AI and not letting real life get in the way of (learning) theory – Part 2

In Part 1, I focused on a case where AI made it easier to step away from the real world in order to teach a model more clearly. That felt useful, but also something to handle with care, and there was a trade-off to be considered. This post is about the flip side. Situations where AI actually helps anchor our comparisons more firmly in reality, rather than pulling us away from it.

This came up for me recently when a colleague posted in the Edexcel Facebook group (highly recommended!) asking if anyone had any activities on labour markets. It reminded me of a ranking task I’d made a few years ago, where students were given a set of job descriptions and asked to rank them in order of how many applicants they thought each job would attract.

But when I went back to it with fresh eyes, I was less convinced. There weren’t really enough examples and the contrasts between them weren’t always stark enough. I wasn’t confident that all students would reliably take from the task what I wanted them to take from it.

That’s partly because of a more general issue with examples mentioned in Part 1: even when a teacher’s intention is clear, students don’t always attend to the same things. They notice what stands out to them, not necessarily what we hoped would stand out.

In labour markets, that problem is particularly acute. Real-world job adverts are nuanced and multi-dimensional. Wages vary alongside location, hours, qualifications, job security, and working conditions. Change one thing, and several others often change with it. That makes it very easy for students to latch onto unintended features and miss the underlying pattern.

There was also a very practical constraint. I have a vague memory of this activity taking me an inordinate amount of time to design the first time round. Guessing numbers, realising I have no idea what an office secretary in Scunthorpe earns, then noticing two examples were too similar, tweaking them, then realising I’d introduced a different problem instead. It probably wasn’t a great use of time. There also just weren’t enough examples for the task to really breathe. If I’m going to ask students to cut things out and physically rank them, I want the task to be deep enough for the discussion to last longer than it took me to cut the dang things out in the first place.

I wanted to redo it with more examples, clearer contrasts, and enough material to make faffing around with cutting worthwhile. I also found myself thinking about it from an observation mindset. If I were being observed, I’d want to control as many of the controllables as possible, and make it more likely that students would infer the factors affecting labour supply. I started from the outcomes. What had I actually intended students to infer? Wages. Flexibility. Qualifications. Experience. Location. Non-pecuniary benefits. From there, I used AI to help me find cases where one of those factors was particularly high, or particularly low, while the others were broadly moderate. (You can take a look here)

Previously, I was sanitising or contorting real examples to make them fit the activity. This time, I was finding real-world examples that already did the job I needed them to do. In this context, AI wasn’t fabricating an artificial world, it was a tool for finding those rare, cleaner, more varied real-world examples much more quickly than I could have done by hand. It also made it easier to put sensible numbers on things like wages and population, to help with students’ sense of scale.

That feels like the best of both worlds. The examples are now realistic, but less noisy. They allow for clean comparisons early on, at the point where I want students to draw out the theory itself and become confident about what shifts labour supply and why.

That doesn’t mean students never need to deal with messier examples, they absolutely do. Later on, once the theory is more secure, I might want students to go the other way: to take what they know about labour supply and use it to explain real-world wage differentials, job shortages, or recruitment problems, where lots of factors are interacting at once. At that stage, the messiness is the point, but here, my aim was different: to identify factors affecting labour supply. For that, having cleaner examples made the task both clearer and more worthwhile and AI made that far more doable.