Two guests on “This Week with EdSurge” explain how AI can be used to assess when students think through their answers and when they talk with chatbots.
Can AI Give Teachers Feedback on What They Notice? Molly Fisher of Rowan University studies professional noticing: attending to what a student is doing, interpreting what it shows about their understanding, and deciding what to ask next. Her earlier research found that teachers can learn to notice, but the review comes slowly, usually only after a colleague observes a class or someone watches a recording. Starting in spring 2027, her team will test whether AI tools can speed that up. The team will score teachers' written responses to classroom video first, then use that scoring to train the models, with a goal of sending each teacher a report within a day. Fisher calls it a fact-finding mission. She urges teachers to stay skeptical of AI output and to be careful about the student information they upload. Should Teachers Grade the Chat? Mike Kentz, who runs AI Friction Labs and has taught middle school, high school and college, starts with a study in which middle schoolers were offered an AI math tutor and most rarely used it. Kentz says a tutor that asks questions is harder than closing the tab or asking a chatbot for the answer, and that requiring it does not help if students never learn what good engagement looks like. His fix is to grade the chat itself. He caps conversations at around ten messages to keep grading manageable, and builds criteria with students by comparing a strong chat with a poor one. He also asks students to reflect before and after, and has them explain a concept like tangent lines to a confused AI classmate. Kentz compares this moment to the early days of composition studies, when no one yet agreed on how to evaluate an essay.
Can AI Give Teachers Feedback on What They Notice?
Molly Fisher of Rowan University studies professional noticing: attending to what a student is doing, interpreting what it shows about their understanding, and deciding what to ask next. Her earlier research found that teachers can learn to notice, but the review comes slowly, usually only after a colleague observes a class or someone watches a recording. Starting in spring 2027, her team will test whether AI tools can speed that up. The team will score teachers' written responses to classroom video first, then use that scoring to train the models, with a goal of sending each teacher a report within a day. Fisher calls it a fact-finding mission. She urges teachers to stay skeptical of AI output and to be careful about the student information they upload.
Should Teachers Grade the Chat?
Mike Kentz, who runs AI Friction Labs and has taught middle school, high school and college, starts with a study in which middle schoolers were offered an AI math tutor and most rarely used it. Kentz says a tutor that asks questions is harder than closing the tab or asking a chatbot for the answer, and that requiring it does not help if students never learn what good engagement looks like. His fix is to grade the chat itself. He caps conversations at around ten messages to keep grading manageable, and builds criteria with students by comparing a strong chat with a poor one. He also asks students to reflect before and after, and has them explain a concept like tangent lines to a confused AI classmate. Kentz compares this moment to the early days of composition studies, when no one yet agreed on how to evaluate an essay.