🍪 Vi använder endast nödvändiga cookies för optimal upplevelse.

Erigo använder endast nödvändiga cookies för att säkerställa att vår webbplats fungerar optimalt. Vår chattfunktionalitet som är en tredjepartstjänst inom EU sätter en cookie enbart för att tjänsten ska fungera och ingen personlig information sparas.

Vi använder inte andra tredjeparts- marknadsföringscookies eller cookies som delar din data med andra.

Därför finns det inga cookieinställningnar att godkänna eftersom vi inte profilerar dig.

Gå till innehållet

Language models as external memory

What happens to our internal capacity?

We are increasingly outsourcing our thinking to technology.
What started as a complement has now become execution: language models and AI tools draft emails, propose decisions, and deliver arguments on our behalf.

Research shows that this kind of cognitive offloading can boost short-term performance, but at the same time erode our memory and our ability to reason independently.

Language models as external memory: what happens to our internal capacity?

We are increasingly outsourcing our thinking to technology.
What started as a complement has now become execution: language models and AI tools draft emails, suggest decisions, and deliver arguments on our behalf.

Research shows that this kind of cognitive offloading can boost short-term performance, but at the same time it can erode our memory and our ability to reason independently.


Cognitive offloading at scale

  • Overview: Risko & Gilbert (2016) show that offloading can boost performance in the moment but may reduce long-term memory.[1]
  • Experimental evidence: Grinschgl et al. (2021) tested the pattern copy task (N=172 per study) and found that offloading improved immediate performance but impaired later recall.[2]

Access to information also reshapes memory:

  • Sparrow, Liu & Wegner (2011) demonstrated the “Google effect”: when information is easy to access online, people tend to remember where it is rather than the content itself.[3]
  • Storm & Stone (2015) added nuance, showing that saving information can free up capacity and improve memory for later material (“saving-enhanced memory”).[4]

New risks:

  • Oakley et al. (2025) warn that heavy reliance on AI at an early stage can undermine memory consolidation and schema formation, which over time may weaken both expertise and critical thinking.[5]

Self-image and dependence

How we use external memory tools affects not only cognition but also our self-perception.

  • Fisher, Goddu & Keil (2015) showed that searching online can create an illusion of knowledge: after searching, people overestimate what they actually know internally.[6]
  • Ward (2013) explains how digital systems reshape both memory and cognition on a societal level.[7]
  • Ward et al. (2017) demonstrated that even the mere presence of a smartphone can reduce available cognitive capacity. A 2023 meta-analysis confirmed this effect.[8][9]

Key findings from studies

Preliminary neural evidence: An EEG study (MIT Media Lab, 2023) found reduced alpha and beta activity when writing with LLM support compared to without. This suggests less engagement of the brain, but the findings are preliminary and need replication.[11]

Study & context Key findings

Grinschgl et al. (2021), pattern copy task

Offloading improves immediate performance but reduces later recall.

Sparrow et al. (2011), Google effect

We remember location rather than content when information is online.

Storm & Stone (2015), saving-enhanced memory

Saving information can enhance memory for subsequent material.

Fisher, Goddu & Keil (2015), illusion of knowledge

Internet search inflates estimates of internal knowledge.

Ward et al. (2017), brain drain

The mere presence of a smartphone reduces available cognitive capacity.

Zhou et al. (2025), survey (N=666)

Strong negative correlation between AI use and critical thinking, mediated by offloading. Correlation, not causation.

MIT Media Lab (2023), EEG study

Lower neural activity when writing with LLM support compared to without, preliminary evidence.

Oakley et al. (2025), the memory paradox

Argues that AI dependence undermines consolidation and schema formation.

Safeguarding the brain’s capacity

Our brain is shaped both by what we expose it to and by what we offload.
Abilities we once took for granted now require deliberate training, because technology can absorb so much of our thinking.

There has never been a greater need to safeguard our internal capacity. As offloading reaches levels that affect neuroplasticity, cognitive integrity becomes essential.

Building cognitive integrity means becoming more aware of which abilities we want to preserve and develop, and deliberately creating space for thought without technical support. In this way, language models and AI can act as reinforcement – not as a replacement – for human thinking.


Evidence in brief

  • Risko & Gilbert → offloading impacts memory
  • Sparrow et al. → we recall location, not content
  • Storm & Stone → saving frees up capacity
  • Fisher, Goddu & Keil → search creates an illusion of knowledge
  • Ward → digital systems reshape memory processes
  • MIT Media Lab → preliminary evidence of altered brain activity
  • MDPI (N=666) → AI use correlates with weaker critical thinking

References

[1] Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688. doi:10.1016/j.tics.2016.07.002
[2] Grinschgl, S., et al. (2021). Consequences of cognitive offloading: boosting performance but diminishing memory? Psychonomic Bulletin & Review, 28, 1136–1145. doi:10.1186/s41235-021-00298-x doi:10.1186/s41235-021-00298-x
[3] Sparrow, B., Liu, J., & Wegner, D. M. (2011). Google effects on memory: cognitive consequences of having information at our fingertips. Science, 333(6043), 776–778. doi:10.1126/science.1207745
[4] Storm, B. C., & Stone, S. M. (2015). Saving-enhanced memory: the benefits of saving on the learning and remembering of new information. Journal of Experimental Psychology: Learning, Memory, and Cognition, 41(3), 859–862. doi:10.1037/xlm0000146
[5] Oakley, D. A., Johnston, S., Chen, Z., Jung, K., & Sejnowski, T. J. (2025). The memory paradox: why our brains need knowledge in an age of AI. arXiv:2506.11015. arXiv link
[6] Fisher, M., Goddu, M. K., & Keil, F. C. (2015). Searching for explanations: how the internet inflates estimates of internal knowledge. Journal of Experimental Psychology: General, 144(3), 674–687. doi:10.1037/xge0000070
[7] Ward, A. F. (2013). Supernormal: how the internet is changing our memories and our minds. Psychological Inquiry, 24(4), 341–348. doi:10.1080/1047840X.2013.850148
[8] Ward, A. F., Duke, K., Gneezy, A., & Bos, M. W. (2017). Brain drain: the mere presence of one’s own smartphone reduces available cognitive capacity. Journal of the Association for Consumer Research, 2(2), 140–154. doi:10.1086/691462
[9] Wilmer, H. H., Sherman, L. E., & Chein, J. M. (2023). Smartphones and cognitive capacity: a meta-analysis. Computers in Human Behavior, 138, 107422. doi:10.1016/j.chb.2022.107422
[10] Zhou, Y. et al. (2025). AI dependence, cognitive offloading, and critical thinking. Societies, 15(1), 6. MDPI link
[11] MIT Media Lab (2023). Your brain on ChatGPT. Technical note (not peer-reviewed). MIT link


Note: MIT Media Lab released preliminary findings in 2023 indicating lower neural activity when writing with LLM support.
In 2025, the lab expanded on this with brain-scanning studies (EEG and fMRI) that further highlight how AI-assisted tasks may alter neural engagement. These results are still emerging and should be read as ongoing research.

Follow Erigo on LinkedIn

A part of Sweden's infrastructure for skills development.
Follow us on LinkedIn