What an adaptive algorithm is
An adaptive algorithm is a set of rules, or a statistical model, that changes what a learner sees based on how they have responded so far. The simplest version is a rule: if a learner misses two questions on a topic, show a short review before moving on. More complex versions estimate how likely a learner is to know each skill and choose the next activity to fit that estimate.
Adaptive course design predates the current wave of AI tools. What may change is how finely the adjustments are made and how easily they can be added to ordinary courses.
A small, hypothetical example
Imagine an open course on basic fractions. A learner answers a set of questions. The system notices that mistakes cluster around finding common denominators, while other skills are secure. It offers a short worked explanation and a few more practice items on that skill, and skips ahead past the parts the learner has already shown they know.
This example is invented for illustration. Real systems differ, and each has its own way of deciding what counts as “secure”. The point is only that the course path is not the same for everyone.
Why open courses are a natural setting
Open courses are often built from reusable pieces: short readings, exercises, videos and quizzes with open licenses. Pieces that can be separated and recombined are easier for an algorithm to arrange. Open licenses also let course teams fix or replace a weak piece, rather than waiting on an outside owner. A course based on competency-based education, where progress depends on showing a skill rather than spending time, fits especially naturally with adaptive routing.
What adaptation may do well
- Reduce repetition for learners who already know part of the material.
- Give extra practice where a learner is struggling, at the moment they need it.
- Offer a gentle pace for people studying around jobs and family, who may not follow a fixed timetable. This suits self-paced study.
- Spot patterns for course teams, such as a question most learners get wrong, which may point to a flaw in the question rather than in the learners.
How much any of this improves learning is a matter for evidence, and findings so far are mixed and depend on the course, the learners and the design.
What it may do badly
- Mistake performance for understanding. A learner may answer correctly by guessing or by pattern-matching.
- Over-narrow the path. If a system only gives you what it predicts you need, you may miss useful background or alternative viewpoints.
- Reflect poor data. Estimates are only as good as the questions and the data behind them.
- Reduce motivation. Some learners dislike a path they cannot see or control.
Algorithms also work poorly on things that are hard to turn into right-or-wrong answers: discussion, creative work, extended writing, collaborative projects and ethical judgment. These are central to many courses.
Where human teachers remain essential
It is tempting to imagine adaptive courses running by themselves. Several things suggest that teachers will remain important.
First, someone must decide what the course is for. Goals, standards and what counts as good work are human choices. Second, teachers notice what data cannot show, such as distress, confusion hidden behind correct answers, or a learner’s circumstances. Third, feedback on complex work is a skill that current tools do only partly, and can do wrongly. Fourth, people learn from other people: encouragement, modeling and discussion matter, and a teacher or facilitator often provides these. A shift towards the role of teacher as facilitator may be the more realistic change, with teachers spending less time on routine marking and more on guidance.
Fairness and transparency
Learners should be told when a course adapts and, ideally, how. If a system routes people to easier or harder material, it should be possible to ask why and to choose differently. Course teams should check that adaptation does not systematically disadvantage some groups, for example learners who read more slowly in a second language or who use assistive technology. Where outcomes differ, the first suspect may be the course design, not the learner.
Data and privacy
Adaptation requires recording what each learner does. Open courses that run on shared infrastructure should say what is recorded and for how long, and limit it to what is needed. Learners can reasonably ask for access to their own records. The wider topic of personal data sharing is relevant here: if a course asks for more than it seems to need, that is worth questioning.
Access still comes first
Adaptive features do not help people who cannot get online. In places where connections are expensive or unreliable, a well-designed course that works offline may serve more people than an advanced one that does not. Open education has often aimed to widen access, and adaptive technology should be judged by whether it supports that aim, not only by how clever it is.
Predicting carefully
Forecasts about technology in education have a mixed track record, with many predicted changes arriving slowly or in a different form. A reasonable guess is that adaptive features will become common but modest: better practice, smarter reviews, clearer progress views. A reasonable hope is that they free teachers for the human parts of teaching. Beyond that, it is wiser to watch how real courses perform than to rely on promises.