AI, the Workforce, and the Economy
David Atkinson
AI Legal Counsel | A.I. Ethics and Law | University Lecturer | Veteran
Part 1 of our bite-sized series on GenAI’s potential impact on the workforce.
We are not going to cite estimations of potential job losses caused by AI because (i) they vary widely, (ii) there is no accountability for the organizations that publish them (i.e., they could be off by an order of magnitude in either direction, and it won’t matter to them), and (iii) these organizations have not proven accurate in the past. They instead seem to be a little more helpful than reading some random day trader’s top stock picks for the upcoming year.?
Let’s look at some relevant factors.
Factor 1: Transition Time
For the workforce, we can assume AI will improve over time. The rate at which it improves is more important than many assume. AI proponents often point out that people have always adapted to new technology, which is accurate, and they tend to lean on the analogy of cars replacing horse carriages. However, what they tend to neglect is that, while the car was invented in the 1880s, there weren’t more cars than carriages on the roads of New York City until some 30 years later. It was another decade after that before carriages were virtually obsolete there. So, carriage makers had forty years to adapt.?
A similar story unfolds for telephone operators replacing telegrams and automated switchboards replacing switchboard operators. The adaptation time frame only grows as we reach further back in history: the horse-drawn plow replacing human labor, steam engines replacing the horse-drawn plow, internal combustion engine tractors replacing steam engines, and so on. In 1962, 90% of Americans were farmers; today, only about 2% are, and society hasn’t collapsed.1 But for that transformation individuals and the economy had time to react and adapt.
Factor 2: Occupational Transformation
The evolution of Automated Teller Machines (ATMs) in the 1970s and their impact on the banking industry offers valuable insights into how technology can transform job roles rather than eliminate them. The introduction of ATMs, while initially perceived as a threat to bank tellers, led to an unexpected increase in teller jobs. This was due to ATMs reducing the operational costs of branches, allowing banks to open more branches and thus increasing the demand for tellers. This example demonstrates how labor-saving technology can sometimes lead to job creation, a phenomenon that could be relevant to other industries facing automation.
The transformation of the bank teller’s role from routine cash handling to more complex tasks such as customer service and account management also highlights how technology can free up human workers to focus on tasks that require human skills like empathy and interpersonal communication. This shift suggests that the integration of technology in the workplace doesn’t necessarily mean eliminating human jobs but rather a change in job descriptions and requirements.
However, the rise of digital banking and fintech presents a new challenge to the banking industry. The demand for physical bank branches and tellers may decrease with the increased use of online banking, mobile banking, and digital payments. The COVID-19 crisis accelerated this shift toward digital banking, indicating that industries need to adapt quickly to changes brought about by external factors. The predicted 12% decrease in teller jobs by 2028 serves as a reminder that while technology can create jobs, it can also make certain roles obsolete.2
Yet even bank tellers have had 50 years to adapt.
Factor 3: AI Capabilities
In contrast, the first widely available LLM chatbot, ChatGPT, was only released in November 2022 and could only handle text as inputs and outputs. Now, it can converse with audio like a smart speaker (but using much more natural intonations than Siri), generate images that are often better than most humans who try to do art (and much, much quicker), and identify images with superhuman ability. These advances are so rapid that they even took the leading AI researchers by surprise.?
AI may eliminate many dull and dangerous jobs, but there is no principled reason it would stop there. AI is just as likely to replace fulfilling and meaningful jobs. It’s impossible to know which jobs will be replaced and when, but assuming some replacements will happen soon seems reasonable. It also seems AI will likely replace non-manual labor before replacing plumbers, electricians, janitors, and other hands-on labor heroes because robotics is not improving at the same rate as software (code running on servers). Instead, it seems jobs that require less skill, are repetitive and are more like humans acting as algorithms will be the first to be replaced (e.g., customer support for commodified goods).
But what do we mean by job replacement? Another line people tend to mention is that while AI can do some tasks better than humans, virtually all jobs require multiple distinct skills (usually dozens). AI isn’t close to performing every task.?
Let’s set aside that AI’s inability to perform tasks today doesn’t mean AI won’t figure out (with billions of dollars and lots of human help) how to perform the task tomorrow or in some reasonably near future. In fact, for the sake of argument, let’s assume there are some tasks AI will never be better at than humans.?
AI may not be able to replace entire jobs, but it also doesn't need to in order to have a tremendous impact on the workforce. For example, it need not replace every task a radiologist can do to replace radiologists. If it can replace a significant chunk of the time-consuming work (like analyzing images, annotating, writing notes about the images, making preliminary diagnoses and treatment plans, and other administrative tasks), and the remaining human-necessary work remains (talking to and comforting patients, reviewing and approving or modifying diagnosis and treatment plans), the division of labor means you still need fewer radiologists overall.?
A proper way to look at this concept is that AI is not meant to replace jobs, but rather AI could be implemented into the workplace to enhance jobs. Why might this be the case? It's the same reason why trade is more efficient than a nation, company, or person handling all tasks themselves. Sure, I could learn how to build my own house and everything in it (all appliances, furniture, and plumbing), but that's not the best use of my time. Instead, I'm better off paying people who specialize in those fields to do those tasks/create those things while I work in an area in which I'm better suited. If we were all entirely self-reliant, we’d all be much worse off and far less productive.?
领英推荐
If a hospital can save its radiologists 50% of their time by having AI perform the tasks it’s better at, then the hospital can have the radiologists focus on the tasks they are better at, and one radiologist could accomplish the work of two radiologists. Meaning, AI replaced a job without having to know how to perform all the tasks associated with that job competently.?
In this way, one could agree that, yes, AI may never be capable of performing all the tasks of [insert job title here], but it could still replace [insert same job title here].?
Factor 4: Rate of Change
The rate of the change could compound issues. If someone spends 11-13 years becoming a board-certified radiologist (from the time they start college), it’s not so easy to shift to another specialty. Imagine what it would be like to be 8 years into that journey and learn that your job prospects have shrunk by a significant margin.?
Now imagine that happening for all kinds of professions simultaneously (meaning, over the span of, say, five years). It would scramble everything we believe. What should they do to gain a different job? Go back to college? Who will pay for that? What if, by the time they graduate, the new profession they were training for now also has dim prospects as well? Do they go back to school again?
It seems probable that we may enter an era in which significant chunks of society will have to complete one—or two-year degree programs every five years or so just to remain relevant and keep pace with AI developments and capabilities to limit job loss risk.?
A key question becomes: Can AI create a job that AI cannot eventually replace in a sub-career-length timeframe? Perhaps it can. But why should we believe that's the case??
This isn’t entirely theoretical. The Washington Post noted that GPT-4 was trained on Stack Overflow, a forum for programmers to ask and answer coding questions, which meant GPT-4 could now answer those same questions. Traffic to the site slid by 15%, and Stack Overflow laid off nearly a third of its staff as a result.3?
In a more forward-looking example of the impact of AI on corporate thinking, IBM reportedly paused hiring for nearly 8,000 positions that it believes AI will replace in the coming years.4 They are saving the costs of salaries, benefits, equipment, real estate, and job training on the belief that AI will allow them to accomplish at least as much as new hires at far lower costs.
As the leader of Israel noted, “This is like having nuclear technology in the Stone Age. The pace of development [is] outpacing what solutions we need to put in place to maximize the benefits and limit the risks.”5
This is generally a good thing, by the way. We get more food per acre than ever before. And people aren’t forced into being farmers just because they were born into a family of farmers. Farming in the U.S. | American Experience | Official Site | PBS
The following students from the University of Texas at Austin contributed to the editing and writing of the content of LEAI: Carter E. Moxley, Brian Villamar, Ananya Venkataramaiah, Parth Mehta, Lou Kahn, Vishal Rachpaudi, Chibudom Okereke, Isaac Lerma, Colton Clements, Catalina Mollai, Thaddeus Kvietok, Maria Carmona, Mikayla Francisco, Aaliyah Mcfarlin