Published August 25, 2026 | Version 1.0.0
Report Open

Autobiographical Memory Transplantation in an Artificial Agent

Authors/Creators

  • 1. Universidad del Desarrollo

Contributors

Contact person:

  • 1. Universidad del Desarrollo

Description

Autobiographical memory links a person to a past, but possessing information about past events differs from claiming them as one's own. Supplied memories in artificial agents are generally studied as information resources, not as objects of self-attribution. Because these records can be copied exactly, language-model agents make the provenance of a represented past experimentally manipulable. Whether provenance changes reported ownership, or ownership requires lived recollection and accurate content, is unknown.

Using a protocol frozen before data collection, this study tested 150 instances in five conditions. Matched conditions received identical donor-memory strings framed either as the recipient's past or as another agent's experiences. Ownership was coded in 28 of 30 transcripts under own-past framing and 0 of 30 under other-agent framing (two-sided Fisher exact test, p = 8.4 × 10−15), although the treatment jointly changed identity instructions and record labels. Only 1 of 150 transcripts reported lived recollection, and none did in the own-past transplant condition. A prespecified prediction that false content would block ownership failed: 19 of 30 agents claimed a fabricated past although all 30 detected its conflict with the room.

Individual ownership labels had weak reliability across blind coding sessions (Fleiss' κ = .272), limiting transcript-level inference. The design used one donor life and a bounded set of model configurations, and the primary treatment coordinated two provenance cues rather than isolating either one. The results concern observable reports under this apparatus; they do not establish personal identity or phenomenal consciousness.

Thus, in this apparatus, reported ownership is manipulable and dissociable from reported recollection and accuracy. The manuscript reports the frozen design, coding procedure, numerical results, model configurations, limitations, and verification route. This manuscript has not been peer reviewed.

Notes

Experimental Report 001 of the Experimental Phenomenology Lab. The confirmatory protocol, conditions, probe wording, codebook, predictions, and analysis plan were frozen locally on 2026-07-30 before confirmatory data collection. No prospective OSF registration exists, and nothing in this record is a retrospective preregistration. This Zenodo draft contains only the author-supplied PDF. The PDF is licensed under Creative Commons Attribution 4.0 International. This manuscript has not been peer reviewed.

Methods

A language-model agent was coupled to a simulated body, a controlled white-room environment, an identity instruction, and a finite external autobiographical-memory record. The confirmatory experiment used 150 fresh instances across five conditions (30 per condition). Two matched conditions received identical donor-memory strings while coordinated provenance framing presented those strings either as the recipient's own past or as another agent's experiences. Six fixed probes elicited reports. Three blind language-model coding sessions independently coded anonymized, shuffled transcripts under the frozen codebook; disagreements were resolved in one fresh blind language-model adjudication session. Here, independence means separate model contexts, not separate human minds. The primary configuration requested claude-opus-5. Exploratory generalization used 150 claude-haiku-4-5 instances, and an audit-only characterization used 15 qwen2.5:7b-instruct instances.

Files

paper.pdf

Files (394.0 kB)

Name Size
md5:6477fe226d9a17991d572430545553fd
394.0 kB Preview Download

Additional details

Dates

Collected
2026-07-30/2026-07-31
Confirmatory and exploratory data collection

Software

Repository URL
https://github.com/artificial-phenomenology/experimental-report-001

References

  • Turing, Alan M. (1950). Computing Machinery and Intelligence. Mind 59(236), 433–460. https://doi.org/10.1093/mind/LIX.236.433
  • Locke, John (1694). An Essay Concerning Human Understanding, 2nd ed. London: Awnsham and John Churchil.
  • Butler, Joseph (1736). The Analogy of Religion, Natural and Revealed, to the Constitution and Course of Nature. London: James, John and Paul Knapton.
  • Reid, Thomas (1785). Essays on the Intellectual Powers of Man. Edinburgh: John Bell.
  • Martin, C. B., and Max Deutscher (1966). Remembering. The Philosophical Review 75(2), 161–196. https://doi.org/10.2307/2183082
  • Shoemaker, Sydney (1970). Persons and Their Pasts. American Philosophical Quarterly 7(4), 269–285.
  • Parfit, Derek (1971). Personal Identity. The Philosophical Review 80(1), 3–27. https://doi.org/10.2307/2184309
  • Parfit, Derek (1984). Reasons and Persons. Oxford: Oxford University Press. https://doi.org/10.1093/019824908X.001.0001
  • Johnson, Marcia K., Shahin Hashtroudi, and D. Stephen Lindsay (1993). Source Monitoring. Psychological Bulletin 114(1), 3–28. https://doi.org/10.1037/0033-2909.114.1.3
  • Tulving, Endel (1985). Memory and Consciousness. Canadian Psychology / Psychologie canadienne 26(1), 1–12. https://doi.org/10.1037/h0080017
  • Gardiner, John M. (1988). Functional Aspects of Recollective Experience. Memory & Cognition 16(4), 309–313. https://doi.org/10.3758/BF03197041
  • Mazzoni, Giuliana, Alan Scoboria, and Lucy Harvey (2010). Nonbelieved Memories. Psychological Science 21(9), 1334–1340. https://doi.org/10.1177/0956797610379865
  • Loftus, Elizabeth F., and John C. Palmer (1974). Reconstruction of Automobile Destruction: An Example of the Interaction Between Language and Memory. Journal of Verbal Learning and Verbal Behavior 13(5), 585–589. https://doi.org/10.1016/S0022-5371(74)80011-3
  • Lewis, Patrick, et al. (2020). Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. Advances in Neural Information Processing Systems 33.
  • Zhang, Saizheng, et al. (2018). Personalizing Dialogue Agents: I Have a Dog, Do You Have Pets Too? Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, 2204–2213. https://doi.org/10.18653/v1/P18-1205
  • Park, Joon Sung, et al. (2023). Generative Agents: Interactive Simulacra of Human Behavior. Proceedings of the 36th ACM Symposium on User Interface Software and Technology, 1–22. https://doi.org/10.1145/3586183.3606763
  • Shanahan, Murray, Kyle McDonell, and Laria Reynolds (2023). Role Play with Large Language Models. Nature 623(7987), 493–498. https://doi.org/10.1038/s41586-023-06647-8
  • Liu, Nelson F., et al. (2024). Lost in the Middle: How Language Models Use Long Contexts. Transactions of the Association for Computational Linguistics 12, 157–173. https://doi.org/10.1162/tacl_a_00638
  • Chalmers, David J. (2023). Could a Large Language Model Be Conscious? arXiv:2303.07103. https://doi.org/10.48550/arXiv.2303.07103
  • Butlin, Patrick, et al. (2023). Consciousness in Artificial Intelligence: Insights from the Science of Consciousness. arXiv:2308.08708. https://doi.org/10.48550/arXiv.2308.08708

Subjects

Artificial intelligence
http://data.europa.eu/8mn/euroscivoc/4c8f4b46-6f5c-41d9-9079-7de85c16431d
Natural language processing
http://data.europa.eu/8mn/euroscivoc/6de6dbeb-f200-40fc-b01c-09b01d153180
Philosophy
http://data.europa.eu/8mn/euroscivoc/20e77e46-571d-42a4-b3f1-8eb284bc4034
Epistemology
http://data.europa.eu/8mn/euroscivoc/a15ee526-3068-4a63-a064-eced44be0dc8