Artificial intelligence has made its way into classrooms faster than almost any previous technology. From automated feedback tools to lesson-planning assistants, AI is being used to boost productivity and personalise learning. But as with any tool, timing and purpose matter.
A growing body of research suggests that while AI can strengthen learning when used carefully, it can also undermine it if overused. The key lies in how it interacts with one of the brain’s most important systems: working memory.
What Is Working Memory and Why Does It Matter
Working memory is the mental “workspace” where students hold and manipulate information. It’s what helps them solve equations, write essays, or remember instructions long enough to use them. When this system becomes overloaded or bypassed, learning suffers.
According to a recent study, Working Memory in the Age of Artificial Intelligence (Akar, 2025), AI can either scaffold or sabotage memory depending on how it’s applied. When students rely on AI before thinking for themselves, their brains don’t fully encode information. The result is a false sense of fluency, so they feel like they’ve learned, but they haven’t truly remembered.
For teachers, that means AI should act as a scaffold, not a substitute. Support that arrives too early, or in too much detail, interrupts the cognitive struggle that helps ideas stick.
The Risk of Too Much Help
Teachers are balancing two competing goals: improving learning outcomes while managing workload. AI promises to solve both, but it’s not always a shortcut worth taking.
When students receive detailed AI feedback before they’ve attempted a task, they miss the opportunity for retrieval practice, which is one of the most powerful, evidence-based learning techniques. Research from Dunlosky et al. (2013) shows that retrieval, spacing and feedback cycles strengthen long-term recall far more than immediate correction.
Too much AI assistance can also lead to cognitive overload. If a chatbot or generator provides a complete solution all at once, working memory is overwhelmed, and transfer to new tasks is weakened.
A Practical Classroom Routine
Teachers can build AI into classroom routines without losing the learning value. Try this simple structure:
- Make an independent attempt first. Ask students to solve or write before turning to AI.
- Targeted AI support. Allow brief, step-by-step help, e.g. one prompt or example at a time.
- Delayed review. Revisit the material through a low-stakes quiz a day or two later.
In STEM subjects, AI can show one calculation or derivation at a time. In writing, it can model outlines before students produce whole paragraphs. The golden rule: keep AI help short, specific and fading as students gain skill.
Teaching Students to Think Before They Click
AI should never replace human reasoning or memory. Instead, it should nudge students toward deeper understanding and reflection. Schools can support this by:
- Embedding retrieval practice after AI use
- Training students to ask for help only after independent effort
- Discussing cognitive load and timing in digital-literacy sessions
As the Akar (2025) paper concludes, AI becomes powerful when it scaffolds learning rather than short-circuiting it. The challenge for teachers is not whether to use AI, but how to design routines that let memory do the heavy lifting.
