Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Upgrade to the On EdTech+ newsletter.

I have been in wonderful Norway this week, and most of my reading has consisted of little cards describing the five different kinds of smoked salmon at the breakfast buffet.

And amazement and amusement at the vast array of different kinds of cheese slicers you can buy. They take categories seriously here.

But I did manage to squeeze in a couple of EdTech related things on the train ride up north. As it turns out, categorization was also the theme running through much of my reading this week. Synthetic students require us to think about the difference between prediction and simulation. Richard Merrick asks us to distinguish between knowledge and judgment. Sam Illingworth explores the difference between educational support and compliance. The Norwegians may be onto something.

While at the Open edX Conference in Salt Lake City (more from me about that next week) one of the most interesting conversations I had was with an institution that is experimenting with simulated students using AI. I am hoping to learn more about that initiative soon, but I read an interesting report from Bain & Company about their use of simulated customers. I found it useful both for understanding the potential applications of simulated users and for thinking through the limitations of the approach, especially as we begin to see more use of simulated students and digital twins in online learning and student success.

Synthetic students represent a shift from predicting student behavior to simulating student behavior. For years higher education has invested heavily in either marketing data or predictive analytics that attempt to identify what students are likely to do or which students are at risk. Synthetic students point toward something different: the ability to test policies, interventions, messages, and program designs on AI-generated versions of students before implementing them in the real world.

Synthetic customers or students rely on a mix of AI models and proprietary and open data to create realistic representations of the choices real customers would make, as described by the Bain report.

Bain argues that improvements in LLMs have made synthetic customer models more reliable and capable of testing ideas far more quickly and cheaply than before. They point to applications ranging from pricing and product design to testing messages and concepts with hard-to-reach audiences.