I went to the Ai4 conference expecting to think about artificial intelligence through the lenses I know best: learning, design, and human development. Across sessions, I kept hearing versions of the questions learning scientists already ask: What should humans keep doing when AI can do more? How do we know an AI system is working? What happens when we outsource too much thinking? Who’s accountable when it fails, and who benefits when productivity rises?
But these questions were often framed through the lenses of technology, business, national competitiveness, healthcare, finance, or existential risk. An informal hierarchy emerged where education and learning science were not always central.
Engineers talked about observability and testing. Business leaders talked about redesigning workflows. Psychologists talked about sycophancy and human connection. Economists and technologists talked about labor, capital, regulation, and corporate incentives.
When I listened closely enough, though, a pattern emerged.
The biggest questions about AI often contain learning problems
Even when the biggest questions about AI get filtered through the lenses of other industries, they often contain a substantial learning problem.
If workers need to transition into new roles as jobs and tasks change, that is a learning problem.
If professionals need enough AI literacy to evaluate unreliable outputs, that is a learning problem.
If organizations need to redesign workflows and help employees develop new practices, that is a learning problem.
If citizens need to distinguish AI hype from proven evidence, understand economic incentives, and participate meaningfully in debates about governance, that is a learning problem.
Learning science can help organizations rethink how they teach new technologies
When employees need to learn a new technology or workflow, organizations might schedule webinars, demonstrate prompts, distribute guides, and hope that adoption follows.
But there are better ways to teach new workflows.
One alternative from an Ai4 session rebranded the familiar “Lunch and Learn” session as a “Bite and Build.” One is passive; the other asks people to learn AI while doing authentic work.
As a learning scientist, I would push this reframing even further.
Even before designing AI training, audit the current work. Find the people already using AI effectively. Study what they are doing. Identify where they are saving time, where quality is improving, where they are struggling, and what tacit knowledge makes their workflows successful. Then redesign the workflow and train people to perform that workflow.
Instead of:
Choose tool → train people → hope work changes,
we can ask ourselves to:
Understand work → redesign workflow → determine AI's role → develop human capability
Again and again, I have found that the quality of an AI-generated learning experience depends on the quality of the context we give it—not merely the sophistication of the model.
If we redesigned our workflows knowing what AI can do, would we design them the way they exist today? What would we do differently?
Learning and development have an opportunity here. We can become the department that teaches people how to operate the newest tools—or we can help organizations rethink how humans learn and work alongside increasingly capable systems.
Those are very different ambitions.
Learning scientists know what makes feedback work
At this year's conference, Dr. Joseph Lee described a pattern he is seeing in AI and mental health: sycophancy. AI systems tend to affirm users' ideas, validate their beliefs, and reinforce self-perception.
But decades of research on feedback and motivation (e.g., Carol Dweck’s growth mindset research, Edward Deci & Richard Ryan’s Self-Determination Theory, Kim Scott’s Radical Candor framework) point to the same conclusion: Growth requires feedback supportive enough to keep someone motivated and honest enough to help them improve. Too much friction, and people give up. Too little, and they stop growing.
Researchers like Manu Kapur suggest we aim for what he calls “productive struggle” or what Elizabeth and Robert Bjork call “desirable difficulties,” a challenge that builds capability rather than one that makes someone feel good in the moment. A mentor that never pushes back does not produce a better student. A writing assistant that calls every draft excellent does not produce a better writer. The risk lies in people trusting a machine’s freely given approval more than their own judgement.
This is another place learning science has real guidance to offer the teams building these systems. Learning scientists know what makes feedback work, and we can apply that to design in order to better support learning.
Learning scientists belong in early build discussions
The learning problems contained in some of AI’s most pressing questions can benefit from learning scientists, educators, designers, psychologists, and policymakers’ early input. Learning science in particular affirms the need for critical human thought, for questioning, creating, and exercising agency while increasingly powerful systems surround them.
That may be the learning problem of our time.
Perhaps education should not be waiting downstream for technologists to build the future and then asking how we train people to live in it.
Learning science adds nuance to discussions about the future of AI
I left Ai4 more optimistic about what AI can make possible and more skeptical of simple stories about what happens next. When I say simple stories, I mean both the techno-utopianism that asks us to believe that greater technological capability will naturally create better human outcomes, and the catastrophic rhetoric that can make a dystopian future feel more immediate than the material harms already going on in the world.
Neither position is sufficient. Taking AI risks seriously is not fearmongering. Neither is asking who benefits from AI anti-innovation.
Critical AI literacy requires being able to hold several truths at once. This kind of complex, nuanced approach can benefit from the principles of learning science.
AI can help us work faster. It can expand access to individualized learning. It can process volumes of information that humans could never reasonably examine alone. It can help us find patterns, monitor changing conditions, identify gaps, and create things that previously required far more time and specialized expertise.
But capability is not the same thing as progress.
For learning science, perhaps our contribution to the AI era lies in insisting that the measure of intelligent technology is not simply what the technology can do, but what people are able to learn, become, and do because of it.
Want to support or partner on this work? Contact us at info@wgulabs.org.
