Buying Our Way into Bondage: The Risks of Adaptive Learning Services

The Perfect Storm

Much of the education technology world – and many of the foundations and venture firms that provide the funding for it – are obsessed with adaptive learning. The Gates Foundation’s Adaptive Learning Market Acceleration Program RFP is the most recent evidence of this trend. The fascination largely stems from the fact that, because these systems are completely automated, they can scale. Scale matters to foundations because it means broader impacts for the work they fund. And, of course, scale matters to investors because it means more customers and, consequently better returns.

But some educational content publishers love the idea of adaptive learning services for a different reason. Open educational resources (OER) are driving the cost of educational content to zero. In fact, you can now graduate from high school (e.g., Open High School of Utah) and complete an associates degree (e.g., Tidewater Community College) without ever spending a penny on textbooks – because of the pervasive use of OER in these programs.

Adaptive learning services are a perfect response to the business model challenges presented by OER to publishers. While the broad availability of free content (e.g., and OER have trained internet users to expect content to be free, many people are still willing to pay for services. Adaptive learning systems exploit this willingness by deeply intermingling content and services so that you cannot access one with using the other. Naturally, because an adaptive learning service is comprised of content plus adaptive services, it will be more expensive than static content used to be. And because it is a service, you cannot simply purchase it like you used to buy a textbook (particularly useful for publishers given the Court’s recent decision upholding the first sale doctrine with regard to textbooks). An adaptive learning service is something you subscribe to, like Netflix. And just like with Netflix, the day you stop paying for the service is the day you lose access to the service.

The Attack on Personal Property

Given the Court’s decision, it makes sense that some publishers would zero in on this leverage point. Whether it’s music on Spotify, movies on Netflix, or TV shows on Hulu, the content industry is engaged in an active campaign to undermine the idea of ownership of personal property. Why would a publisher sell you a CD or DVD, for which you pay only once, when they could persuade you to subscribe to a service for which you will pay every month for the rest of your life? Why would they sell you a CD or DVD which you can listen to or watch forever, loan to a friend, or sell to a used record store, when they could have you subscribe to a service by which they deprive you of any first sale rights?

In short, why is it in a content company’s interest to enable you to own anything? Put simply, it is not. When you own a copy, the publisher completely loses control over it. When you subscribe to content through a digital service (like an adaptive learning service), the publisher achieves complete and perfect control over you and your use of their content.

To the extent that publishers actually have these motivations, the attack on ownership of personal property is annoying in the context of entertainment, but becomes profoundly disturbing in the context of higher education. But in some sense, whether these are the publishers’ motives or not, the end results for learners are the same – the move to subscription models results in a number of significant problems.

How the Past Differs from the Future

In the past, students bought textbooks. Because students owned the books, they could be sold back, loaned to a friend, or students could opt to keep them for future reference. But when you subscribe to an adaptive learning service you own nothing, you can keep nothing, there’s nothing to loan to a friend or sell back, and there’s nothing to reference in the future. When your subscription ends, everything goes disappears. Need to review the material from that math class last year for this semester’s science class? Sorry! Your subscription expired at the end of last semester. Would you like to rent another four months of access for $129.99?

In the past, students could highlight and take notes in the books they owned. This kind of intensive, structured studying resulted in the creation of personalized artifacts that were a meaningful portion of what students’ knew at the conclusion of class. Many adaptive learning services encourage learners to highlight, take notes, and build other learning artifacts by annotating their content. However, because students own nothing, the day their subscription ends all of their notes, highlights, annotations, and other study artifacts are unceremoniously deleted. An important part of what they learned in the class is gone forever, because they couldn’t afford to keep subscribing forever. The situation essentially becomes “You will pay, or you will forget.”

In the past, when a publisher went out of business students could continue learning from the books they had purchased from the publisher. But when one of the companies providing an adaptive learning service goes out of business, “pivots” to focus on other products, gets acquired, or for other reasons end-of-life’s the service, what happens? Even if you could afford to continue paying for a subscription, everything vanishes and you have literally no recourse.

From Content to Data

There is no analog in the old publishing world for the models of learners that adaptive learning services create in order to do what they do. However, it is clear that these models begin as empty algorithms, and are entirely dependent on the learner creating and contributing data to the system in order to function. If the learner does not contribute data to the system, the service cannot build a model of the student upon which it can adapt its instructional, assessment, and other features.

The utility of an adaptive learning service is a function of the amount of a student’s data to which it has access. And while these data are created by the students, and therefore would typically be the property of the students, publishers claim ownership of these data through Terms of Use and other legal tactics and refuse to provide students with access to their own data. Consequently, the longer a learner uses a particular adaptive learning service, the higher the switching cost becomes to move to a different service – because publishers will not allow students to take their data with them, they will have to train the new system from scratch. What happens when Johnny transfers to the school across town that uses a different service? What happens when Sally graduates and goes to college? What happens when Pat transfers from the community college to the university? In these and all other cases, the student is back at square one.


Through a general strategy of preventing students from owning educational materials as personal property, including taking away learners’ rights to their own data, publishers could have a ready-made solution to the problem of price pressure from open educational resources. And whether this is any specific publisher’s motivation for the move to subscription-based adaptive learning services or not, the resulting impacts on students are the same.

Because some of the research on these systems suggests that they can be very effective at supporting learning, publishers can claim to be “doing the right thing for students” while increasing revenue and decreasing degrees of freedom for students and institutions. As a comparison point, migrating from one learning management system to another would be a pleasant walk in the Sunday afternoon park compared to the switching costs associated with moving from one of these services to another. This before we consider the drastically increased “cost of ownership” of the subscription model, in which you don’t actually own anything.

I am not arguing in favor or against the instructional effectiveness of adaptive learning services. I am simply pointing out the completely unprecedented risks involved in betting an entire school, district, university, or state system on a service with the properties described above.

If creating a system of “super lock in” and perfect control over students’ use of content are not primary design criteria for adaptive learning systems, then we should see the emergence of multiple adaptive learning systems that do not have these characteristics.

Openness is the Solution

Each of the problems with adaptive learning services evaporates when principles of openness are applied to these systems.

  • When the source code of an adaptive learning service is openly licensed (open source), even if a company or hosting service goes out of business, or gets acquired, etc., your institution can continue to utilize the service.
  • When the content in an adaptive learning service is openly licensed (OER), that content, together with students’ notes, highlights, annotations, and other work within the system can be exported, archived, and used by students forever.
  • When students own and can download the data they create and contribute to an adaptive learning service, they can maintain their own backups and make multiple uses of it – including potentially using that data with other systems.

Openness is the skeleton key that unlocks every attempt at vendor control and lock in.

Inasmuch as vendors are just beginning to encourage institutions to make their first adoptions of these adaptive learning services, there is still plenty of time for institutions to stand up for their students’ and their own best interests. Institutions should require guarantees regarding openness in the RFPs they create for the acquisition of these systems. No school has to race to adopt an adaptive system that doesn’t provide the guarantees necessary to protect the legitimate needs of the school and its students.

4 thoughts on “Buying Our Way into Bondage: The Risks of Adaptive Learning Services”

  1. This is an excellent analysis of a specific problem of adaptive systems that I had not really thought about before now. A key phrase refers to the risk of adopting a closed adaptive system. In times of rapid change the best strategies include agility – or the ability to undo a decision easily in the future. This also tends to favour low-investment strategies as they are easier to reverse. So what have I learned? If in doubt, avoid adaptive systems until an open system is developed. Any ideas on when that might happen?

  2. ” If creating a system of “super lock in” and perfect control over students’ use of content are not primary design criteria for adaptive learning systems, then we should see the emergence of multiple adaptive learning systems that do not have these characteristics. ”
    I’m working on an adaptive system at the moment. Not a very sophisticated one, I have to admit, but I’ll state now that my user data is staying closes. Not to control my students directly, but because my data model and my learning algorithms are very closely tied together. The user data would therefore be of very little use for the migration of users from one system to another. The only practical use for the user data would be to assist in the reverse-engineering of my algorithms.
    That is not to say that there won’t be any options for migration — in the long term there will have to be, for precisely the reasons you say, but practical, technical considerations mean that any migrated data will be a mere subset of the data that individual systems store.
    In the meantime, the same algorithms that are used to teach the individual can be modified to allow an assessment of the user’s current competence, and this can be supplemented with basic assumptions about syllabus ordering (eg addition is always taught before multiplication) used to find the smallest set of questions required to verify student knowledge.

    And given that testing and retesting is one of the most effective ways of consolidating learning, I don’t see why it should be a problem that a learner has to sit a small entrance test at the start of any new course.

  3. This critique seems to center around particular types of adaptive learning companies’ business models, not adaptive learning per se. There are adaptive learning companies which allow students indefinite access to his or her past work after a semester ends, for example. Certainly in those cases students’ data are not “unceremoniously deleted”.

    It’s true that if one of those companies folds, students will lose access to their past work — unless the program allows for a data export/download or something similar. I haven’t seen robust data export features before.

    The principle of the article seems sound but it seems to highlight the wrong problem..This is not a problem with adaptive learning; it is a problem with certain companies’ business models.

Comments are closed.