AI as Capital Deepening: Substitution and Complementarity in Labour Markets

Will AI come to replace workers, or give workers the tools to boost their productivity?  This debate is a hot topic among economists, business leaders, and labor organizations.  The rapid emergence and spread of generative AI is seeing software approach the competency of many white collar workers, at least those at the entry level.  Theorists are trying to determine whether AI will mostly turn out to be a complement to labor or a substitute for labor.

AI as Substitute for Labor

Unfortunately for entry-level workers, many routine tasks, such as managing an office calendar or handling written correspondence with customers and clients, can likely be fully automated by AI within the next several years.  Already, many customer service tasks have been automated online, with customers having to work hard to get through to a human employee.  Ultimately, AI will not replace all customer service workers, but only a handful will be necessary to handle the few tasks AI cannot.  

Similarly, AI will be able to generate documents and correspondence for a large firm with only one or two human workers overseeing the rapid product creation.  Instead of a team of white collar workers generating documents, one or two can put in the appropriate queries and data sets and check the massive files of results.  Thus, human labor may be largely replaced for routine cognitive tasks, which require little critical thinking or creativity.  

The Output Effect May Save Some Jobs

The above scenario is often seen as an “AI apocalypse” for entry-level white collar jobs, with a large percent of office workers replaced by high-speed software.  This apocalyptic scenario also extends AI dominance to transportation, with self-driving cars and transportation drones replacing drivers and pilots.  And, most recently, fast food may become fully automated thanks to AI, with at least simple products created entirely by machine after ordered via smartphone app.  So, what will become of our entry-level white collar workers, drivers, pilots, and fast food employees?

Historically, most displaced workers find other jobs, often new jobs created by the increase in output caused by the same automation that created the disruption.  This is known as the Output Effect: increased output creates a demand for new jobs.  And, we are more than three years into the AI age…and there has been no statistical increase in unemployment.  Therefore, it is more likely that labor is being relocated rather than eliminated.  For example, if AI is able to fully automate all scheduling for an office, those secretarial positions could be moved to other departments to assist with sales, product design, or even physical tasks.

AI as Complement to Labor

Currently, many workers use various AI platforms to assist with daily research and correspondence.  It is common for many white collar workers to use ChatGPT, for instance, to improve their emails, which can aid in clearer communications.  For sensitive or difficult subjects, the use of AI may significantly reduce the time spent crafting emails.  To summarize large tracts of data, AI can reduce time tremendously and create charts and graphs that can copy-and-pasted into emails to customers, clients, coworkers, and supervisors.

Time spent researching can be reduced substantially, as AI can pore through thousands of online posts, documents, and research papers much faster than individual human searches via search engine.  However, the human employee, such as a lawyer, manager, or administrator, is still needed to apply the historical data and trends to the current, perhaps unique, situation.  Most observers would be uncomfortable with AI having the final say, so there is strong demand to keep critical-thinking supervisors, especially those in expert roles, at the helm.

Labor-Leisure Balance Shift

Instead of causing mass unemployment by replacing human workers, AI may be utilized to keep output similar and give human workers more leisure time.  For example, a white collar worker may be able to complete 50 hours worth of pre-AI labor in 30 hours after becoming proficient with Chat GPT, Google Gemini, Grok, or Claude.  Instead of simply demanding more output, the firm may allow the employee to take these hours as leisure and keep his or her pay the same.  

Why would a firm provide such generosity?  It could be seen as an investment in attracting and retaining talent, especially among younger, tech-savvy workers who are more likely to have young children and, therefore, more need for flexible schedules.  Generous leisure or vacation time policies may deprive a firm of some output in the short run, but result in lower turnover costs in the long run as employees become highly loyal.  The most talented workers are far more likely to move to firms that use AI as a complement and offer more leisure than firms that are known to use AI automation as a substitute.  

Long-Run Growth Implications of AI Adoption

Realistically, there will likely be a mix of AI adoption strategies, with some firms using it to automate and replace labor and others using it to complement skilled labor.  This can happen even within the same firm, depending on the type of labor.  Low-skill white collar labor, which often handles routine tasks like scheduling and basic correspondence, will likely be reduced sharply.  High-skill labor, which often requires nuance and expertise, will likely remain largely intact.

Increased Economic Growth Likely

AI’s ability to perform human tasks faster than a human will inevitably lead to greater efficiency for many tasks.  This will result in stronger economic growth than otherwise.  However, it remains to be seen whether a majority of firms will pursue AI-to-maximize-output or AI-to-improve-worker-satisfaction policies.  Both options have advantages and disadvantages and depend largely on societal utility (satisfaction).

Limits on AI Expansion due to Non-Uniformity of Industries

Many also argue that AI adoption will be slower, as not all sectors of the economy operate uniformly.  This will put a limiter on the demand for AI adoption, as improving efficiency and output in some sectors will be less necessary if there are not enough firms or industries “downstream” in the economy to handle the increased output effectively.  For example, a highly automated company that makes construction tools and materials will have less demand to AI-optimize if a stagnant construction industry does not demand more tools and materials.

Warning: A Glut of Output Leads to Falling Prices

AI-to-maximize-output may be avoided by firms and industries as they come to fear a rapid increase in output causing reduced prices.  Historically, this was seen in both the 1880s and 1920s as mechanization and automation drove down prices in agriculture and other industries, resulting in firms struggling to pay off debt.  Ironically, the increase in efficiency harmed producers as they made less revenue with each good sold, forcing them to operate at a breakneck pace simply to break even.  Consumers won, but some producers were driven out of business, especially when the Great Depression hit.