AI-personalised learning paths: training that adapts to the person

The typical corporate course treats everyone the same way. Same modules, same order, same pace, regardless of what each person already knows. The expert is bored re-reading things they have mastered for years, the beginner struggles to keep up, and almost no one gets what they actually need at that moment. It is a compromise companies accept because a single course is simple to build and to distribute, yet the price is paid by people's attention and by the effectiveness of the training. Personalisation exists to get past that compromise, and artificial intelligence is what finally makes it possible at scale.

The limits of one size fits all

When the same path is handed to hundreds of people with different histories and skills, a mismatch is hard to avoid. For some the content is obvious, and the time spent reviewing it is time taken away from work; for others it arrives too fast, and what goes ununderstood drags on as a gap. In between there is a band of people for whom the course is more or less fine, and who could still learn more from a path tuned to their own needs.

The result is training that works halfway for almost everyone and brilliantly for very few. The most visible consequence is a loss of motivation, because few things dampen the desire to learn as much as feeling treated like a number in an undifferentiated class, held to a pace that is not your own.

Learning at the right level for each person

The idea that difficulty should be calibrated to the person is not a recent one. Lev Vygotsky described what he called the zone of proximal development, the space in which a piece of content is challenging enough to make you grow and accessible enough to be tackled with a little support. That is where learning pays off most, and every person has their own zone, which shifts over time and with progress.

A fixed course, by its nature, cannot sit inside everyone's right zone at once, because it picks a single one and asks everyone else to adjust. Personalisation turns this logic around and tries to bring each person to their own ideal point, where the effort is the productive kind that comes with learning rather than the sterile kind that comes with boredom or with getting lost.

What the research says about adaptive systems

The idea of training that adapts to the person has a history that begins with the dream of giving everyone a dedicated tutor, far too expensive to actually offer to all. Intelligent tutoring systems were born to approach that result with technology, and the research that has measured them reports encouraging numbers.

Kurt VanLehn, in a wide-ranging 2011 review, first cut a popular myth down to size, because one-to-one human tutoring does not produce the legendary two sigma of improvement but an effect of about 0.79 standard deviations over no tutoring at all. The interesting part is the comparison, since intelligent tutoring systems, able to adapt content and pace to the learner, reached 0.76, nearly as much as a tutor in the flesh. A 2016 meta-analysis by Kulik and Fletcher, across fifty controlled evaluations, found a median effect of 0.66 standard deviations, while noting that the result changes a good deal depending on how it is measured. The figures call for the caution any meta-analysis deserves, and yet the signal holds: a system that adapts training to the person comes remarkably close to what a dedicated teacher would achieve.

What exactly to personalise

A point of method heads off a very common misunderstanding. Personalising does not mean indulging supposed learning styles, the idea that there are visual, auditory or kinaesthetic people each to be served in their own channel. That theory, popular as it is, has found no support in research, since the review by Pashler and colleagues in 2008 found no adequate evidence that matching teaching to a declared style improves results. Personalisation that works rests on something concrete and measurable, namely what the person already knows, how they are doing along the path, the pace they keep and the goals in front of them. Adapting a path on these grounds makes sense and produces effects, while adapting it to a label about a preferred channel is a road science has left behind.

How Evolve personalises learning paths

Evolve, AWorld's Learning Experience Platform, puts personalisation at the centre of the experience, and it does so on a different plane from the AI Co-pilot. Where the Co-pilot generates content from existing materials, here artificial intelligence works on what to offer each person and in what order.

In practice, mandatory paths assigned by the company on subjects such as safety or compliance sit alongside suggested paths, which the AI recommends to each person based on their role, their interests and the progress already made. This way a person does not face an undifferentiated catalogue but a proposal that takes account of where they are. The same logic of adaptation works inside the paths too, because the final assessment brings back to each person exactly the questions they got wrong along the way, with unlimited attempts until they are passed, tuning the review to real mistakes rather than to a programme that is identical for everyone. All of this rests on microlearning <!-- internal link: EN microlearning article -->, which breaks content into short units and so makes it easier to reassemble them into different sequences for different people.

The AI proposes, people set the direction

One note keeps expectations in the right place. Automatic personalisation is a powerful tool, and it works when it rests on well-made content and clear goals, which remain a human responsibility. It is the company that decides what is important to learn and why, while the AI takes care of bringing the right thing to the right person at the right time, easing a job of orchestration that would be impossible to do by hand at scale. The value is in freeing those who lead training from the thankless task of treating everyone the same out of sheer necessity, more than in handing the choices over to an algorithm.

Frequently asked questions

What are AI-personalised learning paths? They are training paths that adapt to the individual instead of offering everyone the same content in the same order. Artificial intelligence recommends what to tackle based on what the person already knows, their progress and their goals, coming close to what a dedicated tutor would do.

Does personalised training work better than one course for everyone? The research on adaptive systems is encouraging. The reviews by VanLehn and by Kulik and Fletcher show that a system able to adapt content and pace comes close to the effectiveness of a human tutor, with effects that vary by context and by how they are measured.

Does personalising mean adapting content to learning styles? No. The idea of learning styles has no support in research. Useful personalisation rests on concrete data such as prior knowledge, progress, pace and the person's goals, not on a label about a supposed preferred channel.

How does Evolve personalise learning paths? Evolve places AI-suggested paths alongside mandatory ones, based on role, interests and progress, and it adapts review through the final assessment, which brings back to each person the questions they got wrong until they are passed. Personalisation therefore covers both what is offered and how what has been learned is consolidated.

Giving each person their own path

For a long time mass training traded effectiveness for convenience, offering everyone the same course because it was the only sustainable way to reach large numbers. Technology has changed the terms of that trade, and today it is possible to give each person a path close to their own, without assigning a teacher to every individual. It is the most concrete way to bring training back to where it pays off most, which is the right point for the person who is learning.

If you want to see how to build paths that adapt to your people, discover Evolve and talk to our team.

Sources

• VanLehn, K. (2011). The Relative Effectiveness of Human Tutoring, Intelligent Tutoring Systems, and Other Tutoring Systems. Educational Psychologist, 46(4), 197-221.
• Kulik, J. A., & Fletcher, J. D. (2016). Effectiveness of Intelligent Tutoring Systems: A Meta-Analytic Review. Review of Educational Research, 86(1), 42-78.
• Vygotsky, L. S. (1978). Mind in Society: The Development of Higher Psychological Processes. Harvard University Press.
• Pashler, H., McDaniel, M., Rohrer, D., & Bjork, R. (2008). Learning Styles: Concepts and Evidence. Psychological Science in the Public Interest, 9(3), 105-119.
• Bloom, B. S. (1984). The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring. Educational Researcher, 13(6), 4-16.