Can Learning Become Predictive?

The strong claim — that a system can know what a particular student will do — is not credible, and institutions are right to be skeptical of anyone making it. Human learning is contingent, contextual, and responsive to intervention. A model that claimed determinism would be wrong about the most important cases.

The weaker claim is both defensible and valuable: given sufficient contextual signal, a system can identify conditions statistically associated with particular outcomes, and can surface those conditions early enough for a person to respond.

01Prediction as a set of pathways, not a verdict

We model possible pathways rather than singular futures. From a current state, a set of trajectories is plausible. Each carries an estimated likelihood, a set of contributing conditions, and a set of decisions that measurably shift it.

The purpose of a prediction in education is to make itself wrong.

This inverts the usual framing. A flagged risk pathway is not a forecast to be validated. It is an invitation to intervene, and the measure of the system is whether the intervention changed the trajectory.

02What responsible predictive learning requires

• 01Explainability. A pathway that cannot be explained cannot be acted on responsibly.
• 02Actionability. If nothing can be done differently, the prediction is surveillance rather than support.
• 03Bias examination. Models trained on historical outcomes can reproduce historical inequities; this must be tested continuously, not assumed away.
• 04Human authority. The system informs a decision. A person makes it.