Multi-Layer Synthesis

A good decision about a learner is never made on a single piece of information.

A teacher sizing up a struggling student in a real classroom is doing something quietly astonishing without really noticing it. They are holding, simultaneously, the student's history of tests, a sense of their mood this week, a memory of something the parent mentioned at pickup, an awareness of what was covered two weeks ago, an intuition about which friends this student is sitting near today, and a read on whether today feels like a push day or a back-off day. None of these inputs on its own would justify a particular response. All of them together, weighted by experience, produce a judgement. It is not quick. It is not procedural. It is the kind of thing a good professional does, and the kind of thing that is very hard to copy.

An adaptive learning system that tries to imitate this has a choice. It can rely on a single signal, like whether the last answer was right, and respond quickly but shallowly. Or it can try to synthesise many signals from many layers, the profile of the learner, the sequence of the content, the modality that suits them, the context they are studying in, the governance rules about what is allowed, and something close to a judgement emerges. The second approach is harder, slower, and more expensive, and it is the only one that has any chance of mattering. Most platforms take the first approach and dress it in the language of the second. A real synthesis is rare, and when it appears it looks less like automation and more like something with a little bit of wisdom in it. When somebody you respect makes a good call, how many separate streams of information do you think they are really combining?

Last week we looked at decisions that are individually correct and collectively confused. This week we look at how a real decision about a learner draws on many sources at once.