AI Interaction is Problem Solving, Which Schools Don’t Teach
Many of the AI literacy teachings say working with AI is a matter of telling it precisely what you want, what your context is, and vetting whether the output is correct. The implicit message is that AI is something you ask a question of.
But those aren’t the most important AI skills, especially now that the AI can be set up to drag more information from you and via agents to do things on its own. The important AI skills relate to problem-solving and its associated judgments.
Consider what actually happens when you tackle a meaningful challenge with AI. You start by deciding whether AI can even help with your specific problem. You choose which AI tool might work best. You figure out how to frame the challenge in a way the AI can understand. When the first attempt doesn't work, you debug the interaction—maybe the AI misunderstood your context, or you need to break the problem into smaller pieces. You iterate through multiple approaches, sometimes discovering that you were asking the wrong question entirely. Eventually, you might realize AI isn't the right tool for this particular challenge and pivot to a different strategy altogether.
Every step involves navigating multiple viable approaches, weighing tradeoffs between competing objectives, and iterating toward better solutions. Working with AI is problem solving, not prompt literacy.
Yet most students have never learned to think this way. Most teachers haven’t either. Whereas most work worlds are all about problem solving, education focuses on knowledge and strips away complexity so that problems focus on individual knowledge silos. People have spent years being taught to optimizing for single correct answers, not construct workable solutions from competing constraints and values. It is practically a playbook for creating binary thinkers. (I speculate that the lack of problem solving in the curriculum also translates to lack of problem solving prowess across schools and the entire system. The talent pool is more skilled at expressing issues than solving them.)
Consequently, many people treat AI like a sophisticated search engine rather than recognizing it as a collaborative problem-solving exercise.
The problem solving I'm describing has three essential characteristics that distinguish it from typical school work:
Multiple viable solutions with tradeoffs. Real problems rarely have a single correct answer. Instead, they present competing objectives—budget versus quality, speed versus thoroughness, individual needs versus group benefits. Students need practice weighing these tradeoffs, understanding how the balance can change by situation, and defending their choices.
Solution demonstration, not just expression of ideas. It's not enough to have good ideas; you need to turn them into actionable plans. This means considering implementation challenges, resource constraints, and unintended consequences. Students should produce something that could actually work in the real world, appropriate to their age of course, complete with contingency plans and success metrics.
Iterative refinement and reframing. The best solutions emerge through cycles of testing, feedback, and revision. Often, this process reveals that the original problem was framed incorrectly or that sub-problems need to be solved first. Students need experience with this messy, non-linear process of discovery and refinement.
These characteristics define both effective AI collaboration and professional success in an increasingly complex world. But they're largely absent from traditional schooling.
How Little Problem Solving Actually Happens in School
To understand the scope of this gap, I recently had AI analyze typical K-12 curriculum time allocation across core subjects. The methodology involved examining district-mandated instructional minutes, observational research on classroom practices, curriculum standards documents, and typical textbook organization to estimate how much classroom time is spent on different types of activities.
The results were stark. By Gemini 2.5’s analysis, approximately 75-85% of classroom time in the U.S. is devoted to knowledge acquisition and skill practice—learning facts, procedures, and established methods. Another 10-15% goes to knowledge application—using learned information to answer questions or solve problems with known solutions.
Genuine problem solving—the kind involving multiple viable approaches, real constraints, and iterative development—accounts for roughly 5-10% of total instructional time. I don't claim these numbers are precise—they could be off by several-fold—but I wanted a ballpark estimate. Even if the true figure for problem solving were 20% instead of 5%, we'd still have a massive mismatch with the world. Even in subjects that seem naturally suited to problem solving, like science and social studies, the majority of time is spent learning about solutions others have already discovered rather than developing the capacity to construct novel solutions.
Walk into most classrooms, and you'll find plenty of thinking, but precious little problem solving. Students analyze literature, solve math equations, and conduct science experiments—all valuable activities that require sophisticated reasoning. But the problems are pre-solved. The methods are established. The criteria for success are clear and singular.
Even in subjects that seem naturally suited to problem solving, we've often sanitized the complexity away. History becomes a sequence of events to memorize rather than a laboratory for understanding how complex systems change over time. Science becomes the application of known formulas rather than the messy process of investigating unknown phenomena. English becomes the analysis of existing texts rather than the creation of new arguments for real audiences.
The result is students who can demonstrate impressive analytical skills within bounded contexts but struggle when faced with open-ended challenges. They can write essays about predetermined topics but freeze when asked to identify and address a real community problem. They can solve calculus problems but struggle to optimize their own study schedules.
Why This Gap Matters More in the AI Era
For decades, the implicit promise of education has been straightforward: master this body of knowledge, and you'll be prepared for adult life. Learn algebra, memorize historical dates, understand cellular biology, and you'll have the foundation you need for college and career.
That model made sense when information was scarce and jobs were relatively stable. If you could demonstrate mastery of established knowledge, employers could reasonably predict your ability to handle similar challenges in the workplace.
But AI has fundamentally disrupted this equation. When any student can access expert-level knowledge on virtually any topic through a simple conversation, the competitive advantage of knowledge acquisition plummets. That had already been happening with Internet information availability, but with AI the knowledge acquisition friction is practically gone. What matters now is the ability to navigate ambiguity, synthesize information from multiple sources, and construct novel solutions to unprecedented problems.
AI amplifies the consequences of this gap. A student with strong problem-solving foundations can quickly learn to leverage AI as a powerful collaborator. They understand how to frame problems, evaluate trade-offs, and iterate toward better solutions, so they naturally apply these skills to AI interaction.
But students without these foundations don't just fail to use AI effectively—they become vulnerable to its influence. They accept AI's first response without considering alternatives. They can't recognize when AI is optimizing for the wrong objectives. They struggle to maintain creative ownership of their work because they've never learned to direct collaborative processes.
The students who thrive with AI are those who approach it like a project manager directing a talented but literal-minded team member. They can break complex goals into manageable sub-problems, provide contextual guidance, and evaluate outputs against multiple criteria. These are precisely the problem-solving skills that traditional schooling often neglects.
These aren't knowledge domains you can master through study—they're capabilities you develop through practice with real challenges that have real consequences.
That’s not a new message but one that underpins the entire branch of constructivist learning theory. But learning through perception and doing, as emphasized by experiential learning advocates, is often watered down to group work or “hands-on” projects that are more assembly or recipe than design or creation. To foster problem solving skills, the end should not be known.
Reimagining How Learning Happens
The solution requires more than adding problem-solving projects to existing curricula. We need fundamentally different approaches to how learning happens.
Authentic project-based learning offers one paradigm shift. Instead of studying historical conflicts and then writing about them, students could design peace-building strategies for current disputes, weighing humanitarian concerns against political realities. Rather than solving predetermined math problems, they could optimize resource allocation for school events, balancing budget constraints with student preferences. The key is ensuring these projects involve genuine constraints, multiple viable approaches, and iterative refinement—not just elaborate ways to demonstrate predetermined knowledge. This is a critical distinction, as many project-based initiatives fall short by offering only a narrow slice of a problem designed to fit a single subject, ignoring the messy, cross-disciplinary nature of real challenges.
But even project-based learning often maintains the traditional sequence: learn the knowledge first, then apply it. What we really need is upside-down learning—flipping the entire paradigm so that complex challenges come first, and knowledge acquisition follows as needed to solve real problems. If those challenges are well designed, much of the existing curriculum still gets covered, but with the more powerful learning experience of knowledge being discovered rather than accumulated.
This concept, which I explored in depth in Wisdom Factories, challenges the fundamental assumptions of how learning should be sequenced. This doesn’t mean throwing students into the deep end without support. Of course, students need prior knowledge. But the mindset of problem-solving can be taught and practiced using whatever existing knowledge students already possess—starting with personal, relatable challenges before expanding to new school topics. It reframes foundational knowledge not as a prerequisite barrier, but as a tool to be acquired on-demand when the need arises.
Think about how professionals actually work. The problem comes first. When an engineer needs to design a bridge, they don't first master every aspect of materials science, then every principle of structural dynamics, then every regulation and environmental factor. They start with the specific challenge—this river, this traffic load, this budget—and pull in knowledge as needed to address each aspect of the solution.
In upside-down schooling, students would start with a big, meaningful challenge—reducing energy consumption in their school, designing a more effective student government, investigating local water quality issues. They'd need to break down these challenges into manageable pieces, figure out what they need to learn to make progress, and seek out authoritative sources of information. Knowledge becomes purpose-driven and on-demand, enabling problem solvers to chip away at solutions rather than stockpiling information they might someday use.
This approach naturally develops the three characteristics of problem solving we desperately need. Students encounter multiple viable approaches because real challenges don't have textbook answers. They must demonstrate solutions that could actually work, not just show they understood the reading. And they iterate constantly as they discover new constraints, gather feedback, and refine their approaches.
Most importantly, upside-down learning develops student agency—the capacity to direct their own learning toward meaningful goals. Rather than dutifully absorbing prescribed content, students learn to identify what they need to know and figure out how to learn it. These are exactly the skills required for effective AI collaboration, where success depends on your ability to direct an intelligent system toward useful outcomes.
There is urgency. Students are forming their AI habits right now, with or without educational guidance. Those who learn to approach AI as a problem-solving partner will have enormous advantages over those who treat it as a fancy search engine. But developing these capabilities requires extensive practice with authentic challenges that start with problems, not predetermined knowledge.
Schools and teachers shouldn’t wait for curriculum committees to catch up or for assessment systems to evolve. Individual teachers can begin incorporating genuine problem solving into their classes immediately. Schools can pilot cross-curricular projects that require students to navigate real constraints and iterate toward workable solutions.
The alternative is sending another generation into a world that requires problem-solving skills they've never been taught to develop.
©2025 Dasey Consulting LLC
Great article! Any ideas about how your suggestions could be scaled? My undergrad classes typically have 80-90 students. It feels like AI could help with the scaling, but I’d have to give the idea more thought. Thanks!