A few months ago, I wrote about which jobs Artificial Intelligence tools like Chat-GPT is likely to come for first. The conclusion was that jobs heavy in writing while low in social perceptiveness and critical thinking lay in the crosshairs. After all, a chat bot can’t easily react to people’s emotions. Indeed, some have found chat bots actively creepy. And, chat bots are only trained on existing knowledge. Thinking critically to come up with alternative solutions isn’t really in their wheel house. So, high-paying jobs like CEO (high on critical thinking and social perceptiveness) and physicist (very high on critical thinking) may be safe. Lower paying jobs like proofreaders or legal secretaries are more likely in danger.
The question for this post is: who exactly works in these jobs? What are the demographics of workers likely to be safe (for now)? And, who exactly is most likely to be affected as generative Artificial Intelligence advances? Will the polarization that characterized the advent of computing happen again, with middle-income workers being hurt more than those at the bottom or top? Let’s turn to the data and give it a look.
(Also, if you like this post and want something to read this summer, buy my book! According to me, it’s an awesome beach read. Or, if you hate the post and want something to burn at a BBQ, buy my book! It’s totally flammable.)
The Demographics of AI Exposure
To explore the types of workers who may be affected by AI, I combine two datasets. First, the American Community Survey, which contains data on workers and their occupations. And second, the O*NET Database, which contains data on the skill content of various jobs. Does the job require writing, interaction with others, thinking critically, etc.?
Then, I bunch workers into five, mutually-exclusive groups. The first is workers whose jobs assign little importance to writing. These workers are not exposed to chat bots, because their jobs don’t rely on what these chat bots have been trained to produce. The next four groups all have jobs that place importance on writing and thus have some exposure to chat-bot induced shenanigans. The first group is most at risk — their jobs place little importance on social perceptiveness or critical thinking. The next two groups are at moderate risk — their jobs place importance on either social perceptiveness or critical thinking. And, the final group is least at risk, as their jobs require high levels of social perceptiveness and critical thinking.
AI Risk and Income
The graph below shows the income of workers in these various risk groups. The graph is non-linear. The jobs at the far left — least at risk because they don’t involve much writing — are also the lowest paying. The jobs on the far right — most at risk because they involve writing but little social perception or critical thinking — are the next lowest paying. And the jobs that involve writing but place importance on social perception and/or critical thinking pay the most.
Figure 1. Average Annual Earnings of Workers by Level of Writing, Social Perception, and Critical Thinking at their Jobs
This graph shows a kind of “polarization” that has occurred in the past. It wasn’t the lowest paid workers who were affected by the last wave of automated technology, nor the highest paid. Robots don’t cut our hair (yet!), and hairdressers tend to be lower paid than other occupations. And, robots can’t yet do complicated statistics, and statisticians are highly paid. But, robots can screw a bolt into a car, and that work used to be squarely middle class. The future looks similar. Jobs that pay the least aren’t as exposed to artificial intelligence as jobs in the middle, which are more exposed than jobs that at top.
AI Risk and Demographics
On the demographic side, things are also somewhat non-linear. The figure below shows the share of people in each of the five risk categories of job by gender, race/ethnicity, and education. The red bars at the top of each column shows that across all these groups, 15 – 22 percent of workers are in occupations that are most at risk of exposure to AI.
Figure 2. Share of Workers in Various Risk Categories, by Select Demographics
So, most groups have roughly a fifth of their workers in the most exposed category. Still, some differences are worth noting. For example, women are more exposed to AI than men. This exposure occurs mainly because women are simply more likely to be in jobs that require writing.
,It’s also worth noting that Black and Hispanic workers appear less exposed than White workers. However, the underlying reason is somewhat disheartening; Black and Hispanic workers are more likely to be in low-paying jobs that require little writing relative to White workers. And, while those with a Bachelor’s degrees appear slightly more exposed than those without a degree, the primary reason is simply that their jobs are much more likely to rely on writing. Indeed, within jobs that place importance on writing, those with a degree are much more likely to be in “safe” jobs requiring both social perceptiveness and critical thinking than those without a degree.
Heading into an AI Future with Real Intelligence
To me, the story these data tell is two-fold. First, the effect of AI is likely to be non-linear with respect to income. Workers at the bottom and the top seem less likely to be immediately affected. Those in the middle will have to cope. You can tell from my language — words like “risk” and “cope” — that I view the likely affect as a negative for these workers. If the past is predictive, then AI seems likely to replace many of the workers it affects, placing downward pressure on their wages. However, people I respect greatly have raised the possibility of the reverse. Perhaps Artificial Intelligence will augment the productivity of middle-class workers, raising their wages. For example, Chat-GPT could help a middling writer be better and thus earn more, whereas AI could hardly help an excellent writer like me (or is it “like myself”? I refuse to ask Chat-GPT). Here’s hoping.
The second part of the story is that AI appears to affect everyone. Across gender, race, and education, about a fifth of workers seem to be in for whatever effect occurs.
Which gets me to my point. We know some kind of broad effect is coming. And, we have time as a society to think about how we want to regulate these technologies, and which types of AI we want to allow to exist. Indeed, the Biden Administration has already started this process, laying out an AI “Bill of Rights.” My favorite of the points laid out in this document is the requirement of “Human Alternatives, Consideration, and Fallback.” The idea would be to allow people to opt out of automated systems, and have access to human help. To me, this sounds good from both a labor market perspective and from the prospective of a customer. So, ideas do exist from the executive branch. Is their the will to do something from congress? I’m not holding my breath.
Instead, I expect that we will just allow the corporations responsible for these technologies to regulate themselves. These are the same sorts of people that brought us Facebook, Instagram, and Twitter. Things the Surgeon General is now considering warning children about. Do I expect these companies to do a good job regulating these things. I’m definitely not holding my breath.