Students Think AI Is Hurting Their Own Thinking

Sixty-seven percent of the students RAND surveyed already believe heavy AI use is dulling their own thinking, and the same research shows more of them reaching for it every semester anyway. A separate study published the same week found that a carefully built AI tutor could still nudge math scores up by a few points, just not on any problem harder than the one it had just practised. Same week, same technology, two findings pulling in opposite directions on what AI is actually doing to how kids learn.

The other two items this week move away from AI and closer to home. One describes what Ontario's youngest learners are working with before kindergarten even starts. The other is about whether 255,000 Ontario teachers and education workers will have a contract by the time most students go back on September 8. Four different subjects, one thread running under all of them: the gap between what a system is built to hand a child and what that specific child actually needs rarely shows up anywhere a parent can see it without asking.

Can an AI Tutor Actually Help Kids Learn Math?

The Hechinger Report published this one on August 17, 2026. Researchers ran a randomized trial with more than 6,000 Tennessee middle schoolers learning fractions, splitting them into four groups: conventional computer-based instruction or the same software enhanced with an AI tutor, and within each group, half the students also had to answer three practice questions correctly in a row before moving on, a technique called mastery learning. The lesson ran 50 minutes during regular math class, with retention checked a week later on a 15-minute test (Hechinger Report, 2026).

Students who got the AI tutor paired with mastery learning scored about 3 percentage points higher than those who got conventional instruction alone, a modest but real edge. What seemed to matter was how the AI tutor behaved: instead of just showing the correct answer, it walked students through their specific mistake, and those students spent more time per question, a sign of deeper engagement rather than guessing faster (Hechinger Report, 2026).

The honest caveat is in the fine print. The gains only applied to problems similar to what students had just practised. On harder, less familiar applications, the AI-tutored group did no better than anyone else. So this is evidence that a well-built AI tutor can sharpen practice on a narrow skill, not proof that it teaches transferable understanding.

Do Students Think AI Is Hurting Their Own Thinking?

RAND put this question directly to a nationally representative sample of 1,214 youth in December 2025, and the results ran in a wider commentary the organization published on August 17, 2026. The question was blunt: does using AI more for schoolwork harm a student's critical thinking skills? Sixty-seven percent agreed or strongly agreed, 27 percent of them strongly (RAND, 2026).

The breakdown by grade is where it gets interesting. Middle schoolers agreed at 68 percent, high schoolers at 64 percent, and college students at 69 percent, so skepticism runs high at every stage, not just among the youngest. But the sharpest split was by how much a student actually uses AI. Among students who use it for three or more classes, 53 percent still agreed it harms critical thinking. Among students who do not use it for schoolwork at all, that jumped to 78 percent (RAND, 2026). In plain terms, the people most worried about AI's effect on thinking are the ones opting out of it, and even the heaviest users are not convinced it is harmless.

This is a self-reported belief, not a measured change in how well anyone actually thinks, and RAND is careful to frame it that way. But a majority of teenagers holding this view while using the tool more anyway is exactly the kind of tension school boards are now being forced to write policy around, including in New York City, where a comprehensive classroom AI policy was expected before students return for a year starting September 10 (Chalkbeat, 2026).

What Pre-K Teachers Say About Materials for Kids Who Learn Differently

This item is older but still unpublished in a form most parents ever see: a fall 2024 RAND survey of public-school pre-K teachers, folded into the same August 17, 2026 commentary. Eighty-five percent of pre-K teachers said they mix and match multiple instructional materials in the same classroom, largely because no single curriculum covers the range of needs sitting in front of them (RAND, 2026).

The gap shows up clearly once you ask what teachers reach for to support specific kids. To support students with an IEP or IFSP, only 28 percent of teachers said they used their comprehensive core curriculum, while 56 percent said they used materials they created themselves. For English learners, it was 24 percent using the core curriculum against 37 percent relying on self-created materials (RAND, 2026). In both cases, the materials of unknown quality, self-made worksheets and activities the district never reviewed, were doing more of the differentiation work than the vetted curriculum was.

None of this means pre-K teachers are failing their students. It means the tools the system hands them were not built with every learner in the room in mind, so teachers are quietly filling the gap themselves. For a parent of a child with an IEP or an English learner heading into kindergarten, it is a reasonable question to ask a teacher directly: what materials do you actually use to support my child specifically, versus the class as a whole?

Where Ontario's Teacher Contract Talks Stand as School Starts

Collective agreements for Ontario's five education unions, ETFO, OECTA, OSSTF, AEFO, and CUPE-OSBCU, together representing more than 255,000 teachers and education workers, expired August 31, 2026 (ETFO, 2026). Most students return to class on September 8 without a finalized deal in place. The unions' stated priorities include stronger special education supports, smaller class sizes, mental health supports, and addressing understaffing, alongside wages (ETFO, 2026).

Bargaining has not reached any of that yet. The talks are stuck on a procedural question: whether certain issues get negotiated centrally with the province or locally with individual school boards. After four days at the central table, CUPE-OSBCU rejected a Ministry proposal to centralize items that have historically been settled locally and referred the scope question to the Ontario Labour Relations Board, a process that can take weeks to resolve before substantive talks on wages or class sizes can begin (CP24, 2026).

Strike votes are the next step. All of Ontario's major education unions are holding them in November: OSSTF members vote November 6 to 16, ETFO's central voting window runs November 9 to 18, and OECTA votes November 12 and 13 (CP24, 2026). A strike vote is not the same thing as a strike, and union leadership has been explicit about that. OECTA president Rene Jansen in de Wal noted that "a strike vote is not a strike," pointing out that roughly 98 percent of labour disputes settle without one (CP24, 2026). Ontario's education minister, Paul Calandra, has said he remains "very optimistic" about reaching agreements and sees "no reason for parents to plan for a strike."

Ontario has been through difficult bargaining rounds before, and they have not always ended quietly: in November 2022, a walkout by CUPE education workers closed schools across the province (CBC News, 2022). For now, the practical answer for families is that the school year is proceeding as scheduled, and this is worth watching rather than reacting to.

What This Means for Ontario Families and Teachers

These four items sit in different corners of a school system, a Tennessee classroom, a national panel of teenagers, thousands of pre-K rooms, and a bargaining table in Toronto, but they share a blind spot. Whether AI is actually helping or quietly undermining a specific child, whether a specific classroom has the right materials for how that child learns, and whether a specific support role survives this year's negotiations are all things a report card was never built to tell you. You mostly find out by asking, or by having someone else watching closely enough to notice first.

Connect Education works exclusively with licensed teachers. They work with your child directly, not a general classroom average, so they notice early when a skill is not sticking, when a tool like AI is being leaned on instead of understood, or when extra support is not landing the way it should. If anything in this week's research has you wondering whether your child's specific gap is being watched closely enough, you can book a free 15-minute assessment or browse certified tutors to get started.