Beyond the Answer Key: How AI Can Change the Way We Teach and Learn Math

Image courtesy of IEEE Spectrum

Artificial intelligence is already reshaping how students interact with mathematics. From apps like Photomath that provide instant solutions to chatbots like MathGPT that generate explanations on demand, today’s students have unprecedented access to help. For teachers, though, this raises an important question: Are these tools actually supporting learning, or are they short-circuiting the thinking process?

Much of the research and practice emerging over the last two years suggests that AI’s potential in math education lies less in efficiency and more in its ability to support productive struggle—the process of grappling with a problem, making mistakes, and refining strategies until understanding takes hold. Vicki Davis, a math teacher who has been experimenting with AI tools, argues that students “don’t need an answer; they need help with the process” (Davis, 2023). As Davis realized, early tools like Photomath or Wolfram Alpha are accurate, but they rarely show students where their reasoning broke down. In light of this, she advocates for pairing conversational AI with computational engines. ChatGPT, for example, can explain why a particular step is incorrect, while Wolfram Alpha ensures the math itself is right.

Generative AI can also support more ambitious instructional practices. Teachers can use tools like ChatGPT and SchoolAI to generate “rich tasks” and contextualized problems that connect math concepts to students’ real-world experiences, while also differentiating assignments for diverse learners (Koehler & Sammon, 2023). A teacher might ask for ratio problems set in the context of sports or music, or request open-ended geometry challenges that allow multiple solution strategies.

The emotional side of math learning deserves attention as well. Recent work in the npj Science of Learning argues that AI systems could play a role in reducing math anxiety—feelings of fear or tension when faced with math—by providing feedback that emphasizes effort and progress rather than just correctness (Gabriel et al., 2025). In practice, that could mean nudges like, “You set up the problem well, now check your square root again,” which reframes a mistake as a manageable step instead of a failure. For a teacher with thirty students in the room, it’s difficult to provide this kind of moment-by-moment reassurance, but AI has the potential to help scale that kind of support (Gabriel et al., 2025).

At the same time, we need to be realistic about limitations. The National Council of Teachers of Mathematics (NCTM) warns that AI tools often generate errors and can mislead students into thinking information does not need to be checked or cited (Klein, 2024). Similarly, the AI for Education coalition emphasizes that general-purpose chatbots like ChatGPT are less reliable for computation than specialized tools like Khanmigo or Desmos, and that in elementary classrooms, AI may be most useful to teachers for planning and preparation rather than direct student use (AI for Education, 2024). Teachers need to model skepticism and critical thinking, showing students how to check AI-generated content.

Equity is another important consideration. The Gates Foundation highlights how AI could expand access to high-quality math instruction, particularly for Black and Latino students and students in under-resourced schools (Gates Foundation, 2024). However, as experts from UNESCO pointed out, such tools must be designed with inclusion in mind—problems that are culturally relevant, data that is used responsibly and cleared of social and cultural bias, and teachers supported in how to integrate these systems effectively (Miao and Holmes, 2023). Otherwise, AI risks reinforcing existing inequities rather than reducing them.

I have also had the chance to try out MathVoyagers, an AI platform currently being developed by Dr. Jie Chao at the Concord Consortium, an EdTech innovation nonprofit based in Concord, MA. Unlike many general-purpose chatbots, MathVoyagers is deeply pedagogically grounded. It does not give away answers—even when asked directly—but instead poses a sequence of guiding questions that push students to explain their reasoning and reconsider their strategies. In my experience, it functions more like a thoughtful instructional partner than a shortcut, and it certainly does not replace the hard work of mathematical thinking. In this sense, it feels much closer to what teachers would want from an AI tool: something that facilitates learning without undermining it.

Looking ahead, I believe there is still much room to innovate in how AI is designed for math education. UNESCO has noted that generative AI has the potential to stimulate students’ higher-order thinking by acting as an opponent in Socratic dialogue (Miao and Holmes, 2023). Inspired by this, I see possibilities for AI in math classrooms that go beyond guiding students step-by-step through problem solving. For example, AI could deliberately generate flawed solutions that students are asked to critique, pushing them to exercise critical thinking and deepen their understanding. Another promising direction is to align AI with structured routines like Comparing and Discussing Multiple Strategies (CDMS), a teaching model developed by researchers at Harvard and Vanderbilt. In CDMS, students compare different methods for solving a problem and then discuss the reasoning behind them. AI could support this process by generating multiple solution paths—some efficient, some less so—that students must analyze, compare, and debate. This would situate AI not as a shortcut provider but as a catalyst for richer mathematical conversations. 

To realize these possibilities for AI in math education—whether in reducing math anxiety, broadening equity, enriching classroom discussions, or sparking creative problem-solving—will require genuine collaboration among teachers, learning scientists, policymakers, technologists, school psychologists, and others committed to learning. Just as calculators once forced educators to reconsider which skills mattered most, AI now challenges us to rethink “what is that critical content that’s necessary for students” and “what we teach [and] what can we de-emphasize,” as NCTM president Kevin Dykema explained in an interview (Klein, 2024). He further noted, “we need to continue to help our students see that math is used in real life…to solve real-world problems.”

For Boston teachers interested in exploring new AI platforms, MathVoyagers is currently available to try. Teachers can request an account and sign up here: https://math-voyagers-test.replit.app/login.