This commentary uses research on AI-scaffolded learning as a springboard for broader reflection on education. The work of Chien, Chan, and Hou (2025) sparked these reflections through their role-playing educational game using generative AI for instant feedback.
The Instant Answer Paradox
We once stood before knowledge and contemplated; now knowledge stands ready to answer before we’ve finished asking.
A question stays with me: If AI can answer questions instantly, what purpose do memorization, research skills, or expertise serve? The findings from Chien and colleagues offer something unexpected. Participants experienced “significantly high flow and low activity anxiety” while learning basic science through AI-scaffolded gameplay. Rather than disengaging when AI provided immediate feedback, students leaned in. This response deserves attention.
This observation challenges assumptions I’ve long held about learning. We’ve believed struggle builds capability. Perhaps we’ve been conflating two distinct forms of struggle. Productive struggle involves grappling with concepts, testing hypotheses, integrating knowledge. Logistical struggle involves waiting for feedback, searching for resources. AI eliminates the latter while potentially intensifying the former. I should note that the game itself focused on teaching AI prompting skills, which somewhat confounds these findings with the learning outcomes.
From Acquiring Knowledge to Building Capability
What struck me about this research was how the generative AI provided “a high percentage of direct answers or indirect suggestions,” creating suitable interactive scaffolding. The study used ChatPDF constrained to specific content, achieving 78-90% helpful response rates. This wasn’t simply automation replacing human instruction. It represented a responsive learning environment adapting to each student’s inquiry in real time.
Education has centered on knowledge acquisition for the past century. We’ve treated students as vessels for facts, theories, and procedures. Success meant retaining and retrieving information on demand. When AI can retrieve facts readily, this model reveals its limitations. Not because knowledge loses importance, but because the fundamental constraint shifts. Access to information no longer limits us. The capability to use information well becomes the bottleneck that matters.
From this research emerges insight into what I’ll call capabilities. Participants weren’t memorizing isolated science facts. They engaged in contextualized problem-solving within narrative frameworks. They formulated questions, evaluated AI responses, made decisions under uncertainty. These represent capabilities, not mere retention.
Three Capabilities That Matter More
Using this research as a starting point, I see three capability domains emerging as increasingly valuable.
Question formulation. When AI provided “direct answers or indirect suggestions,” learning quality hinged entirely on question quality. Knowing what to ask, recognizing when different questions serve better, this becomes foundational. It demands domain awareness, what I think of as curiosity architecture, and epistemic humility. These capabilities can be taught.
Synthesis across contexts. Players applied basic science knowledge to solve puzzles within a fictional cave system. This transfer, moving knowledge from one context into another, remains distinctly human territory. AI generates solutions within defined parameters. Identifying which parameters matter in novel situations requires judgment shaped by varied experience across domains.
Metacognitive awareness. The research measured flow and anxiety during gameplay. Students maintaining high flow weren’t passively absorbing information. They monitored their engagement, regulated learning strategies, maintained productive challenge. This self-awareness grows critical when AI offers scaffolding. Without it, dependency on external structure replaces internal capability.
Implications for Educational Practice
This research raises important questions for those shaping learning experiences. I’ve been considering what this means for practice:
-
Reimagine assessment priorities. When AI handles recall effortlessly, assessments might focus on application in novel contexts, examining whether students can transfer understanding across domains.
-
Develop question literacy. Question formulation could become an explicit skill with deliberate practice and progression, much like we approach writing or quantitative reasoning today.
-
Distinguish productive from logistical struggle. AI might remove barriers to resources and feedback, while we simultaneously increase cognitive challenge through problems requiring genuine synthesis and creative integration.
-
Build metacognitive reflection. Learning routines might incorporate regular reflection, asking not only what students learned but how they learned it, what strategies proved effective.
-
Prioritize depth over breadth. When information becomes readily available, developing one capability deeply may serve students better than covering ten topics superficially.
What This Reveals About Learning
This study demonstrates that generative AI can scaffold learning while sustaining engagement within a specific context. I’ve used it here as a springboard for broader speculation about education’s future. The research itself examined a contextualized game teaching basic science to college students; my reflections extend beyond what the data alone can support.
When answers arrive readily, the most valuable capability isn’t knowing facts. It’s understanding what to do with accessible knowledge. Rather than diminishing learning, this shift represents its elevation to higher purpose.
References
Chien, C. C., Chan, H. Y., & Hou, H. T. (2025). Learning by playing with generative AI: Design and evaluation of a role-playing educational game with generative AI as scaffolding for instant feedback interaction. Journal of Research on Technology in Education, 57(4), 894-913. https://doi.org/10.1080/15391523.2024.2338085
