A litmus test for AI in education

As teachers, how should we make decisions about when AI should and shouldn’t be involved in teaching and learning? These decisions are tricky and important: teachers have to make room in their already packed to do lists to figure out whether large language models (LLMs) might be useful in specific teaching or learning tasks AND sort out the ethics (and sometimes, legalities) involved in those LLM functions within the context of school policies.

Maybe what we need is one overall idea or principle about what role we want LLMs to play in teaching and learning, and that idea can be a litmus test we use to make these decisions. Here’s one potentially useful question we can use to decide whether to get an LLM involved in our classroom: whose thinking are we replacing?

Context

Many cognitive psychology principles relate directly to what some researchers call “the science of learning” (if you want to explore science of learning resources, the APA Top 20 Principles project summarizes important research in this area). There is a LOT to discuss in this research area, but one of the most important generalizations is: thinking in complex ways about important information increases learning.

This generalization can be a useful lens to use when making decisions about how LLMs should and should not be used in education. Since learning is the result of thinking, we should figure out what thinking the LLM is replacing and whether that replacement is in the service of learning or works against learning. Whenever we consider whether an LLM might be an appropriate tool for a learning task, we should think about what thinking needs to be done in order to accomplish the learning and what thinking it might be OK to “offload” (Wexler, 2026). This offloading might involve students using an LLM as part of a lesson (student cognitive work is “offloaded”) or teachers using an LLM for some teaching task (teacher cognitive work is “offloaded.”)

Notice that using the litmus test leads to different decisions about whether an LLM should be involved in these fictional examples. In the first example, the teacher recognizes that allowing students to use an LLM would get in the way of learning, because it “offloads” the cognitive work that students need to do to learn. In the second example, the teacher uses the litmus test to conclude that using an LLM would be useful because the teacher is already expert in the cognitive work being offloaded to the LLM and can use that expertise to evaluate output from the LLM. These decisions are consistent with research about how novices (students) and experts (teachers) think differently (“A Novice Is Not a Little Expert,” 2024). Novices need to do specific cognitive work in order to learn, while experts can often safely offload that thinking to save them time.

Teachers could keep the litmus test question “what thinking is being replaced?” at the heart of discussions about how LLMs should and should not be used during learning. Humans learn by doing our own thinking, and considering what cognitive work is being offloaded to AI can be a useful way to evaluate when it should and shouldn’t be involved in teaching and learning tasks.

Barcaui, A. (2025). ChatGPT as a cognitive crutch: Evidence from a randomized controlled trial on knowledge retention. Social Sciences & Humanities Open, 12. https://doi.org/10.1016/j.ssaho.2025.102287opens in new window

Kirschner, P. and Hendrick, (2024) A novice is not a little expert. In How Learning Happens: Seminal Works in Educational Psychology and what They Mean in Practice (pp. 58-68). Routledge. https://doi.org/10.4324/9780429061523opens in new window

Kosmyna N. et. al (2025) "Your Brain On ChatGPT: Accumulation Of Cognitive Debt When Using An AI Assistant For Essay Writing Task." arXiv preprint arXiv:2506.08872opens in new window

Top 20 Principles for Teaching and Learning. (2025). American Psychological Association (APA). Retrieved March 3, 2026, from https://www.apa.org/ed/schools/teaching-learning/top-twenty

Wexler, N. (2026, Jan 18). Is Relying on AI Cognitive “Offloading” or “Outsourcing”? Minding the Gap. https://nataliewexler.substack.com/p/is-relying-on-ai-cognitive-offloadingopens in new window

Rob McEntarffer taught high school English, psychology, AP psychology, and philosophy for 13 years. While teaching, he became interested in educational measurement issues and worked as an assessment/evaluation specialist with Lincoln Public Schools starting in 2005. Rob earned his PhD in teaching, learning, and teacher education in 2013, focusing his research on how teachers make room for formative assessment processes in their classrooms. He now teaches in the College of Education at Doane University.