No bad poetry this week, but I am talking about AI again. What? I don’t make the higher ed news; I just provide the commentary. In this week’s issue, we’re going to put on our economists' hats again and dive deeply into a new paper from the U.S. Census Bureau that attempts to find connections between how AI-exposed a major is and the employment outcomes of its graduates. I then close with three Sparks about measuring learning, skill erosion, and private college tuition. 

Is AI Disrupting the Labor Market for New Grads? 

From Graduating into Disruption: Labor Market Outcomes for AI-Exposed College Majors | U.S. Census Bureau 

The Center for Economic Studies evaluates how different college majors have fared in the labor market based on their levels of AI exposure. 

Our Thoughts  

At this point, it’s starting to feel like I need to go back to campus to take some economics courses if I want to continue commenting on these stories. If I had known higher education and economics were to be so entwined during my career, I might have paid more attention to my roommate’s macroeconomic ramblings during my sophomore year. While both Inside Higher Ed and The Chronicle of Higher Education provided news coverage of this paper, I’ve chosen to link to the paper itself this week because I’m going to go further into the paper than either news site did. Both did a great job providing an overview of the paper, but they gloss over some of the juicier bits. Let’s dig in! 

In short, the researchers used graduation data from more than 350 participating institutions and linked that data at the individual level to state unemployment insurance wage records to see how those graduates fared in the labor market. Using prior research on the level of AI-exposed occupations and linking majors to occupations using national survey data, the researchers were able to see how the level of AI exposure for each major translated into employment opportunities and wages. The authors performed some economic math to translate all of this into deciles to see how each major performed before and after the release of ChatGPT in late 2022. Unlike the research from Issue 119, whose methodology I poked holes, this one is pretty good and uses government wage records rather than private payroll or startup data.  

So, when you read the news headlines, the data is hard to ignore. Compared to the less exposed majors and measured against 2022, graduates in the top decile of AI-exposed majors were five percentage points less likely to have a job in the quarter after graduation, and those who did get a job were paid about 13 percent less on average. Over time, these gaps do begin to close, with graduates two years out showing a two percentage point employment gap and earnings about five percent lower. While this is a step in the right direction, it’s still not great for these recent graduates as studies of graduating into a recession have found that these early gaps are hard to recover fromt as a person’s career progresses.  

But, when we look at the paper more deeply, we see the headlines only tell part of the story. First, this is primarily a computing story as most of the majors in the top decile are computing-related (e.g., computer science, information systems, computer engineering, etc.). Other majors such as finance and accounting that had high AI-exposure scores didn’t take the same hit as computing-related majors and saw only small effects in the job market. The authors don’t fully explain the gap and concede it may be difficult to separate AI’s effects on graduate employment from the broader technology market slowdown. Though they still argue that they believe AI played a substantial role.  

Next, we see that the research tells us less about higher education overall in the U.S. and more about specific segments of the sector. The data set used for the study covers about 29 percent of bachelor’s degrees, and 93 percent of the participating institutions are public, meaning they are overwhelmingly represented in the sample. In the authors’ national comparison, public institutions comprise a little over half of four-year institutions. This means the paper tells us a lot about the labor market experiences of graduates from big public universities and almost nothing about the experiences of graduates from private institutions. Additionally, the study counted graduates who went to graduate school or those who are self-employed as not employed because they did not have any wage data. The authors concede that this could account for as much as half of the initial drop in employment while also arguing that graduates may have taken these routes because they couldn’t find employment. This doesn’t mean the study is bad; it just means it tells a different story from the reporting. This is an important distinction should the paper come up in a committee report or board meeting. 

However, there is one idea the authors float that I think deserves more attention. In their paper, they state that the effects on new graduates in the labor market showed up before most employers had adopted AI. Their study compared the time before AI was generally available to after it was generally available in late 2022. Most employers didn’t adopt it immediately, especially not in a way that AI could be the sole reason for hiring declines in summer 2023. One possibility the authors suggest is that what we’re seeing is less of an AI impact from company adoption and more of an impact on what grades and coursework signal to potential employers. Perhaps graduates struggled to find jobs because employers doubted they could actually do the things their transcript said they could. Perhaps employers were concerned that graduates had simply been handing in AI-generated work to earn their A’s and B’s. While there have been some studies connecting the availability of AI to grade inflation, I’m not sure that would explain the drop so quickly in summer 2023. Then again, ChatGPT’s ability to write a passable essay was front-page news by 2023 so perhaps it was a contributing factor. 

Even so, these are important questions to consider, and this study does provide some data to begin answering questions about AI’s impact on the employment outlook for graduates and majors. An institution could replicate one of those research studies against their own data to see how their courses and majors perform. If you do, be sure to start the discussion on your campus about what you might do about it. Conversations about oral exams and assessment redesign have mainly been framed around academic integrity, but if you believe portions of this paper, getting it wrong could have impacts well beyond fair grading.  

Sparks 

• The Way(s) Forward on Measuring Learning (Inside Higher Ed) - In this three-part series, Doug Lederman discusses how higher education could measure learning. While I’ve linked to the third part, I recommend giving the entire series a read. Probably one of the more important questions we need to soon.
• Report: AI May Be Eroding the Very Skills Employers Need Most (Campus Technology) - A new report from IBM finds that AI may be eroding the critical thinking and judgement skills needed in the modern workplace. While this is not the first study to find these results, I appreciate that this study was global and not focused solely on the U.S.
• More Private Colleges Are Trying to Woo Students With Public-Flagship Prices (The Chronicle of Higher Education) - An increasing number of private institutions are competing with neighboring public institutions on price for select student populations. So, not quite a tuition reset, but a market indicator of what may be to come around price.