@TeacherToolkit
Ross Morrison McGill founded @TeacherToolkit in 2007, and today, he is one of the ‘most followed educators’on social media in the world. In 2015, he was nominated as one of the ‘500 Most Influential People in Britain’ by The Sunday Times as a result of…
Read more about @TeacherToolkit
Is artificial intelligence improving working memory, or just supporting students to forget?
When AI acts as a scaffold—not a replacement—it boosts memory. But too much help too soon? That’s a problem.
Hidden cognitive risks
This research challenges the idea that more help is always better. Instead, it offers a practical model showing how AI should time its support and keep it short to protect students’ memory.
AI tools like ChatGPT are widely used in schools, praised for accelerating learning and easing workload. But these tools also pose a hidden cognitive risk. According to Working Memory in the Age of Artificial Intelligence (Akar, 2025), the real issue is how and when these tools offer help.
The paper introduces the concept of working-memory co-regulation.
This means AI shouldn’t do the thinking for the student, but instead scaffold it—offering help only after students try to solve a problem themselves. Key ideas like the timing of support and the granularity (detail level) of feedback are what make AI helpful or harmful.
Improving outcomes vs. managing workload
Teachers are caught between two pressures: improving outcomes and managing workload. AI promises to solve both challenges. But the research is clear: poorly timed or overly verbose AI support displaces retrieval practice, one of the most effective learning strategies teachers use (see Dunlowsky et al., 2013).
Students who use AI before they attempt a task, don’t encode memories as well.
Those who get too much help in one go suffer from overloaded working memory and reduced transfer to other tasks. The result? Superficial fluency and weak recall. Teachers need to be more intentional about when and how students use AI.
Adopt a simple classroom routine
Teachers can adopt a simple workflow:
- Ask students to complete a task or quiz independently.
- Provide short, specific, and step-based help using AI—no more than one prompt at a time.
- Ask students to revisit learning using a delayed low-stakes quiz, ideally 24–72 hours later.
For STEM subjects, AI should explain one calculation or derivation at a time. In writing, AI should be used to scaffold outlines before students attempt full paragraphs. Teachers should ensure help is brief, local, and fades as students gain confidence.
Reflection questions for teachers:
- Are teachers asking students to try before turning to AI for help?
- Is AI support kept short and structured step-by-step?
- How often do students revisit material after using AI?
- Is retrieval practice used after AI-assisted tasks?
- Do teachers see AI as a scaffold or a surrogate?
- Are students being taught how to use AI wisely and sparingly?
- Is AI used differently across disciplines (e.g., STEM vs. writing)?
- Is the level of support faded as students gain expertise?
- Could more feedback be delayed rather than immediate?
- Are school leaders discussing cognitive load in AI planning?
Teachers should not ban AI, but they must manage its use carefullly. It’s time to design classroom routines where AI supports memory—not shortcuts it.
The research concludes:
Positioning AI as a scaffold rather than a surrogate offers a tractable route to convert short-term performance gains into durable learning and transfer.
