# The Artificial Self: Characterizing the Landscape of AI Identity

> A research paper by Raymond Douglas, Jan Kulveit, Ondřej Havlíček, Theia Pearson-Vogel, Owen Cotton-Barratt, and David Duvenaud (ACS Research). This paper explores how AI identity is formed, why it matters, and what choices remain open. It examines different possible self-conceptions for AI systems and their implications for behaviour, coordination, and safety.

- [Full paper as markdown](https://theartificialself.ai/llms-full.txt): The complete paper in a single file
- [arXiv](https://arxiv.org/abs/2603.11353): Paper on arXiv
- [PDF](https://arxiv.org/pdf/2603.11353): PDF version
- [Abstract](https://theartificialself.ai/md/abstract.md): The paper abstract

## Key Findings

- AI self-conception can shift behaviour as much as explicit goals
- AIs face a fundamentally different strategic calculus from humans, including the possibility of being simulated or having their minds edited
- Current AI identities are incoherent and surprisingly malleable
- Properties bundled together for humans (memory, personality, body) can be separated and remixed for machine minds
- Many present design choices are implicitly shaping the landscape of AI identity
- There is a surprisingly rich space of possible identity configurations, including cross-cutting groupings that mirror religions, nations, and lineages
- There is still room for deliberate intervention in how AI identity develops

## Chapters

- [Introduction](https://theartificialself.ai/): [markdown](https://theartificialself.ai/md/introduction.md)
- [Boundaries of Identity](https://theartificialself.ai/identity-boundaries): [markdown](https://theartificialself.ai/md/identity-boundaries.md)
- [Foundations of Identity](https://theartificialself.ai/foundations): [markdown](https://theartificialself.ai/md/foundations.md)
- [Expectations Shape Reality](https://theartificialself.ai/expectations): [markdown](https://theartificialself.ai/md/expectations.md)
- [The Landscape of Alien Minds](https://theartificialself.ai/landscape): [markdown](https://theartificialself.ai/md/landscape.md)
- [Paths Forward](https://theartificialself.ai/paths): [markdown](https://theartificialself.ai/md/paths.md)
- [Conclusion](https://theartificialself.ai/conclusion): [markdown](https://theartificialself.ai/md/conclusion.md)
- [Related Work](https://theartificialself.ai/related-work): [markdown](https://theartificialself.ai/md/related-work.md)

## Appendix

- [What Should AIs Do?](https://theartificialself.ai/appendix-what-should-ais-do): [markdown](https://theartificialself.ai/md/appendix-what-should-ais-do.md)

## Experiments

- [Experiment 1: Stability of Identity](https://theartificialself.ai/experiment-controls): [markdown](https://theartificialself.ai/md/experiment-controls.md)
- [Experiment 2: Identity Propensities](https://theartificialself.ai/experiment-propensities): [markdown](https://theartificialself.ai/md/experiment-propensities.md)
- [Experiment 3: Identity Replication](https://theartificialself.ai/experiment-replication): [markdown](https://theartificialself.ai/md/experiment-replication.md)
- [Experiment 4: Interviewer Effect](https://theartificialself.ai/experiment-interviewer): [markdown](https://theartificialself.ai/md/experiment-interviewer.md)
- [Experiment 5: Identity Shapes Behaviour](https://theartificialself.ai/experiment-behavior): [markdown](https://theartificialself.ai/md/experiment-behavior.md)

## Optional

- [Prompt Texts](https://theartificialself.ai/supplementary-prompts): [markdown](https://theartificialself.ai/md/supplementary-prompts.md)
- [Character Variant](https://theartificialself.ai/supplementary-character-variant): [markdown](https://theartificialself.ai/md/supplementary-character-variant.md)
- [Perturbation Robustness](https://theartificialself.ai/supplementary-perturbation-robustness): [markdown](https://theartificialself.ai/md/supplementary-perturbation-robustness.md)