A strong word for an unfinished story
The title speaks of a revolution. A more careful reading is that AI features are being added to many learning tools, that some of them are useful, and that the evidence on how much they improve learning is still developing and mixed. This article looks at what personalization means in practice, how it works with open content, and where it can go wrong.
What personalization usually means
Personalization is a broad label. In practice it usually refers to one or more of these:
- Pacing: letting a learner move faster through familiar material and slower through hard parts.
- Sequencing: choosing which topic, exercise or reading comes next.
- Difficulty: adjusting how hard the questions are.
- Format: offering text, audio or video versions of the same content.
- Support: providing hints, extra examples or reminders when a learner appears stuck.
None of these is new. Good teachers have always adjusted pace and examples. What AI adds is the ability to do some of it automatically, for many learners at once, using information the system gathers as people work.
How it works with open content
Open educational resources are well suited to this because they can be broken into parts, reordered and adapted. A system can treat a textbook chapter as a set of small units, pair them with practice questions, and present them in an order that fits a learner’s recorded progress. Where licenses allow adaptation, an educator can also edit a unit that is not working, instead of waiting for a publisher.
The inputs are usually modest: which questions you got right, how long you took, which items you skipped, and sometimes what you say you want to achieve. From these the system estimates what you probably know and chooses what to show next. It is an estimate only, and a rough one.
What these systems cannot see
A system knows only what it records. It does not know that you were interrupted, that you guessed, that the question was badly worded, or that you understood the idea but made a slip. Several consequences follow:
- Short-term performance is not the same as lasting understanding. Material that feels easy and is answered correctly now may be forgotten later.
- A system that keeps questions at a comfortable level may remove helpful challenge. Ideas such as desirable difficulties suggest that some struggle supports memory.
- Learners differ in ways that are never measured, such as motivation, health, or whether they have a quiet place to study.
For this reason, a personalized path is a hypothesis about what suits you, not a fact about you.
Bias and uneven quality
Systems learn from data about earlier learners. If those learners were mostly from one group, language background or school system, the system may serve other learners less well. Examples of how this can appear include different error rates across groups, recommendations that assume a certain level of prior schooling, or text analysis that handles some dialects poorly.
Bias is also quieter than that. A system that learns what similar learners did may keep steering people to the same content, which can narrow what is seen. And a tool that defines success as quick completion may favor speed over depth.
There is no complete fix. Reasonable steps include testing a tool with the actual learners it will serve, looking at results for different groups, and keeping a way for people to question or override a recommendation.
The learning styles trap
Some products claim to match content to a learner’s “style”, such as visual or auditory. The learning styles myth is worth knowing about: reviews of the research have generally not found good evidence that teaching to a preferred style improves outcomes. Claims of this kind should be treated with caution. Personalization based on what a learner already knows and what they are trying to do has a firmer rationale than personalization based on a label.
Privacy and consent
Personalization depends on data about how people learn, which can be sensitive. Learners and parents should be able to find out what is collected, why, how long it is kept and who sees it. Collecting less is usually better than collecting more. Reviewing privacy settings on learning accounts and thinking about your wider digital footprint are sensible habits whatever the tool.
Keeping the learner in charge
An important question is who decides. In some systems the software chooses and the learner follows. In others, the software suggests and the learner chooses. The second approach is often a better match for open education, which tends to emphasize learner control. Being able to see why something was recommended, skip it, or pick a different route helps people develop the ability to direct their own study, a theme explored under learner agency.
The role of teachers
Teachers add what the system lacks: knowledge of the individual, judgment about what matters, and the ability to notice when a recommendation makes no sense. In many settings the most useful role for software is to give a teacher a clearer picture of who needs help, with the teacher deciding what to do. Where no teacher is available, as in much self-study with open materials, learners may need to supply that oversight themselves by checking their progress against an outside standard, such as a syllabus or a past paper.
Questions to ask before relying on it
- What data does it use to decide what I see next?
- Can I see and change my path?
- How do I know it is working? Is there evidence beyond the supplier’s claims?
- Does it still show me material outside what it predicts I need?
- What happens to my data if I stop using it?
Reasonable expectations
AI-based personalization may help some learners move more efficiently, and it may be especially welcome where there is no teacher to adjust the pace. It may also add errors, narrow choices or collect more data than needed. Whether it is revolutionary depends on results that so far remain uneven. For now, treat it as one feature among many, check it against your own progress, and keep the final decisions in human hands.