This is a term paper I wrote for a computer science class I took at a community college over the summer of 2020. We were free to pick any topic related to computer science, provided that it was approved by the professor. I do admit, halfway through writing the paper, I began getting lazy and drastically reduced the effort I put into my research, interested only in getting an A for the semester. If this degradation in quality bothers you, I apologize, but I nonetheless hope to introduce more people to the ideas explored in this research. Also, since the focus of this paper was not intended to be Soviet history or economics, I did not delve into the broader policy changes that occurred over the years following Stalin’s death, and instead focused on how today’s computer technology can facilitate socialist economic planning.
One of the most oft-repeated interpretations of political developments of the past century is that the Soviet Union’s demise was the inevitable result of its economic system — namely, its socialist planned economy. By the 1980s, the rapid economic growth that took place during the first few five-year plans that transformed an underdeveloped agrarian society into a world-class superpower had long been overshadowed by years of economic slowdown, and the planned economy, which had once been considered by many to be a model for others to follow, began to be considered by many to be an unrealistic and utopian ideal that inevitably results in failure.
While the Soviet Union’s economy did have its shortcomings, there were and still are political incentives to attribute the failures to its guiding ideology without analyzing them in the context of the domestic and international situations faced by the country throughout the course of its existence. Rejecting the dominant narrative and the assumption that all planned economies to be inevitable failures, the purpose of this paper is to briefly review the successes and failures of the Soviet Union’s socialist planned economy and examine how modern-day computer technology, especially machine learning, can solve problems faced in the past by Gosplan’s economic planners, with the understanding that the Soviet economy had grown too complicated by the time Brezhnev took power to continue planning with methods that were largely unchanged from the days of Stalin’s leadership.
II. Brief Introduction to the Soviet Planned Economy
The October Revolution in 1917 resulted in the formation of the Russian Soviet Federated Socialist Republic, which was later joined by other union republics to form the Union of Soviet Socialist Republics (USSR). Following the economic policy of War Communism during the Russian Civil War and the subsequent era of the New Economic Policy, the Soviet Union embarked on the first attempt in world history to manage an economy not by market signals, but through central planning. As a socialist state, the majority of the USSR’s means of production (factories, mines, railroads, etc.) were publicly owned, and its economy was centrally planned in the main at various levels of the government.
The State Planning Committee of the USSR, more commonly known as Gosplan, was the agency responsible for the overall planning of the economy at the All-Union level. Additionally, each of the fifteen Soviet Socialist Republics had its own republican gosplan. The production of certain commodities was directed by the All-Union Gosplan, while republican gosplans oversaw the production of other commodities. Various ministries that oversaw specific industries existed at the All-Union, union-republic, and republic levels to facilitate the transfer of information between the planners and enterprise managers and also allocate resources to enterprises that fell within their supervision (Gregory & Stuart, 2001). Each year, the techpromfinplan (the technical-production-financial plan) was issued to Soviet enterprises that were responsible for fulfilling production goals.
Whereas resources in a market economy are allocated predominately in accordance to market signals, resources in a socialist planned economy are allocated primarily through an economic plan that aims to define production goals that are not only feasible given the economic realities of a country, but also optimal. As an economy grows to be more complicated, this becomes a more difficult task as not only increasing amounts of information are needed, but also the ability to interpret such information in time to draw up five-year plans and annual plans on schedule.
Economic plans in the Soviet Union were created by a process called material balance planning, in which available inputs were balanced with desired outputs. Each spring, state planners would set growth targets consistent with the five-year plan which are then communicated to the ministries. The ministries then communicated these figures to the relevant enterprises. Afterwards, the ministries negotiated with the planners for the inputs necessary in order to attain production targets over the summer. Come autumn, the state planners would balance the input requests with the production goals, and the plan is sent to the Council of Ministers and the Supreme Soviet for approval and ratification (Abegaz, 2018). To facilitate this process, Gosplan had teams of workers each tasked with the responsibility of obtaining information about a particular product group. Evidently, this was a process that involved several layers of bureaucracy each relying on estimations based on imperfect information.
III. Successes of the Planned Economy
Despite its shortcomings, the Soviet Union was nonetheless able to sustain economic growth for decades. The first five-year plan laid the foundation upon which future economic plans were built, and with the exception of the 1932–33 famine and the period during and following the Soviet Union’s participation in the Second World War, the daily Calorie intake per person generally grew following the conclusion of the first five-year plan until 1970, when the figure stabilized at approximately 3400 (Allen, 2003). The Central Intelligence Agency (1983) reported in a declassified document that “American and Soviet citizens eat about the same amount of food each day but the Soviet diet may be more nutritious.”
According to mathematical models created by economist Robert C. Allen (2003) in his book Farm to Factory, a Reinterpretation of the Soviet Industrial Revolution, a capitalist alternative would not have achieved the same levels of growth experienced by the Soviet Union due to the lack of the necessary prerequisites of economic development as an agrarian society. An attempt to follow a capitalistic model of growth, according to Allen, would have resulted in high levels of unemployment and low growth. The Stalin-era policy of directing resources towards heavy industry created the necessary conditions for the improvement in the quality of life of Soviet citizens.
Though the era of Brezhnev’s leadership is known for economic stagnation in comparison to the decades that preceded his ascension to power, economic growth nonetheless continued in the Soviet Economy; whereas real wages fell from 1973 to 1989 in the United States, they rose in the Soviet Union (Cockshott & Cottrell, 2002). Allen argues that the economic slowdown of this era was not an inherent feature of the planned economy, but rather the poor decisions made by Soviet planners in the face of an evolving economy, the arms race, and other factors.
IV. Problems with Planning
When the Soviet Union first began to industrialize, its methods of planning were adequate as the economy was simple in comparison to later times. In an age before modern computer technology, shortcuts had to be taken in order to draft economic plans on time. Central planning was limited to a relatively small number of commodities while the rest were produced at the republican levels. In addition, assumptions were made to make planning feasible — one glaring example was the practice of assuming a linear relationship between inputs and outputs that remained unchanged as production scale grew (Gregory & Stuart, 2001).
Intensifying problems was the fact that individual enterprises were not in communication with one another, and were instead kept informed by ministries and central planner (Cottrell & Cockshott, 1993). Consequently, an improperly informed center would result in issues such as product shortages for consumers and insufficient inputs for producers, who were susceptible of knowingly or unknowingly misrepresenting necessary inputs. If an enterprise were to discover that it lacked sufficient inputs to meet production goals, there were several options it could pursue. An individual enterprise requesting more inputs would result in either another enterprise that manufactures intermediary goods being requested by a ministry to increase production, which necessitates more inputs elsewhere, or the use of existing stock if one were to exist. Alternatively, an enterprise could reduce the quality of its products while still technically meeting annual production goals; for example, if an enterprise were tasked with producing a million cinderblocks but only had the materials necessary for producing 800 thousand, a million cinderblocks of subpar quality could still be produced by reducing the density of the product.
Evidently, one major cause of these predicaments was the lack of information technology that could provide quick and accurate feedback to central planners, who in turn did not have computers capable of processing this information for the purpose of revising production plans, which could not be done in real time due to technological restraints. These issues did not go unrecognized in the Soviet Union, and developments in computer science did spark discussions regarding the role computers could play in helping planners overcome these issues.
V. Computers in Economic Planning
As was the case in the United States, a major driving force of computer technology in the Soviet Union was the military constantly keeping up with the Cold War. With the development of military networks, some computer scientists began envisioning non-military uses for computer technology, and in 1962 Soviet computer experts began developing what was intended to become a nationwide network called the National Automated System for Computation and Information Processing, more commonly known as OGAS. One of the major proposed uses of OGAS was streamlining the process of economic planning, with many noting the potential of increasing the accuracy of economic planning while reducing the time required with large amounts of data available to planners from the central to enterprise levels (Gerovitch, 2008). Owing to government decisions, however, funding for the project stopped in 1970, hence planning continued with the use of existing methods and information channels.
Roughly coinciding with OGAS’ death in 1970 was the birth of another project that attempted to integrate computer technology with economic planning. That year, the idea of incorporating cybernetics in economic planning was raised in Salvador Allende’s social-democratic Chile. Because Chile was an underdeveloped country with only about 50 computers that were not yet interconnected, the project, named Proyecto Synco, was not intended to be anything more than a telex network that enabled various sectors of the economy to share information in a timely fashion for the purpose of aiding economic planners in making decisions (Goertzel, 2014). Proyecto Synco met an abrupt end in 1973, though, when Allende’s government was overthrown by a military coup led by Augusto Pinochet.
Even though computer technology was never able to realize its full potential in socialist or social-democratic economic planning, it has nonetheless become vital in today’s economy, and lessons can be drawn from its application in capitalistic economic planning. Companies like Walmart rely on large-scale and intricate planning, data collection, and data analysis with the help of modern-day computer technology to manage their logistics and perfect their supply chains. Amazon collects customer data to create personalized product recommendations and hence boost sales, while companies like UPS uses computers to determine optimal delivery routes that save fuel and time. Financial institutions rely on artificial intelligence to perform tasks such as risk assessment and fraud detection. Exemplifying the extent real-time data is utilized in today’s economic planning is the fact that Amazon keeps track of time spent by individual employees in the restroom and factors in the data when calculating their paychecks. All of these tasks involve the collection and analyses of quantities of data that cannot feasibly be handled by humans, enabling tasks that previously required humans and often involved guesswork to be undertaken by computers instead with great accuracy.
VI. Machine Learning
Artificial intelligence is a branch of computer science that aims to develop computer technology capable of tasks that traditionally required human intelligence. This is accomplished through the devising of advanced algorithms that attempt to emulate the human thought process. Machine learning, a subset of artificial intelligence, enables the development of these advanced algorithms by computers themselves by learning through experience with varying degrees of human intervention, and is already implemented in a wide range of technology today, from email spam filters to self-driving cars. In the introductory chapter of his book Machine Learning, computer Scientist Tom Mitchell (1997) states: “a computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E” (p. 2).
The three major categories of machine learning are supervised learning, unsupervised learning, and reinforcement learning. Additionally, there is a type of machine learning that falls between supervised and unsupervised learning, appropriately named semi-supervised learning. Supervised learning is the most ubiquitous subcategory of machine learning. Through analyzing labeled pairs of input and correct output parameters, known as training data, functions are inferred by methods such as regression and classification, which can in turn be used to predict outputs for new inputs (Das, Dey, Pal, and Roy, 2015). Regression is typically used in situations involving quantitative outputs, such as time or price, while classification is used in when the outputs are qualitative, such as email type (priority, normal, spam, etc.) or color.
Whereas supervised learning involves providing the learning model with both input and correct output, unsupervised learning models are only fed input with the aim of finding unknown patterns among the data. Because there is no known expected output that can be taught to unsupervised learning models, they do not perform classification or regression analyses. Instead, unsupervised learning models organize data into clusters by identifying similarities, and can identify anomalies through the process. Unsupervised learning algorithms allow social media websites to identify potential friends of users, financial institutions to detect fraudulent activities, and companies to identify market segments (Das, Dey, Pal, and Roy, 2015). YouTube keeps track of and study viewer preferences and uses unsupervised learning to recommend videos based on user behavior while also considering factors such as device type, recommending shorter videos on mobile devices and longer videos on televisions (Newton, 2017).
The third major category of machine learning is fittingly called reinforcement learning. Reinforcement learning models are not provided with training data. Instead, they are rewarded and penalized for desirable and undesirable outputs, respectively, as they operate. By experimenting with different actions, reinforcement learning models improve their performance by performing actions correlated with rewards and avoiding those correlated with penalties. AlphaGo, the first computer program to defeat a world champion in the game of Go, a strategy game with possible board configurations, utilized reinforcement learning to master the game by first playing games with amateurs to learn how humans play the game then playing against various versions of itself to learn from its successes and failures (DeepMind Technologies, n.d.). DeepMind, the Google subsidiary that developed AlphaGo, utilized reinforcement learning when it developed AI that enabled Google to cut cooling costs for their data centers by 40% (Evans & Gao, 2016).
These learning models are not mutually exclusive to one another, evidenced by the existence of semi-supervised learning, whose learning algorithms are trained with a small proportion of labeled data and a larger proportion of unlabeled data. This is useful in the real world due simply to the fact that most data that exists is unlabeled. Through identifying the suitability of each model for specific tasks and creatively integrating them into computer programs, solutions to complex problems can be developed as more information is gathered and more patterns are recognized.
VII. Potential Applications of Machine Learning in Central Economic Planning
In the past, economists ideologically opposed to socialist forms of economy deemed central economic planning infeasible due to the millions of equations that needed to be solved, and as mentioned previously, Soviet planners did make simplifying assumptions when calculating material balances so that plans could be drafted in time. The capabilities of today’s supercomputers challenge this assertion, with the U.S. Department of Energy’s Summit supercomputer capable of carrying out 200,000 trillion calculations per second (McCorkle, 2018). Furthermore, the power of machine learning has aided humans in quickly making accurate predictions from existing knowledge. Drawing inspiration from machine learning’s invaluable role in the planning processes of capitalistic corporations, we now explore how issues faced by Soviet planners of the last century can be alleviated with the help of machine learning technologies introduced in this paper.
A successful economic plan is one that is capable of allocating resources in accordance to pre-defined goals. In the case of companies like Walmart and Amazon, the goal is to maximize profits by reducing cost inefficiencies and increasing sales. Correspondingly, the goal of a socialist planned economy is to ensure a certain standard of living with sufficient access to goods and services with attention paid to the particular society’s values, which may include principles such as environmental protection and guarantees regarding things such as housing, education, and healthcare. Consumer goods, housing, medical services, and education, etc. all require human and nonhuman resources to produce or provide. For target outputs to be attained, available inputs need to be accurately predicted such that targets are feasible, and for target outputs to be socially beneficial, there needs to be an accurate estimation of demand to minimize the occurrence and magnitude of shortages and overproduction, as well as limit the impact of the two on the broader economy. Today, unsupervised learning algorithms have successfully carried out market segmentation and identified consumers’ tastes. In a socialist society, this information would help planners more accurately infer the demand for different commodities and determine appropriate price and supply levels.
Many of the problems faced by Soviet planners involved incomplete information and limitations on predictions that could be made from the data, and are precisely the types of problem machine learning is designed to overcome. While past planners made simplifying assumptions for practicality purposes, such as assuming linear relations between inputs and outputs, a supposition that knowingly ignores economies of scale, planners today can train supervised machine learning algorithms with past data and use unsupervised learning algorithms to recognize previously unknown trends and patterns. Issues stemming from the inefficient flow of information in the past can easily be corrected with today’s internet infrastructure and computer technology.
With large quantities of historical data, and with statistics being constantly collected and shared in real time, enterprise managers will have a harder time overstating the necessary inputs or understating the productivity of their enterprise. Planners can accurately evaluate the productive capacity of each individual enterprise, the feasibility of goals, and the inputs necessary for fulfilling the goals. Additionally, anomalies can be detected and addressed in their nascency by ministries and Gosplan, rather than at the end of the year, as was often the case in the Soviet Union. Should an issue that prevents production goals from being reached occur, whether it be at the enterprise, ministerial, or societal level, numerous potential courses of action can be assessed with their possible effects predicted with reliable precision, reducing the instances of managers technically meeting production goals by cutting corners. Whereas Soviet planners had trouble reliably adjusting economic plans in response to unforeseen circumstances, planners with access to machine learning algorithms and today’s information technology are better prepared to adapt plans to the ever-changing and imperfect world in which they operate.
Machine learning algorithms are already being used to preventatively minimize equipment failure by catching defects early on before problems worsen, allowing major repairs that cause intrusive downtime to be avoided. This is especially vital for enterprises that manufacture intermediate goods in a planned economy, as underproduction in these enterprises lead to shortages of inputs in other enterprises, which are ultimately passed to society. Similarly, machine learning algorithms are used today to aid manufacturers in quality control, helping them perfect their manufacturing process and waste less inputs. While the impact of these factors such as machine breakdowns and wasted product was estimated in Soviet economic plans, they can now be systematically analyzed and factored into plans with great precision while manufacturing methods are perfected to reduce such occurrences.
The optimal plan for a capitalistic corporation is one that maximizes its profits. Similarly, the optimal plan for centrally planned economy is one that helps a society reach its stated goals. If something is not viewed as a goal, then it will not be addressed. For example, under both capitalism and socialism, the reduction of greenhouse gases does not occur as a result of human intervention if it is not considered by shareholders or central planners, respectively, to be a goal. If this is seen as a societal goal in a capitalistic society, then corporations must be incentivized one way or another to modify their production processes. In addition to resolving issues stemming from inaccurate estimates of supply and demand as well as unforeseen externalities, central economic plans in a socialist society can be adjusted with goals such as reducing greenhouse gas emissions or water pollution. Once set, the likelihood of such goals being fulfilled increases as correlations between the specific steps of production of specific commodities are discovered in collected data. Additionally, since unsupervised learning can identify previously unknown patterns, it is plausible that trends seemingly unrelated to economic activities, such as cancer rates in a specific geographic area, could be linked to human activities, allowing authorities to take countermeasures in order to maximize the well-being of society.
This paper has summarized the Soviet planning process and issues it faced. It has also introduced the major categories of machine learning, provided existing examples of their uses, and identified ways machine learning can be used to overcome the inefficiencies that hindered the Soviet planned economy. The purpose of this paper is not to present computers, artificial intelligence, and machine learning as economic panaceas, but consider how these technologies can reconcile some of the issues experienced by Soviet planners of the last century. The availability of these technologies alone cannot guarantee the success of an economy, whether it be capitalist or socialist, especially if these technologies are not accompanied by policies that favor the maximization of their potential, as can be seen by the Soviet Union’s reluctance to adopt cybernetics for its economic planning. However, the capabilities of today’s technology do challenge claims of unfeasibility made by critics of socialist planned economies, and I believe that this is a topic worth further investigating as the efficient use and allocation of resources becomes an increasingly important consideration in light of issues such as climate change and the cyclical nature of recession.
Abegaz, B. (2018). Pedagogical notes — the centrally planned Economy: pre- and post-socialist. Williamsburg, VA: College of William and Mary Department of Economics.
Allen, R. C. (2003). Farm to factory: A reinterpretation of the Soviet industrial revolution. Princeton, NJ: Princeton Univ. Press.
Central Intelligence Agency. (1983). [Declassified document approved for release in 2007, FOIA number: CIA-RDP84B00274R000300150009–5]. Retrieved from https://www.cia.gov/library/readingroom/docs/CIA-RDP84B00274R000300150009-5.pdf
Cockshott, P., & Cottrell, A. (2002). The Relation between Economic and Political Instances in the Communist Mode of Production. Science and Society, 66(1), 50–64.
Cottrell, A., & Cockshott, W. P. (1993). Socialist Planning after the Collapse of the Soviet Union. In P. Bridel (Ed.), The socialist calculation debate after the upheavals in Eastern Europe: Papers given at a conference held at the Centre d’etudes interdisciplinaires Walras-Pareto, University of Lausanne (pp. 167–185). Revue Europeenne des Sciences Sociales, vol. 31, no. 96. Cahiers Vilfredo Pareto.
Das, S., Dey, A., Pal, A., & Roy, N. (2015). Applications of Artificial Intelligence in Machine Learning: Review and Prospect. International Journal of Computer Applications, 115(9), 31–41. Retrieved July 06, 2020.
DeepMind Technologies. (2016). AlphaGo: The story so far. DeepMind. Retrieved August 01, 2020, from https://deepmind.com/research/case-studies/alphago-the-story-so-far
Evans, R., & Gao, J. (2016). DeepMind AI Reduces Google Data Centre Cooling Bill by 40%. DeepMind. Retrieved July 31, 1, from https://deepmind.com/blog/article/deepmind-ai-reduces-google-data-centre-cooling-bill-40
Gerovitch, S. (2008). InterNyet: why the Soviet Union did not build a nationwide computer network. History and Technology, 24, 335–350.
Goertzel, T. (2014). The path to more general artificial intelligence. Journal of Experimental & Theoretical Artificial Intelligence, 26(3), 343–354.
Gregory, P. R., & Stuart, R. C. (2001). Russian and Soviet economic performance and structure. Boston, MA: Addison-Wesley.
McCorkle, M. L. (2018). ORNL Launches Summit Supercomputer. Oak Ridge National Laboratory. https://www.ornl.gov/news/ornl-launches-summit-supercomputer
Newton, C. (2017). How YouTube perfected the feed. The Verge. https://www.theverge.com/2017/8/30/16222850/youtube-google-brain-algorithm-video-recommendation-personalized-feed