Can AI Tools Transform Low-Demand Math Tasks? An Evaluation of Task Modification Capabilities

2026-04-14 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, AI in Education · Depth: Expert, quick

Summary

A recent study evaluated the capacity of eleven AI tools, including general-purpose models like ChatGPT and Claude, and specialized tools such as Khanmigo and coteach.ai, to upgrade low-cognitive-demand mathematics tasks. Researchers prompted these tools to modify two types of tasks, using a strategy reflecting typical teacher approaches. On average, AI tools successfully upgraded tasks only 64% of the time, with individual tool performance varying from 33% to 88%. Specialized tools showed only a moderate advantage over general-purpose tools. Common failure modes included "undershooting" by maintaining low demand and "overshooting" by creating overly ambitious tasks. The study also found a small negative correlation (r = -.35) between an AI's ability to classify task demand and its ability to modify tasks, indicating distinct capabilities.

Key takeaway

For mathematics teachers considering AI tools for curriculum adaptation, understand that current AI capabilities for upgrading low-demand tasks are inconsistent, averaging 64% success. You should anticipate potential "undershooting" or "overshooting" in task complexity and plan to review and refine AI-generated modifications carefully, as specialized tools offer only a slight edge.

Key insights

AI tools show moderate success in upgrading low-demand math tasks, with generative modification distinct from classification.

Principles

• Generative AI task modification differs from classification.
• Specialized AI tools offer only moderate advantage over general-purpose.

Method

AI tools were prompted to modify low-cognitive-demand math tasks, with performance evaluated using the Task Analysis Guide framework to assess upgrade success and identify failure modes.

In practice

• AI tools can assist in curriculum adaptation.
• Expect varied success rates (33%-88%) from AI task modification.

Topics

• AI Tools
• Mathematics Education
• Task Modification
• Cognitive Demand
• Curriculum Adaptation

Best for: AI Scientist, Research Scientist, Domain Expert

Related on AIssential

• Measuring the impact of learning with AI in Sierra Leone and beyond — AI, when designed for "Socratic" scaffolding, significantly improves student math learning and engagement without replacing teachers.
• How AI Helped Me Turn a Long Family Drive Into a Full-Blown Music Game — Accessible AI tools empower non-coders to rapidly develop complex digital applications from novel ideas.
• Examining the Usage of Generative AI Models in Student Learning Activities for Software Programming — GenAI significantly improves programming task performance but often impedes knowledge gains, particularly for beginners who over-rely on it.
• Kathleen deLaski on reimagining higher education, generational mobility, building AI skills, and human originality (AC Ep43) — Higher education must adapt to AI and "skillification" by emphasizing human judgment and originality alongside technical competence.
• AI is inflating student grades, and the effect points to outsourced work, not better learning — AI-driven grade inflation in writing and coding assignments indicates outsourced work, not improved student learning, eroding grade validity.
• How AI is changing the nature of mathematical research — AI tools are revolutionizing mathematical research by automating proof generation and contributing novel ideas, accelerating discovery.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.