How to update your teaching in a world with AI.

Technology advancements and greater access to information online have posed challenges for how teachers educate students for years. Depending on the content you teach and the educational environment there are different options for how to incorporate or react to the existence of new AI Large Language Model tools like ChatGPT. At that point, it makes sense to take a step back and explore what options you might have to address your current dilemma.

Complex problems are often vague, and that means they have many possible solutions. Following the Meta-Problem Method may lead you to a distant dilemma from the one that started your quest. A key part of the method's value is that it forces you to clarify what you really want and what you are willing to give up. It also enables you to compare objectively the possible pathways and their trade-offs. It prevents you locking into solutions mode too early, and then doubling down on solving a low-yield problem that does not serve your goals as well as the alternatives. At the end of this process, you will have a better understanding of your priorities and how to achieve them.

Step in the Meta-Problem Method	Illustrative Example
Dilemma The high-level issue you are trying to address	How to update your teaching in a world with AI.
Goal The changes you would like to make to address the dilemma. There are usually many options. Selecting the best set comes after you learn what is possible.	Supporting Goals Students learn valuable skills. Students learn the course content. Avoid students choosing to cheat. Less work needed per student.
Problem Space While goals tell us what we want, our next step is to understand what is holding us back from making progress on them. This approach is borrowed from calculus as we explore the neighborhood of the current dilemma.	For each goal that you are considering, ask yourself: How much progress is possible? How much effort would it take to make progress? What methods might help to make progress? What might the positive or negative effects be on the other goals as you make progress towards the current one? Example Problems: How can we teach students to use AI as a useful tool? How can they recognize and develop other skills when AI is not the right tool? For example, in order to ensure students still learn how to write for themselves before learning to be an editor of AI-generated content, some educators are returning to hand-written assignments. How can we teach the foundational course content when students might use AI to bypass thinking? How will we ensure they understand when they are missing the nuance in the higher-level material if they missed the foundational knowledge? Is the current course material something these tools do well, or is it something where they do poorly? For example, a student who submits an incorrect AI generated answer may have not tried to check their work, or they may have genuinely missed the differences. Depending on the reason for their error they may need different lessons. How do we make it hard for students to cheat without being caught? Should we assign easier homework so it is less tempting to cheat? Assignments where it is easier to detect cheating? Technology-free assessments to validate understanding? For example, since most knowledge builds we need to provide both easy questions (to develop competence) and harder questions (to develop excellence). The easy questions may be topics AI can handle, but harder questions should provide the reinforcement to learn for themselves. How do we manage curriculum development and grading effort? What are the minimum updates that need to be done? What might change again soon? For example, there are many subjects which we already have plenty of best practices documented which could be used to create custom AI-generated lesson plans. But educators need to check their work just like students to ensure the right messages are being shared. Since using AI is a tempting shortcut, how can we teach students the weaknesses and strengths? For example, including a unit in middle school or a workshop for first-year college students emphasizing the need to learn how to learn.
High-Yield Problems Sometimes solving one problem helps make progress towards several goals. In this step, we identify these “two-for-the-price-of-one” problems.	Which Options Will Advance More Than One Goal? Design course material and assignments to trip up AI tools. If the topics support this approach, it will help students learn valuable skills and the course material while making it harder to cheat, though it may take more effort. Design some assessments to be technology-free. This method may increase the effort per student, but can ensure students are evaluated for their knowledge and practice those skills. Teach a focused lesson on the limitations and strengths of AI. This may lead some students to have more confidence in their ability to cheat, but would also help them understand when not to use AI. Design assignments to depend on setting the problem up instead of providing a solution. This approach is typically easier to manually-grade and tests reasoning instead of computation. Et cetera.
Problem Selection Which of the many possible options in the high-yield problem step is the best set to address the dilemma?	Selection Criteria Which solutions will best address the dilemma? Which solutions will deliver the best outcome for the least amount of time, effort and money? Which solutions is the student most excited to take on?  By this point in the Meta-Problem Method, you have clarified your goals, identified some options you could take, weighed the trade-offs that come with each of those options, rejected some options because they will take more time, effort or money than the results are worth, and you have identified a set of high-yield problems that will advance several of your goals at once. Now you are ready to start solving a problem knowing what you expect to achieve.
Implement, Learn and Adapt	Observe and learn as you go. New information may reveal itself as you implement your chosen solution, so check continuously that you’re still solving the right problem.