At a moment when technologies like AI are reshaping education and workforce training, questions about how people learn — and how learning environments can be improved — remain central. Bror Saxberg is the Founder of LearningForge LLC and a learning engineer whose work focuses on applying the science of learning, motivation, identity, and stress to the design of educational and training systems. Over the past three decades, he has worked with schools, universities, philanthropies, investors, and product developers to translate evidence from learning science into more effective practice. In this 5 Questions interview, Saxberg reflects on why education has been slow to incorporate what research already shows, the risks of over-relying on new technologies, and what it takes to build learning systems grounded in evidence rather than intuition.
What do you do?
I have been a learning engineer for 30 years, meaning I’ve been helping improve learning environments by bringing in the science behind learning, motivation, and identity and belonging to help make learning environments at many levels more successful. I’ve also helped developers, investors, philanthropies, and more improve their existing learning environments through inclusive design — working with their stakeholders (especially students and teachers) and science to find the next best things to try to improve learning outcomes.
Why is this work important?
Unlike some other ecosystems, like medicine, education has not taken advantage of the many empirical findings about learning, motivation, identity, and stress that have emerged in the last several decades to give guidance to how to do better. Some domains, like chemical engineering, show how a field can evolve beyond pure science by combining scientific knowledge with practical concerns like economics and safety to decide how to build and improve real-world systems.
We need more of this blending of empirical findings with technical capabilities to improve learning at scale — especially now, with new technologies like AI looking like they “ought to be helpful.” The trick is to not rely on our gut, or our personal experiences, to decide what can help, but rather start from the science we have so far, add to it, and then find how technology can deliver, affordably and reliably, what will work better for learning.
What’s been the biggest surprise so far?
How slow education, and especially job training, has been to use what we already know about expertise, learning, motivation, identity, and stress to build out better learning environments. Corporations, with their supposed “dread focus” on value to shareholders, should be early adopters of ways to 1) identify groups of high-value, high-volume, high-variability job categories, 2) analyze who are top performers in groups, 3) deeply analyze what these top performers decide and do, and then 4) use evidence-based methods to train people in the job category to work the way top performers do. That hasn’t happened.
This same approach is available to higher education and high schools, though the economic-value case is more indirect. Still, we rarely do that detailed analysis of who are top performers, coupled with changes to what we then train everyone to be able to decide and do.
Especially now, as technology changes what “being good at your job” means for people, our ability to identify and train for this seems incredibly important — and we have not yet built these muscles into schools, universities, or job training.
Where do you see your work in five years?
In five years, I imagine continued work with people across many kinds of organizations — philanthropies, school innovation drivers, product development organizations, universities, and venture capital firms — who deeply care about empirically improving what learners decide and do. This work often starts from the developing sciences of learning, motivation, identity, and stress, and then considers whether emerging technologies might make previously inaccessible high-quality methods much more scalable.
High-quality one-on-one human tutoring is one example of the kind of evidence-based approach this work may increasingly seek to scale. It draws on a number of well-established learning and motivation science principles: starting from where an individual student already is; using what they already know and care about as an entry point into new and complex skills; providing efficient opportunities for practice and feedback; and building a personal relationship over time through continuity, trust, and shared experience.
More than one of these elements may be facilitated by new tools — especially AI. But the key is not to start from the technology itself. Rather, the starting point should be the levers that evidence suggests actually make a difference, and then identifying technologies that can make those levers more affordable, reliable, available, and data-rich over time. From there, the work is to continue iterating and improving — and to weave the technology together with educators and learners in the right way.
What else should people know?
Many technologies have been brought to bear on education and training over decades, all the way back to radio in the 1930’s. Yet, too often people assume that “this technology will do it,” without connecting to what we have learned in the last decades about learning, motivation, etc. It is as if we were developing new chemicals for medicine, and then just giving those to sick people to see if they worked. Even with a carefully constructed trial, we would conclude “chemistry doesn’t help sick people.” Of course, the key is understanding the science of the body, the disease, and the body’s reaction to disease, and then building on a hypothesis about how a specific chemical can usefully intervene somewhere in these processes.
Too often, we spam learners at any level with “technology that will be better than what you’ve had before,” with no connection to the sciences of learning, motivation, etc. We all need to get better informed about “learning engineering” to help us more efficiently diagnose problems, and then select from technologies (new, old, classroom, individual) that have a better chance of improving learning – and collect the right kind of evidence to continue to improve from there.
