The emergence of ChatGPT and other generative technologies has given fresh impetus to concerns about AI taking our jobs. A recent paper from the MIT Sloan School of Management explores how valid these new concerns are.
The researchers explore how technology affected income between 1981 and 2016, which, while outside of the realm of generative AI, nonetheless saw the introduction of personal computers, mobile phones, the internet, industrial robotics, and various other technologies that had a significant impact on both blue- and white-collar work.
Leveling the playing field
The key to determining the impact was whether the new technology could do the job instead of the worker or whether it complements the worker. In the former, there was an obvious loss of income as livelihoods became automated. In the latter circumstance, the best workers saw their incomes dip as less experienced workers were able to close the ability gap, and therefore see their incomes rise.
The researchers kicked off their study by collecting job descriptions from two widely used sources, ONET (Occupational Information Network) and the 1991 edition of the Dictionary of Occupational Titles. These sources provide detailed information about various professions and the tasks they involve. For instance, the ONET entry for a “Kindergarten Teacher, Except Special Education” outlines 37 tasks, including activities like “Demonstrate activities to children” and “Read books to entire classes or to small groups.”
They then tasked ChatGPT with categorizing each job’s tasks into two groups: routine (tasks that require minimal experience and are likely easy to automate) and nonroutine (tasks demanding extensive experience and likely challenging to automate). To ensure accuracy, the researchers double-checked ChatGPT’s results against other classification methods.
Technology’s march
The researchers then examined the march of technology during that period by trawling through the patent databases. They were particularly interested in breakthrough patents that were not only very different from what came before but also highly influential on future patents.
This enabled the researchers to explore whether the breakthrough patents were disrupting routine or nonroutine tasks. If the technology matched with a routine task, the researchers considered it to be a labor-saving technology, whereas if it matched with a nonroutine task, it was a labor-augmenting technology.
With this, they were able to gauge whether each of the technologies was likely to have a positive or negative impact on the labor market, as if a job was exposed to a labor-saving technology it was highly likely to have been automated, whereas exposure to labor-augmenting technology will likely have corresponded with a period of growth for that role.
Exposure to change
Finally, the researchers analyzed government data on people in various occupations. This data included things like their income and their education levels, and the researchers were interested in any changes after the breakthrough patents were issued.
In a broad sense, the data showed that within a specific occupation, being exposed to technology that replaces labor tends to lead to lower wages and decreased job opportunities. Conversely, exposure to technology that enhances or supports labor tends to result in higher wages and increased employment for that particular job.
However, when the researchers switched their focus to examining the impact of technology exposure at the individual worker level instead of the broader occupation level, the narrative became more intricate—especially concerning technology that amplifies labor.
Interestingly, however, as wages were increasing across these labor-augmented occupations at the same time that the average worker was also seeing a small decrease in their income and a small increase in their likelihood of losing their job, the researchers conclude that most of the gains from the new technology were going to entry-level people.
What of the future
Of course, a common argument among tech fans is that the past is no guide to the future and that this time is different from what has come before. They argue that generative AI promises to be much more wide-reaching than previous technologies, and therefore will have an even greater impact.
To see whether this might be possible, the researchers ran a fresh analysis, this time using ChatGPT itself, which they asked to explain whether a particular task could be performed with or without significant human intervention.
This information was then compared with occupation descriptions to gauge whether the routine and nonroutine elements of a job could be either augmented or replaced by the latest generation of AI.
Obviously, the methodology demands a pinch of salt, but they nonetheless came to a similar conclusion, that less experienced workers in labor-augmenting roles are likely to see a boost in their earnings as the technology allows them to catch up with the productivity of their more illustrious peers.
The results also indicate a growing significance of soft skills in the workplace. Jobs heavily dependent on interpersonal skills seemed to be minimally influenced by technology. So, while generative AI will almost certainly change the labor market, it’s unlikely to render vast swathes of the workforce redundant.
