An Inquiry into Labor in the Age of AI: An Economic Perspective

Harshavardhan Srijay
9 min readDec 2, 2020
Photo by Birmingham Museums Trust on Unsplash

Since the Industrial Revolution, a period characterized by the shift of society’s economic structure towards that of industrial capitalism, new automated machines have continuously been introduced to industries and have tested the ability of labor markets to withstand fundamental shifts in the roles of workers. While past advancements in automation are not new, none seem to be as unpredictable as the most recent iteration of automation — artificial intelligence (AI). This technology is gradually being introduced into the workforce, and has the potential to disrupt or even replace a significant percentage of the workforce, due to its ability to combine the computational powers of a computer with the learning abilities that (though crudely) can approximate the human brain. The uncertainty about how labor roles will change in response to this new technology motivates the re-examination primarily of Adam Smith’s ideas about the division of labor and specialization, with regards to their applicability to AI. Here, we will analyze the impact of AI specifically on labor in the healthcare industry. To do this, we must first assess the degree of AI assimilation within healthcare, from which we can then better understand how this impacts the traditional concepts of division of labor and Marx’s labor theory of value. While impossible to accurately predict the future of labor, changes that are already occurring and likely will occur in the healthcare industry will hopefully provide an informative proxy for the inevitable changes in labor that will occur due to AI in society as a whole.

Inherent to healthcare is the abundance of data, giving rise to the integration of AI into the normality of basic healthcare practice, as it can infer patterns and relationships from large volumes of data much better than humans can. Considering the technological strengths and limitations of AI, researchers see 3 major avenues through which it can be used in the industry to maximize productivity and efficiency: task automation, pattern recognition, and contextual reasoning, in roles including but not limited to automated patient registration and discharge, diagnosis from medical images, and automated ordering/restocking of healthcare supplies using pattern recognition of demand, respectively (Spatharou, Hieronimus, and Jenkins 2020). By using AI to fill these roles, rather than humans, employers can reduce costs related to human-error, improve patient experience and outcome, and increase profits through reduced labor costs. For example, many radiologists spend hours of time tediously going through thousands of medical images and diagnosing patients using these images. From an economic perspective, the time invested to perform this work should be “skilled labor” as Marx defines it, as each unit of a skilled radiologist’s labor results in higher quantities of the labor product (correct patient diagnosis) than any un-trained person (Marx 1887, 32). However, relative to AI, the same labor power expended by the trained radiologist is in fact quite “simple,” as each unit of “AI labor” can generate more value. In one study for example, researchers built a deep learning model that trained on breast cancer images and found its diagnostic ability was approximately 12% stronger than that of trained human radiologists (McKinney et al. 2020). While the technology is still likely a few years away from entering mainstream practice, this example offers reasonable optimism for AI’s ability to consistently become a standard part of healthcare in general, and highlights the need for a more robust definition of labor value/specialization in the age of AI. Ultimately, both the cost and labor benefits afforded by AI make it too attractive for capitalist executives not to continue to implement it in practice, meaning AI in healthcare is most likely to only grow in presence.

Since AI is going to be part of healthcare in some capacity, the question now arises about the upper extent of this assimilation. Will it be at the expense of current doctors/workers, completely relegating human labor to a mere redundancy? Many experts agree that almost all repetitive tasks in the healthcare sector such as billing, claims, patient enrollment, scheduling, etc. will be replaced by AI at some point. However, the extent to which AI can do so may not be limitless, as McKinsey & Co. estimates that only 35% of occupations and time spent in the healthcare sector is automatable (Spatharou, Hieronimus, and Jenkins 2020). Let us consider the pharmacist profession, and the extreme scenario in which all pharmacists are fully replaced by automated, intelligent robots, meaning approximately 300,000+ people in the US will become unemployed (U.S. Bureau of Labor Statistics 2020). In this society, while the production and distribution of pharmaceutical commodities (prescriptions, medicine, etc.) is fully optimized for maximal utility using automation, which acts almost as the hypothetical limit of man’s labor power, these commodities will only be accessible to those who can afford them (as in all economies). However, due to the necessarily reduced spending power of many Americans, industry revenue will decrease, and the increased quantities of product will be useless. This hypothetical in pharmacy can easily be extrapolated to the healthcare sector in its entirety, and the same inherent limit in productive utility will exist. While extreme, this hypothetical is instructive in illustrating the limit to labor replacement, as the market is fundamentally dependent on humans, not robots, for consumption. For this reason, AI will most likely not replace all jobs, but reach an equilibrium in terms of job replacement that both minimizes labor displacement and maximizes production/efficiency. Additionally, in terms of labor availability itself, the World Health Organization (WHO) estimates that overall demand for healthcare workers will rise to 18.2 million across Europe by 2030. Even after correcting for 10% of nursing occupations that are expected to be replaced by automation, a 39% increase in demand for nursing occupations is expected by 2030 (Spatharou, Hieronimus, and Jenkins 2020). Thus, despite concern, human labor opportunities will still exist, even in a future world “ruled” by AI.

Now that we see both the potential and limits for the role of AI in the future of healthcare, how exactly will AI integrate with healthcare? While we considered and effectively ruled out the two extreme possibilities of AI inculcation, the intermediate degree to which AI replaces jobs and changes the distribution of labor roles is unclear. Many experts agree that while job replacement by AI is inevitable and must be expected, automation will also create many more types of jobs, providing alternate avenues for the division of labor for humans (Spatharou, Hieronimus, and Jenkins 2020). These increased opportunities will most likely be in the form of more quantitative roles in knowledge creation like data architecture, bioinformatics, and others in data science and engineering. Professionals in these roles will likely need to have significant expertise in quantitative fields because they will be tasked with building AI-powered technology and implementing data-driven approaches to healthcare in its entirety to improve patient outcomes. Because of this, it may be more appropriate to interpret AI’s role in the future of healthcare as an agent that changes the labor itself, rather than simply expressing the degree of AI involvement in terms of a linear gradient. In this case, there is still an abundance of opportunities for humans to pursue specialization and the division of labor, through these newly created jobs in high demand. The nature of labor is not a zero-sum game; there is not a finite amount of jobs. Every instance of automation in the past was met with initial skepticism and fear over the future of labor, but ultimately these fears were not realized, primarily due to the role of automation in changing and creating new jobs, not replacing jobs. 20 years ago, who would have predicted that making YouTube videos in 2020 could generate millions of dollars of income? Throughout history, what society deems valuable in terms of the products of labor has constantly been in flux, and AI is no exception. For example, when intelligent robots inevitably become capable of autonomously performing neurosurgery, the market demand and compensation received by other specialists like robotics engineers, data scientists, and neuroscientists to train and build the surgical machines will rise. While the demand for neurosurgeons themselves will likely decrease, the inculcation of AI opens many more doors for not only others to specialize into, but for the neurosurgeon themselves to oversee the robot. This is analogous to an airline pilot flying on autopilot, where the machine acts as the manual laborer of sorts, reducing human-error and optimizing outcomes, while the human oversees the operation and maintains responsibility. So, it seems that a future of AI and humans working in tandem is most probable, where both agents fulfill equally crucial roles in the production of tradeable commodities.

In this re-defined division of labor, a modification of Smith’s theory, both human and machine specialize in labor whose combination maximizes utility. Smith theorized that in order to maximize productivity of the individual (and as a consequence, society as well), individuals “specialize” in the production of specific goods/services, such that they become increasingly skilled in the production of this commodity. So, traders can generate more potential exchange value by specializing (because of increased quantity of their commodity), than by creating all necessary commodities for themselves. In this way, people can acquire more assets and pursue wealth accumulation, the fundamental drivers of capitalistic progress (Smith 1776, chap. 2). This idea is premised upon the assumption of “learning” a skill/trade and thereby increasing productivity — something that AI is capable of in healthcare settings, and is often superior to humans in efficiency, making it an effective “laborer” in this manner. The combination of the newly available free time for humans generated from this labor of AI, and the new job opportunities that AI manifests allows for increased specialization for humans in an altered manner. This specialization of humans would not come in the form of physical labor, as Smith seemed to have intended, but in intellectual labor — human ingenuity to frame, build, monitor, and train artificial intelligence technologies that are limited in their dependence on trainable data. This would maintain Smith’s theory to an extent — specialization in humans would occur in the process of knowledge creation and creativity, rather than physical labor. Machines on the other hand occupy the opposite side of the production chain, by “learning” and increasing productivity in the manual production of goods/services, just as Smith described. They execute the tasks whose output when attempted by humans is limited in productivity by the time spent. Thus, this modified division of labor maximizes the productive output of human and machine in tandem.

The gradual assimilation of AI in the healthcare sector is inevitable, as are the changes in labor it brings along. Human labor will likely be allocated to knowledge creation and creativity, a more efficient use of human time that also allows for the maintenance of specialization. While the ability of this analysis to extend beyond healthcare is unclear, one interesting possibility that could arise from this is the potential for a complete transformation in labor itself, with demand shifting towards that of task-based compensation for workers. In this modified division of labor, in which the majority of the traditional labor necessary in the production of products will have similar efficiencies since they primarily use the common technology of AI, the differentiated exchange values of commodities could be re-defined through the value of the specialized creative and intellectual power used in their production, not necessarily the value of the physical labor power as Marx describes (Marx 1887, 29). And due to the looser correlation between time spent and creative output, relative to the relationship between time and manual labor, the future of labor compensation for humans might be task-based, rather than time-based, as the majority of labor whose value is defined by time will likely be performed by artificial intelligence.

References

Marx, Karl. 1887. Das Kapital: A Critique of Political Economy. Translated by Samuel Moore

and Edward Aveling. Moscow: Progress Publishers.

McKinney, Scott Mayer et al. 2020. “International Evaluation of AI System for Breast Cancer

Screening.” Nature, 577, 89–94. https://doi.org/10.1038/s41586-019-1799-6.

Smith, Adam. 1776. “Wealth of Nations.” Accessed July 19, 2020.

http://geolib.com/smith.adam/woncont.html.

Spatharou, Angela, Solveigh Hieronimus, and Jonathan Jenkins. 2020. “Transforming Healthcare

with AI: The Impact on the Workforce and Organizations.” Accessed July 15, 2020. https://www.mckinsey.com/industries/healthcare-systems-and-services/our-insights/transforming-healthcare-with-ai.

U.S. Bureau of Labor Statistics. 2020. “Pharmacists: Occupational Outlook Handbook.” Last

modified April 10, 2020. https://www.bls.gov/ooh/healthcare/pharmacists.htm.

--

--

Harshavardhan Srijay

An undergraduate student interested in the science of optimizing decision-making to promote social good.