A Portland State University team will spend two years building tools to measure whether artificial intelligence can actually help students think through math problems, not just solve them.

The project, called "Seeing Through Data: Benchmarking Adaptive Feedback on Grades 6–12 Data Reasoning," is backed by a $2 million grant from the Gates Foundation and Learning Commons through the Digital Promise K-12 AI Infrastructure Program, according to PSU. That program is a $26 million multi-year initiative to improve how AI is built and used in K-12 classrooms.

The grant lands as Portland Public Schools works through its own reckoning with AI in the classroom. The PPS board unanimously paused new generative AI contracts in June 2026, as Willamette Week reported. The Nov. 4 PPS board meeting agenda included a GenAI Systems Inventory and Policy Update.

Three PSU professors lead the research. Eva Thanheiser, professor of mathematics education, heads the project. Ameeta Agrawal, associate professor of computer science, and Amanda Sugimoto, professor of education, round out the team.

Their central question: Can AI recognize what a student is thinking, suggest a useful next step and support learning without doing the work for the student or compromising privacy?

"The feedback is grounded in the student's thinking rather than in the correct answer," Thanheiser said, according to PSU.

How it works

The team plans to collect 1,200 pieces of student and teacher data from partnering school districts. All data will be stripped of student identity. Some will train an internal model on what good feedback looks like. Most will test whether AI systems have learned to deliver it.

The benchmarks will score AI models on specific questions. Did the system accurately read the student's reasoning? Did it pick the right instructional move? Did it preserve meaningful math work for the student instead of giving away the answer?

PSU offered an example. A student makes a bar graph showing School A has 80 suspensions and School B has 50, then claims School A has a "bigger discipline problem." A weak AI response would praise the graph and suggest adding labels. A stronger response would prompt the student to consider whether raw counts or suspension rates better support the claim, since the schools have different enrollment.

"AI is more than happy to just do the work for you, but we don't want to do that," Agrawal said, according to PSU. "We want AI to enable your thinking."

Beyond math class

The tasks focus on data visualization reasoning: interpreting data, deciding what to include, determining fair comparisons and evaluating what claims the data actually support.

Sugimoto is extending the work into English language arts, building tasks around infographics and data-based claims in informational text. Her component focuses on linguistically and culturally diverse students, testing whether AI can recognize sound reasoning even when it doesn't arrive in polished English.

Agrawal said the benchmarks could eventually give school districts an independent way to evaluate AI products. "For a school or district wanting to adopt an AI model in the future, this project will provide an independent way to examine specific capabilities and limitations that matter when AI systems respond to student reasoning," she said, according to PSU.

The PPS board's Sept. 22 Policy Committee meeting included a preliminary discussion on developing a formal AI policy, according to committee minutes. Director Julia Englesman, the committee chair, asked staff to prepare early policy ideas. Committee members noted that some high school educators have restricted computer note-taking because of AI use and that student feedback on AI has been "overwhelmingly negative."

The PSU benchmarks, once completed, would be openly available for any district to use.