced to a number in order to compare alternatives. Consumers determine prices of consumption goods through their decisions to buy or not to buy. Producers are also concerned with the prices of intermediate factors in earlier stages of production. In the market, these prices are set by competition among entrepreneurs to use the available productive resources.
In the debate, Lange and other “market socialists” suggested that computers could make these decisions by solving a large system of equations. The equations would have to incorporate consumer demand curves for all products as well as labor supply curves. The combinations of required inputs to produce some quantity of outputs would be represented by a production function.
There are multiple problems with this idea. If, as Lange suggested, a computer could solve the equations, then does the solution constitute a “plan” for economic production? The solutions of the equations provide the quantities of inputs, outputs, and production functions. The normal meaning of “plan” is closer to the word “recipe.” A list of quantities and inputs that could produce certain outputs does not constitute a plan.
The solutions of the equations play the role of ingredients in a recipe. What is also needed are the steps for combining them. A plan includes the steps to reach the goal. F.A. Hayek explained that, even if the equations could be formulated and solved, “This 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.”
Economist Joseph Salerno has identified another problem with the equation-solving solution. The equation computations have no relevance to the real world because they state an equilibrium condition at a single point in time. However, the real world is constantly changing. The equilibrium prices would not be stable during the period of production. Every change would create a new set of equilibrium conditions.
A business starts out with a plan for what they intend to sell—an estimate of the required inputs and expected outputs. The entrepreneur has a view of the prices that will be paid for inputs, prices at which outputs can be sold, and how he plans to manage the process. This includes which production methods are appropriate, who to hire, how much to do within the firm, what parts or auxiliary services to obtain on the market, what employment policies to have, how to manage risks, and how to solve problems that come up along the way. The ability to make all these decisions is represented by the production function in the equation.
The prices used in the advance calculation must be “good enough” to consider all the unknowns and allow the firm to make a profit. Business plans—which may be approximate or précis—are based largely on estimated prices. Some prices may be paid up front such as a long-term lease. Because production takes time, many inputs must be purchased multiple times before the products are completed—for example, wages paid each week or month. For each purchase, the prices prevailing at the time are paid. The quantities of inputs needed are also estimates. Some production processes are more predictable than others. The amount of spoilage and accidental loss is not known in advance. The efficiency of a particular process may be greater or less than anticipated. Finally, the selling prices of finished goods depend on customer acceptance of the product, which is not a sure thing.
If one business were to receive a detailed set of instructions from a different firm in the same industry, could the two firms deliver the same product for the same cost and quality? No more than two cooks could realize the same dish from the same recipe or two musicians playing the same score could sound the same. A considerable amount of skill and experience, as well as a multitude of individual decisions, are required. Hayek wrote 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.”
Even something as straightforward as holding the desirable level of inventory for sale requires economic calculation. How long does the inventory last? Inventory for a clothing store could expire due to spoilage, seasonal garments losing value out of season, or fast-moving fashion trends leaving one style behind. What alternative uses exist for scarce resources such as warehouse space? What are the costs of tracking, heating, cooling, and securing the inventory? W.H. Hutt has written about the economics of availability. Some businesses intentionally manage their pricing to ensure that they never sell out. A convenience store would consider it a failure if they ran out of beer. In that type of business, some additional stocks that are not sold are part of the cost of never running out.
As Hayek explains, even the concept of the cost of production involves economic calculation. Yes, there are monetary costs of the inputs that were needed to produce goods. However, every business has durable assets that are used up over time through wear and tear. Assets should be valued at the greater of their resale price or replacement cost. In the mining industry, it is common for a deposit to be purchased before the mine has been fully excavated because another mine developer may have a better ability to mine it than the owner.
Businesses have a range of shared costs such as payroll processing, insurance, utility bills, legal and compliance costs, rent paid for headquarters, and others that bear some relationship to the cost of producing their product. Other downstream value-chain activities such as marketing and advertising are also costs that contribute to revenues. The relationship of these costs to specific revenues is less clear.
Economic calculation, the choice of production methods, and the execution of the plan are not independent steps. Economic calculation does not stop when production begins. Economic calculation informs nearly every decision that is made over the course of a day at work. The motivation for profit and avoidance of loss incentivizes the use of economic calculation at all levels of the firm. Even frontline workers require a general awareness of how much resources cost the business in order to make tradeoffs in their use of the company’s assets. A sous chef in a kitchen must take care not to overcook a prime cut of steak, while a few leaves of lettuce can easily be discarded.
Even if computers can solve equations better and faster compared to eighty years ago, that is of no help in replacing the market economy. The computation of equilibrium solutions for quantities does not give anyone the ability to produce goods and services. Recent advances in AI have developed other computational methods, such as neural networks that can solve problems that cannot be stated in a closed-form equation.
The search for profits and risk of loss drives economic calculation. The entrepreneur strives to earn profits because wealth affects his future. Profits give the entrepreneur the possibility of a better life for himself, his family, and those in the world that have been enriched by his products. To decide to improve your future, you must first have a future. Computers do not have a life. “Having a life” is not an algorithm either. Having a life goes beyond only having goals that you can only reach in the future.
Entrepreneurial skill includes being good enough at economic calculation. Entrepreneurship can be studied and learned to a degree. Looking at the proportion of businesses that fail, it is clear that many people who go into business overestimate their own entrepreneurial skills.
AIs are trained on large datasets to reflect an aggregated view. To the extent that everyone knows certain things within the same industry, this knowledge is not a differentiator. Entrepreneurs undoubtedly look at data. However, each entrepreneur has a unique and differentiated view of how to use existing assets. Many things in a business involve data, but many do not. For a new product, there may be no data. Without data, the entrepreneur must use imagination and empathy to foresee how customers might accept a new product. However, we don’t exactly know what entrepreneurship is—not in such a way that we can train AI.
Businesses have always made calculations. Bookkeeping predates computers. Mechanical cash registers existed before electronic ones. All modern businesses use information technology in some form. Any software, from spreadsheets to enterprise resource planning, that is readily available becomes a type of capital good. Like all capital goods, software can replace some types of labor. A competitive advantage comes from using a piece of software to greater advantage than your competitors do.
Economic calculation is required to quantify this degree of advantage from adopting the new software. Software is costly to build, deploy, and use. A particular decision to use AI may work out badly. The world of information technology is littered with failed software projects, often after incurring enormous costs. According to Forbes, “Big technology projects fail most of the time.” Mary K. Pratt in CIO writes, “Technical and transformational initiatives still fall flat at an alarming rate.”
As AIs become more generally available, they will be adopted where they can help to make decisions, incorporate more information, or save humans time spent on an increasing range of tasks. I have interacted with customer support AIs that could not solve my problem at all or not as well as a person could. However, it might have cost the company from $3 to $6 to have a person help me, and that cost would have been passed on to me in some form. We are constantly faced with choices between better, more costly services and lower quality but cheaper alternatives.
AIs are improving at specific skills humans already have. AIs are trained on worked examples or datasets by the humans who do it. “Planning the economy,” however, is not a specific skill nor is it done by any one person or set of skilled people. What people do instead is to allocate resources within a limited scope of a household or a firm using economic calculation. Computers can help with that, but they cannot become central planners.