Schedule of Events | Search Abstracts | Invited Symposia | Symposia | Poster Sessions | Data Blitz

How does offloading generative processes to ChatGPT impact learning and memory?

Poster Session D - Monday, March 9, 2026, 8:00 – 10:00 am PDT, Fairview/Kitsilano Ballroom

Anna Kazatchkova¹ (), Joshua A. Skorburg¹, Christopher M. Fiacconi¹; ¹University of Guelph

We often use technology to store information externally and reduce the cognitive load required to accomplish tasks, a process known as cognitive offloading. While offloading can improve productivity, it can also impair learning and memory. Famously, global positioning systems (GPS) help us navigate efficiently, but can also hinder spatial learning. Similarly, Large Language Models (LLMs) can improve productivity by generating tailored outputs for a wide range of tasks, whereas self-generating content engages deeper cognitive processes that enhance memory. Nonetheless, given LLMs ability to generate tailored outputs, are generative aspects of human learning still relevant? The present study explores how offloading generative tasks during study to an LLM influences learning outcomes. Specifically, we investigated how offloading mnemonic generation for randomly-generated five-letter acronyms to LLMs influences both recognition and cued-recall memory performance, compared to self-generation of mnemonics using a fully self-paced (Experiment 1) and time-equated design (Experiment 2). On self-generated trials, participants created their own mnemonics, whereas on LLM-generated trials, participants were asked to type out the mnemonic that was generated on-screen by ChatGPT, and were instructed to study each word before moving on to the next trial. We demonstrated that when mnemonic generation was offloaded to an LLM, both recognition and cued-recall performance were significantly impaired in comparison to self-generation, regardless of whether study was self-paced or equal for both trial types. These results underscore the possible trade-offs between efficiency gains from LLMs with the potential cognitive drawbacks of relying on them for tasks that are crucial for learning and memory.

Topic Area: LONG-TERM MEMORY: Other