AI tutoring narrows attention

A classroom can look busy and still miss a child. A new study from North Carolina State University found that when teachers use artificial intelligence tutoring software, they often return to the same small set of students again and again instead of checking in across the whole room. (news.ncsu.edu) That pattern showed up even though the software was built to help teachers spot who needs support. The systems in the study were “intelligent tutoring systems,” which are programs that watch what a student does on a math task, offer hints or feedback, and show teachers a live dashboard of who seems stuck. (news.ncsu.edu, ies.ed.gov) Think of the dashboard like a car panel with warning lights. If a student keeps entering wrong answers, the system can flag “struggle.” If a student stops working for a while, it can flag “idle.” Those signals are supposed to help a teacher decide where to walk next. (news.ncsu.edu) But classrooms are not traffic-control rooms, and teachers do not move by dashboard alone. In interviews, teachers said they decide whom to help using many cues, and two of the biggest were whether a student had needed help before and whether that student looked engaged or disengaged in the moment. (news.ncsu.edu) The researchers tested that against what teachers actually did, not just what they said. They analyzed 1,437,055 student-system interactions from 339 students in 14 middle and high school math classes across 10 United States schools during the 2022–23 school year. (news.ncsu.edu) The result was simple and a little unsettling. Teachers were more likely to interact with students they had already helped before, even after accounting for whether students were flagged as struggling or idle. In other words, earlier attention tended to attract more later attention. (news.ncsu.edu) That kind of loop can be hard to notice while teaching. A teacher may feel responsive because they are constantly helping someone, while quieter students who are not raising their hand, not tripping the right alert, or not already on the teacher’s mental list can fade into the background. That concern fits a broader federal warning that artificial intelligence tools in education need to be designed and used with fairness in mind, not just efficiency. (news.ncsu.edu, ed.gov) The original study was done in middle and high school math classes, not kindergarten through fifth grade classrooms. But the finding travels easily to younger grades, where teacher contact is often even more central because elementary students depend more on adult check-ins to stay academically focused, follow routines, and remain socially connected to the class. (news.ncsu.edu, news.ncsu.edu) That matters because the promise of artificial intelligence tutoring is personalization. If the software personalizes practice for each child but the adult attention around it gets concentrated on the same few children, the classroom can become more uneven, not less. The machine may be individualized while the human support becomes patterned. (ies.ed.gov, news.ncsu.edu) The study does not say teachers are being careless. It says the opposite: teachers are making fast decisions under real classroom limits, and those decisions appear to be shaped by habit, prior interactions, and the practical impossibility of giving every student one-on-one time. (news.ncsu.edu) So the fix is probably not “use less technology.” It is more likely to be small routines that spread teacher attention on purpose, such as rotating first check-ins, tracking who has not had a conversation yet, or redesigning dashboards so they show not only who is struggling now but also who has been overlooked this week. The North Carolina State team explicitly says the findings could inform tools that help teachers track classroom interactions more evenly. (news.ncsu.edu, files.eric.ed.gov) That is the real lesson in this study. Artificial intelligence can tell a teacher which students are clicking, pausing, guessing, or stalling. It cannot, by itself, make sure every child feels seen. (news.ncsu.edu, ed.gov)