Skip to content

AI might not take jobs but it will take tasks

Anthony Butler
3 min read

Artificial intelligence (AI) and specifically Large Language Models (LLMs) are what is defined in the economics literature, as a General Purpose Technology.  These are specific classes of technology that demonstrate improvements over time, become pervasive throughout the economy, and typically lead to the spawning of complementary innovations and positive spillover effects.  Electricity or the internet are sometimes cited as examples.  As such, the impact is likely to be broad and there is much work to be done in quantifying and qualifying the economic impacts of these technologies.

It is, therefore, not surprising that this recently published study found that an estimated 80% of workers in the US economy will have at least one task exposed to LLMs with some 19% of workers in job occupations where over half of the tasks are likely to be automated by LLMs.  However, what is surprising is that, contrary to popular wisdom, the jobs that are most at risk of being impacted by these AI technologies are not the “blue collar” jobs but rather the jobs that involve writing and creativity.  We can see, for example, the impact of generative AI on the design and artistic professions where it is now possible to quickly create large volumes of photorealistic and very creative images using models such as Stable Diffusion or Midjourney.  As these generative AI models become increasingly multi-modal, we can only imagine what this will mean for the broader creative professions.

The research also suggests that those jobs with a high dependence on scientific and critical thinking skills are currently not highly exposed to impacts from these current language models.  The exception, perhaps, are programming skills which, whilst obviously requiring some critical thinking, have also been found to have a very strong positive association with exposure.

The below table shows the the professions that are found to be most exposed to the effects of LLMs.

By contrast, the research highlighted the below professions as being at least risk of exposure to the transformational effects of LLMs.

Does this mean that the impacted professions will "go away" or we will see mass unemployment amongst programmers or artists?  No, it's unlikely.  If AI did indeed cause mass unemployment, it would be the first time in human history that any technology has caused this.  

Boston University's James Bessen produced a fascinating piece of research on the effect of computers since the 1980s (also a general purpose technology).  He found that those professions that were exposed to the transformational effects of computers actually grew faster.  Computers didn't destroy jobs but actually led to the creation of many more jobs within the affected categories.  The below diagram depicts this nicely.

The reason is that automation of a profession increases demand for it because , whilst it allows more work to be done by less workers, there is more work to be done overall and firms can afford to then engage professionals to do this previously unaddressed work.  This is a pattern that we see across the economy and across history.  In an earlier article, Besen had examined the effects of ATMs on the employment of bank tellers and found remarkably that has the number of ATMs increased, the numbers of people employed as bank tellers also increased.

n conclusion, another way to look at this is perhaps not that AI will take over jobs but rather it will take over tasks.  Instead of of asking the question of what is the impact of AI – and LLMs specifically on jobs, it is better to ask what is the imapct on tasks and, for a given role, what percentage of that role consists of these tasks?  If we do this, we find that AI (or robots) can have several effects: it can make us more productive at existing tasks, it can shift labour to new tasks, or it can create new tasks for people to perform (such as labeling).  This excellent paper explores this in more detail.  

Anthony Butler Twitter

Anthony is a Senior Advisor to a G20 Central Bank on emerging technologies and applied research. He was previously Chief Technology Officer for IBM, Middle East and Africa. Lives in Saudi Arabia.