Innovation in education depends not just on bold ideas, but on the systems that help those ideas grow. Jeremy Roschelle, Executive Director of Learning Sciences Research at Digital Promise, is a leading learning scientist focused on building the infrastructure that makes rigorous, real-world research possible. From national collaborations that tap into existing classroom technologies to efforts that ensure AI is guided by evidence rather than hype, his work accelerates how learning tools are designed, tested, and improved. In this 5 Questions interview, Roschelle reflects on what he’s learned, why research infrastructure matters, and how it can shape the future of learning.
What is the nature of your work?
I am a learning scientist. I seek to understand how children learn complicated things and use that knowledge to design better supports for learning. And when innovators create a promising design, I aim to rigorously evaluate whether it actually works, and if it works for everyone and in all settings. There’s a virtuous cycle of developing knowledge of how people learn, using it to design better learning, and rigorously testing whether the designs work — and feeding the evaluation results back into the cycle.
These days, I often work on infrastructure that can accelerate this cycle. Research has been too slow, too expensive, and too disconnected from everyday teaching and learning. Infrastructure holds the key to getting more of what’s needed with less of the downsides.
Why is this work important?
Infrastructure for educational research means two things: ways of organizing people in communities to work together on important problems and ways of supporting the basic steps required in research so that the people can focus on their expertise, and not on so many basic details.
To explain why it’s important, here’s an analogy. There are many amazing open source communities that produce truly important software. One example: the Linux operating system is open source and it runs everything. Open source software development is only possible because of things like GitHub, a shared infrastructure for accessing code, improving it, and integrating the improvements. Without infrastructure, software engineers would find open source projects horribly time consuming and nearly impossible to coordinate. No one would participate in open source software without GitHub, or similar alternatives.
Now in the past 20 years, I got to lead several really major educational research projects. These were projects that collected data at scale, that conducted research rigorously, and did their work in typical schools and classrooms. It’s hard to do that kind of study; it’s called a randomized controlled study because you need to control what happens, and that’s difficult when you are fielding the study in schools. I looked around me and saw how few studies like this ever get completed and how expensive they are to do and realized that there has to be a better way.
What’s been the biggest surprise so far?
The better way starts with creating infrastructure to enable educational research. For example, in the past few years, I’ve been working on SEERnet, which is a collective of 12 teams, all funded by the U.S. Department of Education. Together, we are figuring out how to use the data that is collected in the technologies that are already used at scale in school. How can existing data, which is collected at huge volume every day, support important research? We’re creating infrastructure, which means things like ways to access data, ways to conduct comparisons, and ways to protect privacy while still answering the question “who does this work for and who does it NOT work for?”
Infrastructure means not only technical things, and also social things. How can researchers more rapidly understand this type of data, and what it works for? How can they create the kinds of social agreements and human subjects review processes necessary to use this data in research? How can researchers, developers and educators work together while also dividing effort in ways that work?
The biggest surprise has been twofold. On the one hand, I have been delighted with how many non-profits and companies would like to share the data in their products. And many of the technical problems can be solved, and solved elegantly and inexpensively. On the other, social infrastructure problems are harder than I thought. It turns out that so many institutional processes we have for educational research don’t exactly fit this new type of research—and this fault line emerges in how the research gets reviewed for funding, how it gets approved by human subjects review, and how schools become willing to participate. The new technical infrastructure collides with an old social infrastructure for research, and that’s not so easy to address.
Where do you see your work in five years?
I am super excited about how Artificial Intelligence and infrastructure are coming together in educational research, and that’s what I want to focus on in the next five years. To do AI right in education, we need more of the virtuous cycle I mentioned. We need more educational research, at scale, in real settings, with emerging technologies. We’re only going to get the volume of high quality research we need to guide AI to successful use in education if we double down on creating infrastructure that enables researchers, developers, and educators to come together and do the work efficiently and rigorously. And we also need infrastructure that makes it easier for AI developers to incorporate the learning science into their design and development process. I’m excited to be working on AI infrastructure for education with an awesome set of partners in a new collaborative effort.
What else should people know?
I would love people to know how important federal educational research funding is to the safety, trustworthiness and efficacy of the AI tools their children will use in school. I simply cannot imagine how decision makers are going to guide appropriate use of this rapidly evolving AI technologies in schools without research. And although companies and philanthropies are also super important to funding AI in Education research, I can’t see how companies and philanthropies could possibly make up for a weaker federal role. And it’s true that states and localities run schools in America, and we always involve educators in states and localities as partners in our research projects at Digital Promise. But we also recognize that states and localities do not have the wherewithal to be the primary funders of research. Now, with AI hitting our schools so rapidly and both with promise and concern, I’d like to ask my colleagues to do a better job of explaining how research leads to better opportunities for all learners to flourish. And I’d like to ask voters to advocate for the research they want to see and vote for the federal funds to make it happen.
