Adaptive learning: when education learns from the student

AI-generated summary

Adaptive learning uses digital technology and AI to tailor education to each student’s needs by continuously observing their performance and adjusting the next activity, explanation, or difficulty accordingly. Systems build and update a student model from data such as correct/incorrect answers, response times, and repeated errors, then recommend targeted practice or alternative explanations. This isn’t merely delivering more content: it’s delivering the right content at the right time. AI techniques—machine learning, recommendation engines and generative models—expand these capabilities, enabling immediate feedback, ongoing diagnosis, and large-scale personalization that would be impossible for a teacher alone to provide. Examples like Khan Academy, DreamBox and Squirrel AI show modest but meaningful learning gains when tools are integrated with effective teaching.

Adaptive learning differs from broader “personalized learning” because it refers specifically to algorithm-driven, scalable adjustments; true personalization also includes motivation, collaboration and context that aren’t easily reduced to data. Benefits include faster remediation, targeted progression, and richer diagnostic information for teachers, who remain essential for contextual judgment, emotional support and complex learning tasks. Risks include privacy concerns, potential bias that pigeonholes learners, and widening digital divides tied to device access and implementation quality. Ultimately, adaptive systems are likely to complement—not replace—human educators, improving effectiveness and lifelong upskilling when deployed with teacher training, transparency, and equitable access.

Adaptive learning uses data and artificial intelligence to adjust content, pace, and activities to the needs of each learner. We look at how it works, how the major platforms are performing, and why their future will depend on technology as much as teachers, privacy, and equity.

For decades, much of education has been organized around a seemingly inevitable difficulty: a common teacher, curriculum, and calendar for students who learn in different ways and at different speeds. Digitalization and artificial intelligence are beginning to challenge this limitation.

Adaptive learning uses technology to observe how each student learns and modify their itinerary based on that information. If they master one concept, they can move forward. If they accumulate errors in another, the system can offer new explanations, simpler exercises, or reinforcement activities. The goal is not for each student to simply receive more digital content, but to receive the right content at the right time.

It is an evolution that fits in with some of the major educational trends that are transforming how we learn: more flexible models, continuous learning and an education that is less and less determined by an identical itinerary for everyone. It also converges with the development of hyper-personalized education, in which technology allows training to be adapted to the needs of a person throughout their life.

The relevant question for teachers, universities, EdTech companies, public officials and organizations that manage talent is not, therefore, whether there will be more technology in education. It is what decisions we want to delegate to it, which should remain in human hands and with what guarantees.

What is adaptive learning?

Definition and principles of adaptive learning

Adaptive learning is an educational approach in which a technological system dynamically modifies a student’s formative experience based on data on their performance.

To do this, they can analyse which answers they get right or wrong, how long they need to solve an exercise, how many attempts they make, what concepts they master or what previous knowledge seems to be causing a subsequent difficulty.

With this data, a model of the student is created and updated. The system then decides what the next content, question, explanation or level of difficulty should be.

Simplified, the cycle is:

Student interaction → collection of evidence → diagnosis → adaptation of the itinerary → new activity → new evidence.

The essential difference with a conventional digital content platform is in that last step. An online video or a quiz can be identical for thousands of users. An adaptive system tries to make the next experience depend on what has just happened in the previous one.

Artificial intelligence considerably expands these capabilities. Machine learning models, recommendation systems and, increasingly, generative AI capable of producing explanations, examples or contextualized feedback can be added to the pre-programmed rules. This is a territory related to the debate that we have already addressed when analysing whether generative AI can revolutionise education systems.

Difference Between Adaptive Learning and Personalized Learning

Although they are often used synonymously, they are not exactly the same.

Personalized learning is a broader pedagogical aspiration: to adapt objectives, content, methods, activities or rhythm to the characteristics of a person. This personalization can be carried out by a teacher without using any algorithm. Adaptive learning is one of the tools that allows this personalization to be scaled up through technology.

This distinction is important. Personalized education should not be reduced to getting an algorithm to correctly select the next exercise. It also involves motivation, curiosity, collaboration, creativity, personal context and emotional accompaniment: dimensions that are much more difficult to convert into data.

How does adaptive learning work?

Algorithms and artificial intelligence applied to teaching

Imagine a student solving an equation. The system doesn’t have to just record whether her answer is correct.

It can detect that she has taken much longer than usual, that she has repeated the same error three times, or that she has mastered the necessary algebra but has a previous gap related to fractions.

From there, different technologies can come into operation. A knowledge model estimates which concepts she masters. A recommendation algorithm selects the next activity. A predictive system calculates the probability that she will struggle in a later unit. And a generative AI can reformulate an explanation using another example.

It’s the transition from a technology that delivers content to one that tries to interpret the learning process.

The new intelligent interfaces applied to the classroom point precisely in this direction: systems capable of providing immediate feedback and relieving the teacher of certain repetitive tasks.

The adaptive cycle: assess, personalise and reinforce

A good adaptive system works continuously.

First it gets a diagnosis. Then it proposes an activity appropriate to the estimated level. It analyzes the answer. It updates the student’s profile and decides again.

This makes it possible to intervene in a way that in a large classroom is difficult to do manually hundreds of times a day.

But there is a particularly interesting consequence: the teacher also receives information.

Instead of discovering after an exam that fifteen students did not understand a concept explained three weeks earlier, he can have indicators during the process. The value of the system is therefore not to replace the teacher’s pedagogical capacity, but to increase his or her ability to observe.

Adaptive Learning vs. Traditional Education: Key Differences

Conventional educationAdaptive learning
Common pace for the groupPace that can vary per student
Default sequence of content Itinerarymodifiable according to performance
Evaluation at specific timesContinuous collection of evidence
The error is usually detected laterThe error can lead to immediate adaptation
The teacher interprets a limited amount of dataThe system can analyze thousands of interactions
Personalization difficult to scalePartially automatable customization

This does not mean that the right column is automatically better.

An algorithm can optimize the mastery of exercises and, at the same time, be unable to assess essential aspects of education. Nor does all knowledge lend itself equally to a sequence of measurable responses. Mathematics, languages or certain technical skills offer particularly favourable conditions; a philosophical discussion, a creative project or cooperative work present very different challenges.

The future will probably be less in replacing one model with another than in combining human and digital skills.

Adaptive Learning Examples and Platforms

Khan Academy: personalization and mastery learning

Khan Academy uses a mastery-oriented learning model: the student progresses by practicing competencies and the system can customize the exercises that appear in their Mastery Challenges according to their level and the time that has elapsed since they worked on a skill.

It is especially interesting because there is evidence of its real application in the classroom. Two field experiments carried out with mathematics teachers and students in the United States studied a program that combined computer-assisted learning – with Khan Academy among its tools – and continuous accompaniment to teachers. They found improvements of 0.12 to 0.22 standard deviations in mathematical performance, especially concentrated in classrooms where students accumulated at least 35 minutes of weekly practice.

The nuance is important: these results do not prove that an algorithm, by itself, produces the improvement. They show precisely something more interesting for educational institutions: technology, practice and good teaching implementation work as a system.

DreamBox: Adapt math while the student works

DreamBox has specialized in adaptive math learning. The platform analyzes the student’s interactions during the exercises and modifies their path based on them.

It also has experimental evaluations. In one of the randomized studies conducted with more than 500 first-year students, the group that used DreamBox obtained superior results on certain mathematical measures, with an effect size of 0.16 on a measurement and geometry subscale and 0.11 on the overall result. Other trials have found improvements in certain ages and competencies.

These are positive effects, but also a call to interpret the great promises of EdTech with caution: the impact depends on the context, the time of use, the content and the way in which the tool is integrated into teaching.

Squirrel AI: Decomposing knowledge to locate gaps

China’s Squirrel AI represents a more intensive approach. It divides subjects into numerous points of knowledge and tries to pinpoint where the gap is that hinders each student’s progress.

Two studies of eighth-graders compared their adaptive system with teacher-led teaching models and found greater gains in math for the groups that used the adaptive system.

The case illustrates both the potential and one of the sector’s central dilemmas: the more detailed the personalization, the more information the system needs about student behavior.

Smart Sparrow: a benchmark in the evolution of the sector

Smart Sparrow was one of the pioneering projects in enabling teachers to create adaptive digital experiences. Its trajectory also helps to understand how the market is evolving: Pearson acquired the company to incorporate interactive and personalized content capabilities into its digital products.

More than a collection of isolated tools, adaptive learning is thus beginning to be integrated into broader educational platforms.

The Foundation has already analyzed this convergence between AI, personalization and new educational models in This is what educational trends look like in times of AI.

The new role of the teacher: from transmitter to architect of learning

One of the simplest interpretations of educational artificial intelligence is to imagine a competition: algorithm vs. teacher. It’s probably the wrong question.

An adaptive system can process more interactions than a teacher could manually observe. But the teacher has something different: context.

He knows that one student is going through a difficult personal moment, that another needs to regain confidence, that a third student is bored because the exercise does not pose any challenge, or that a seemingly incorrect answer contains original reasoning.

The combination of both capabilities can transform the work of teachers.

If technology performs part of the routine diagnosis, identifies patterns, and proposes activities, the teacher can spend more time on accompaniment, conversations, collaborative projects, and higher-value interventions.

This requires, however, training. It is not enough to deliver a platform to a school. The results obtained in research on computer-assisted learning show precisely that the teacher’s ability to integrate the tool into his or her practice is a decisive variable.

Advantages and disadvantages of adaptive learning

The main benefit for the student is obvious: reducing the likelihood of spending too much time working on content that they have already mastered or advancing leaving significant gaps behind.

For teachers, it offers a diagnostic capacity that is difficult to achieve at scale. For universities and training organizations, it allows different experiences to be designed for heterogeneous groups. And for companies, it can open up a new generation of corporate training where each professional only covers the skills they need to develop.

But their risks grow at the same time as their capacity.

Privacy: When Learning Means Generating Data

An adaptive system can store academic results, interaction patterns, response times, and inferred profiles about strengths or difficulties. When users are minors, the issue becomes particularly sensitive.

UNESCO calls for a human-centered, age-appropriate, and privacy-based approach to educational AI.

In Europe, moreover, certain AI systems used in education may fall into the high-risk categories of the Artificial Intelligence Regulation. In such cases, the European framework sets out requirements related to risk management, data quality, traceability, human supervision, accuracy, robustness, and cybersecurity.

For EdTech developers, this makes privacy and algorithmic governance more than just a compliance issue: they must be part of the product design.

Biases: personalizing should not mean pigeonholing

There is another, less visible risk. If an algorithm concludes too early that a student is struggling with a subject, it can start permanently offering them easier content. The adaptation that was intended to help them could end up limiting them.

That is why it is essential that teachers and students can understand and question the recommendations of the system. Personalization should open paths, not turn a prediction into an academic destination.

The digital divide beyond having the Internet

Spain has very high levels of connectivity: in 2025, 97.4% of households had access to the Internet and the percentage reached 99.8% among households with dependent children. However, only 83.8% of households had a computer (desktop, laptop or tablet).

This is a relevant difference. The digital divide is no longer just about being connected or disconnected. What matters is the device available, the quality of the connection, digital skills, the family environment and the ability to use these technologies effectively.

If the most advanced tools only reach the most affluent schools and families, personalisation could widen the inequality it was intended to combat.

It is a dilemma closely related to the Foundation’s work on Technology and Inequality: innovation creates social value when its benefits can be extended to society as a whole.

Can adaptive learning be the future of education?

It will probably be an important part of that future, but not the whole future.

Its greatest potential is not in building an education without teachers. It is in solving one of the historical constraints of the classroom: the difficulty of providing individualized attention to many people simultaneously.

AI can turn thousands of small interactions into useful information. It can detect a gap before it becomes a failure. It can allow one student to advance faster and another to receive reinforcement. It can also help ensure that learning does not end with university, but accompanies each professional for decades.

This last dimension is especially relevant for companies and talent managers. In a labour market where skills are changing faster and faster, the same mechanisms can be applied to reskilling and upskilling: diagnosing what a person knows, determining what they need to learn and designing a specific itinerary.

This is a vision consistent with Akademia Future Builders, the Bankinter Innovation Foundation programme that connects university students with the technological trends that will define the future and seeks to develop a critical and proactive look at change. Akademia is not an adaptive learning system, but it shares an essential idea of the new educational paradigm: preparing for a future in transformation requires learning in a different way.

Technology can make education more personalised. The real challenge will be to make it more effective, more inclusive and more humane.

Frequently asked questions about adaptive

learningDoes adaptive learning replace the teacher?

It automates certain functions (diagnosis, recommendation of activities or follow-up), but the teacher is still essential to interpret the context, motivate, design experiences, accompany the student and decide when to follow or question an algorithmic recommendation. The available evidence also suggests that good teacher integration is decisive in obtaining results.

Is adaptive learning only for online education?

It can be used in completely digital models, but also within face-to-face classrooms and hybrid systems. A student can perform an adaptive part on a device and then use the face-to-face time to solve doubts, work in a team or develop projects.

Can adaptive learning be applied to corporate training?

Yes. Its logic is especially useful when an organization needs to train many people with different background knowledge. An initial diagnosis can identify skills already acquired and build specific reskilling or upskilling routes, avoiding that all professionals have to complete the exact same program.

In a lifelong learning context, the ability to detect what each person needs to learn and what their next step should be can end up being as important as the content catalog itself.