I am a software engineer at MathWorks and a Ph.D. candidate at Florida Atlantic University, focusing on machine learning (more…)

4 minute read

One of my favorites sources to keep track of current affairs, The Economist, published in January of 2021 a column on the effect of automation on jobs, “New research shows the robots are coming for jobs—but stealthily” (need a subscription to read in full).

Among the reasons I like The Economist is because it is dense (in a good way). For example, this one-page piece cites three recent academic sources. It takes some effort to follow the complex lines of argumentation for multidimensional, real-world problems like this one, the dynamics of the job market (well, it takes me some effort - your mileage may vary), but it is rewarding in the end.

Rewarding, if nothing else, to gain an appreciation for the multiple facets the topic has. It is not a topic that has a simple answer. The piece does not attempt to answer a simple question (“will automation destroy jobs?”). Instead, it explains the forces that drive the problem and the current understanding of people who are paid to think about it.

The following sections summarize the sources the article cites and their understanding of the “automation vs. job” debate.

AI and jobs: evidence from online vacancies

Daron Acemoglu, David Autor, Jonathon Hazell, Pascual Restrepo https://www.nber.org/system/files/working_papers/w28257/w28257.pdf

It argues, based on recent and detailed job figures, that in aggregate, automation is not reducing the number of jobs out there: “we find no discernible impact of AI exposure on employment or wages at the occupation or industry level, implying that AI is currently substituting for humans in a subset of tasks but it is not yet having detectable aggregate labor market consequences.”

However, the paper concentrates on AI as a task-replacement technology: “our focus on AI adoption that is driven by the task structure of establishments may leave out other types of AI impacts that are less related to task structures, such as the use of AI to launch new products and services.”

This is a good segway for the following paper, which argues a different point.

Artificial Intelligence Technologies and Aggregate Growth Prospects

Timothy Bresnahan https://web.stanford.edu/~tbres/AI_Technologies_in_use.pdf

It argues that “the most valuable applications of AI have nothing to do with displacing humans” [The Economist quote]. In the author’s words, there is “no substitution of machine for human work at the task level”. Instead, these applications are in new areas that create competitive advantages in a way that doesn’t directly involve labor.

For example, Amazon’s recommendation system improves Amazon’s ability to not only sell more but also improve its logistic (distributing products around warehouses, close to where they are needed, anticipating demand). This makes Amazon a stronger competitor against regular stores, even though it doesn’t have stores.

The author calls this application a “system level” disruption. The job of employees of regular stores is in jeopardy while Amazon grows its workforce by using AI. This point is argued more generically by the following paper.

Artificial Intelligence: The Ambiguous Labor Market Impact of Automating Prediction

Ajay Agrawal, Joshua S. Gans, Avi Goldfarb https://www.aeaweb.org/articles?id=10.1257/jep.33.2.31

It argues that the AI application in recommendations, or in “prediction” (as in “machine learning prediction”), will bring about fundamental changes to the jobs that involve decision making.

The argument starts with the fact that we need predictions to support a decision. In other words, the goal is the decision, not the predictions. As predictions get better and better, we will reach a point where they are so good that they will “creat[e] new decision tasks when automating prediction sufficiently reduces uncertainty as to enable new decisions that were not feasible before.”

This effect on decision making, first by augmenting it with better predictions (the stage we are in now), then eventually replacing it with the actual (and novel) decisions, will create dramatic changes to the economy.

This is a very readable paper, with many concrete examples of current jobs and roles and how much they could be affected by the improvements in prediction (and decision making) through AI.

Feb 2022 update: “Economists are revising their views on robots and jobs”

Another The Economist article, Economists are revising their views on robots and jobs.

This article is about robots, not artificial intelligence. However, it is the same general topic of “effect of automation on jobs”.

The article’s main point is that economists are revising how they measure the effect of automation on jobs. Their new methods concluded that automation increases jobs, at least at the company level.

Being The Economist, it cites academic papers to back up the claims:

The new research raises a few questions:

  • Why have economists changed their views? The Economist credits the change to new statistical methods that untangle cause and effect from the numbers.
  • Is this all good news? It is still open for debate. The research cited above does not evaluate the quality of the jobs created or left in place. It also focuses on the effect at the company level, not on the economy (across companies).

I know ending a post without a conclusion is anti-climatic. But in this case, it’s an acknowledgment of the complexity of the “AI vs. jobs” discussion. Going through these different points of view helped me avoid the trap of having a simplistic opinion about a nuanced topic.

You may also enjoy

Using LLMs to summarize GitHub issues

16 minute read

A learning exercise on using large language models (LLMs) for summarization. It uses GitHub issues as a practical use case that we can relate to.

Vision transformer properties

13 minute read

Vision transformers are not just a replacement for CNNs and RNNs. They have some interesting properties.