Evidence for Learning with AI and technology #

Every substantive claim on the Learning with AI and technology page is checked against current research. Here is each claim, how well today’s evidence supports it, and the sources. The full, de-duplicated source list lives on the references page.

Supported · strong evidence — Cognitive offloading is the use of physical action or external tools (including technology) to reduce the internal cognitive demands of a task.

Risko & Gilbert’s 2016 Trends in Cognitive Sciences review is the standard definition and framework for cognitive offloading, and the construct remains well established and widely used in cognitive science through 2026.

Sources: Risko, E. F. & Gilbert, S. J. (2016), Cognitive Offloading. Trends in Cognitive Sciences, 20(9), 676-688 — https://doi.org/10.1016/j.tics.2016.07.002 · full reference ›

Supported · moderate evidence — People readily offload cognitive work onto external aids (such as writing things down or using a device) rather than relying on internal memory, often even when internal capacity would suffice.

Experimental work reviewed and extended by Risko & Gilbert (e.g. preferences for external storage like saving a file rather than memorising) shows offloading is a common default; the general tendency is robust, though its magnitude varies with metacognitive confidence and task framing.

Sources: Risko, E. F. & Gilbert, S. J. (2016), Cognitive Offloading. Trends in Cognitive Sciences, 20(9), 676-688 — https://doi.org/10.1016/j.tics.2016.07.002 · full reference ›

Supported · moderate evidence — Offloading is often adaptive and beneficial: using external aids for routine storage or computation can free internal cognitive resources for other parts of a task.

Risko & Gilbert frame offloading as frequently rational and performance-enhancing in the moment; the page’s distinction between offloading harmless routine work (e.g. a calculator for understood arithmetic) and offloading the effort that produces learning is consistent with this nuanced view.

Sources: Risko, E. F. & Gilbert, S. J. (2016), Cognitive Offloading. Trends in Cognitive Sciences, 20(9), 676-688 — https://doi.org/10.1016/j.tics.2016.07.002 · full reference ›

Mixed · weak evidence — Higher reported use of AI tools is associated with greater cognitive offloading and with lower critical-thinking performance, with offloading partly mediating the relationship.

Gerlich (2025) reports a significant negative correlation between frequent AI-tool use and critical thinking, mediated by cognitive offloading, in a survey/interview sample. The finding is directionally plausible and consistent with offloading theory, but it is recent, correlational (no causal claim), self-report based, and not yet replicated, so the page presents it as a caution rather than a settled result.

Sources: Gerlich, M. (2025), AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking. Societies, 15(1), 6 — https://doi.org/10.3390/soc15010006 · full reference ›

Supported · strong evidence — Retrieval practice — testing yourself to pull information out of memory — produces more durable learning than rereading, so using AI to generate quizzes targets the highest-value study activity.

The testing effect is one of the most robust findings in learning science; Adesope, Trevisan & Sundararajan’s 2017 meta-analysis of 118 studies found a moderate-to-large benefit of practice testing over restudy, and it remains consensus in 2026. The page’s recommendation to have AI quiz the learner applies this directly.

Sources: Adesope, O. O., Trevisan, D. A. & Sundararajan, N. (2017), Rethinking the Use of Tests: A Meta-Analysis of Practice Testing. Review of Educational Research, 87(3), 659-701 — https://doi.org/10.3102/0034654316689306 · full reference ›

Supported · strong evidence — The difficulty and effort of recalling or working something out is itself what builds learning, so removing that struggle (e.g. by accepting an AI-generated answer) removes the learning benefit.

The ‘desirable difficulties’ principle (Bjork & Bjork) holds that conditions introducing effortful retrieval and processing improve long-term retention even as they slow apparent progress; this is well established and directly supports the page’s claim that offloading the effortful step forfeits the learning.

Sources: Bjork, E. L. & Bjork, R. A. (2011), Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning. In Psychology and the Real World — https://bjorklab.psych.ucla.edu/research/ · full reference ›

Supported · moderate evidence — Generating an answer or attempting a problem before receiving feedback or the correct answer improves learning more than studying the answer alone.

The benefit of attempting retrieval before checking is supported by the testing-effect literature and by research on the generation effect and feedback timing; producing one’s own response first, then verifying, is a well-grounded strategy, which the page applies to AI (‘attempt first, check second’).