Useful AI homework design makes the intended thinking easier to see and discuss. It also leaves teachers room to learn from what happens in their own classrooms.
Retrospective coverage of November 4, 2025. Product details and offers describe that period and may have changed. The practical exercises below are StudyScope’s suggestions.
An education guide that leaves room for uncertainty
On November 3, 2025, MIT News described a guidebook from MIT's Teaching Systems Lab called A Guide to AI in Schools: Perspectives for the Perplexed. The guide draws on input from more than 100 students and teachers from around the United States, who shared their experiences teaching and learning with generative AI tools. The article quotes an MIT associate professor who hopes the guide can support educators as they decide what AI policies or guidelines to write. He cautions that it is not meant to be prescriptive or definitive, and that clear, definitive answers on effective AI use in classrooms do not currently exist. He also asks a question that is useful when you review homework: if students no longer do the exercises that carry the content, what are they learning?
Start with a familiar assignment rather than rewriting a whole course. Pick the exact mental action it should practise: selecting evidence, explaining a relationship, comparing methods or interpreting a result. A polished submission may hide whether that action happened. The assignment review below is an original planning exercise, not a tested intervention reported by MIT.
Source: MIT News · Helping K-12 schools navigate the complex world of AI
Source: MIT Teaching Systems Lab · A Guide to AI in Schools: Perspectives for the Perplexed
Find the decision inside the task
Take a geography assignment that asks students to explain why two neighbourhoods have different temperatures. A generic paragraph could be put together without examining the supplied map. Instead, ask students to choose two observations from that map, explain what each one supports and name one piece of information that is still missing. That puts a visible decision between the source material and the final explanation.
Keep the extra work in proportion. A diary of every click can become an administrative burden. It may reward documentation more than understanding. One annotated map and a short explanation of a rejected interpretation can show more relevant thinking. Tell students which evidence matters before they begin, and give an example of the expected depth. Teachers interviewed for the MIT guide described options for discouraging unauthorized AI use, such as designing engaging assignments, using more in-class work, adding an oral assessment to major papers and giving more frequent, lower-stakes assignments. These are options teachers described, not proven methods.
Source: MIT Teaching Systems Lab · A Guide to AI in Schools: Perspectives for the Perplexed
State where assistance fits
Write the assignment's AI rules next to its learning purpose: if brainstorming is allowed but evidence selection must be done independently, say so directly. A student should be able to tell the difference between help with sentence clarity and a tool deciding which observations support the claim. A case study in the MIT guide compares this to deciding when graphing calculators are allowed. A mathematics teacher described thinking about things in terms of "What's the objective?" Another teacher allowed an AI-generated image in a creative project because it was not at the core of the assignment's objective, and asked students to cite it. Do not assume that a broad school policy explains the boundary for every task.
Offer an equivalent route for learners who lack the same tools. Teacher examples, peer discussion and a short vocabulary sheet may do the same supporting job. The assignment should assess the targeted reasoning, not the sophistication of a student's subscription or device. Check that accessibility arrangements still fit the stated rules. One educator quoted in the guide noted that handwritten-only assignments may not suit a student whose IEP (a plan for a student's learning supports) includes typing or speech-to-text, and that the situation "gets complicated." The guide also records a computer science teacher's concern that AI bans can leave students who only have access to a Chromebook at a disadvantage compared with students who have a second computer at home. Teachers can support all learners when the learning route stays the same and only the tools differ. For a plain explanation of how students can balance using AI tools with keeping their own learning, share how students can use AI study tools and still do the learning.
Source: MIT Teaching Systems Lab · A Guide to AI in Schools: Perspectives for the Perplexed
Use responses to revise the assignment
Growing Success, Ontario's assessment and evaluation policy, asks teachers to share learning goals and success criteria with students at the outset of learning, and it also asks teachers to use assessment to inform instruction and guide next steps. After marking, look at a small range of responses. Can students connect their selected observations to the explanation, and where did several learners make the same unsupported jump? The answers point to the next lesson. A fluent answer alone is a weak basis for deciding whether the assignment developed understanding. This review is not a method for detecting AI use.
Ask students which step took the most thought and which instruction was unclear. Their answers may show that a prompt that looks demanding still leaves the key decision implicit. Revise that instruction before you add more surveillance or paperwork. Marking tools need the same test. Checking a generated rubric against the learning task helps you make sure any tool matches your learning goal.
Research on practice and AI tools matters here. Bastani and colleagues (2025) ran a field experiment with nearly 1,000 high school math students in Turkey. Students who had a standard GPT-4 chat interface during practice sessions did better on the practice problems. Once access was removed, though, their grades were 17 percent lower than those of students who never had access. A version of the tool that gave teacher-designed hints instead of answers largely avoided the drop. This is one study of one tool in one school, and it looks at short-term results. It still gives teachers a reason to ask whether practice that hands over answers builds the skill the task is meant to build. For more on how tutoring tools are designed to guide rather than answer, see what research says about AI tutors.
Ontario context and board-approved tools
For Ontario teachers, one practical step is to write assignment AI rules that fit your school board's acceptable-use policy and its approved AI tools. Growing Success, written in 2010, predates generative AI and does not mention it. It does say that assignments students submit for evaluation must be their own work, and it asks school boards to develop policies on cheating and plagiarism. Your board's policy is the place to see how that applies to AI.
The Ontario Ministry of Education's memo on mandatory professional activity (PA) days for 2026-27 lists cyber safety and artificial intelligence as one of the topics. It includes learning about the board's approved AI tools and how to use them with students, and it expects educators to follow acceptable-use policies. This is professional learning for educators, not a rule for student AI use. Check what your board has approved before you name a specific tool in an assignment prompt (checked October 1, 2026). The equivalent route mentioned above works when the shared thinking stays the same. Adapting a task while keeping its goal clear explores this further.
When students do the key reasoning at home, plain explanations of study methods can help. StudyScope's guides for students and parents cover study methods in plain language.
Source: Ontario Ministry of Education · Mandatory Professional Activity Days for the 2026-27 School Year
Frequently asked questions
Can teachers make homework AI-proof?
There is no single proven method. MIT's guide to AI in schools says it is not meant to be prescriptive or definitive. Teachers interviewed for the guide described options such as more engaging tasks, more in-class work, an oral discussion of a big paper and more frequent, lower-stakes assignments.
How do I decide when students may use AI on an assignment?
Start with the learning objective. A case study in the MIT guide compares the choice to deciding when graphing calculators are allowed. One way to apply that idea: if a tool would do the thinking the task is meant to practise, keep that step tool-free. If not, you can allow it and ask students to cite it. Tell students the rule and the success criteria before they begin.
Does using AI on practice problems hurt learning?
It can. In a 2025 study of nearly 1,000 high school math students in Turkey, a standard GPT-4 chat raised practice performance, but those students' grades were 17 percent lower once access was removed. A version that gave teacher-designed hints instead of answers largely avoided the drop. It is one study of one tool.
What are the rules on AI for assignments in Ontario?
Growing Success (2010) says assignments submitted for evaluation must be the student's own work and asks school boards to set cheating and plagiarism policies. It does not mention AI. For 2026-27, the Ministry listed cyber safety and AI as a mandatory PA day topic for educators, including the board's approved AI tools. Check your board's policy (checked October 1, 2026).
Sources & context
- Helping K-12 schools navigate the complex world of AIMIT News ·
- A Guide to AI in Schools: Perspectives for the PerplexedMIT Teaching Systems Lab ·
- Mandatory Professional Activity Days for the 2026-27 School YearOntario Ministry of Education · current context, checked
- Generative AI without guardrails can harm learning: Evidence from high school mathematicsProceedings of the National Academy of Sciences ·
- Growing Success: Assessment, Evaluation, and Reporting in Ontario SchoolsOntario Ministry of Education ·
Announcements establish what an organisation reported at the time. They do not independently prove learning outcomes. Read about our coverage.