Blog

by Joshua A. Taton, Ph.D. | November 12, 2023 | 3 min read

There is a lot of discussion, right now, on how generative AI will reshape education. As with any tool, generative AI offers promise and peril. Here, I explore why I think generative AI is a long way off from helping students develop conceptual understanding.

Let’s break down why generative AI tools, generally, can’t help with developing students' conceptual understanding in mathematics:

• They can’t ask students questions that will authentically assess students’ prior level of understanding

• They can’t give students a mathematical scenario and ask students how they might solve it (and understand the response)

• They can’t respond to students’ problem-solving attempts in ways that maintain productive struggle and provide scaffolding 

• They can’t guide students along the pathway from invented solutions to a connected awareness of more sophisticated ideas or approaches

• They can’t flexibly adjust problem scenarios in response to students’ language-learning needs, interests, and range of understanding

• They can’t assess the depth of understanding students exhibit in several areas simultaneously.

Why not? Because students use invented strategies, invented language, incomplete explanations or unexpected descriptions. And they have widely diverse learning needs.

Anyone who has taught successfully for a long time knows this. And has learned to speak and interpret “kid math🩷.”

No, there’s a lot of evidence that multiple-choice questions simply can’t "get at" conceptual understanding. Further, multiple-choice tests can’t cover all of the cross-standard content, adequately enough to assess full understanding. Finally, as we know well, multiple-choice tests exhibit lots of statistical, linguistic, and often hidden cultural bias that cannot be fully mitigated.

Can generative AI do some of these things, at least in part? Sure! I'm seeing compelling evidence of progress.

While I think generative AI can be a helpful thinking tool, especially for creating rough drafts, for organizing thoughts and guiding further digging, and for light, skills-based tutoring.

But I think we are still a long way off from generative AI being able to help with deepening students conceptual understanding, especially not in engaging and thought-provoking ways. Human-computer interfaces are still clunky with inputting and interpreting handwritten text. Digital manipulatives and tools are a pale approximation of physical ones.

Regardless, generative AI models and ed tech tools, broadly, would need the heavy involvement of math educators, teachers and math-cognition experts. Especially in roles as design leads. 

And I’m certainly not seeing that happen yet—at least not regularly enough. 

One last comment: As a result of the lack of involvement of educators and math-cognition experts in the design of ed tech tools, in my view, such tools still depend on old, narrow, and unhelpful paradigms about math, learning, instruction, and assessment. Generally, these paradigms focus on replication of skills (without understanding) and regurgitation of skill-based understanding via multiple-choice. 

This also needs to change, if we want such tools to be demonstrably more effective than they are now. Personally and professionally, then, I recommend we continue to support teachers and strong, research-based, and student-centered instruction in schools.

Need Help?

I am ready, willing, and able to help your school or system adopt research-based changes to your curriculum, instruction, and assessment practices. If you want to see deep and meaningful change, including renewed excitement about mathematics instruction or learning in your building, please contact me for a consultation.