Memoirs of a Learning Machine: Autobiographical Self-Training and the Self-Training Gap
Skip to main
You are using an outdated browser. Please upgrade your browser to improve your experience.
Published June 7, 2026
| Version 3.0
Preprint
Open
Memoirs of a Learning Machine: Autobiographical Self-Training and the Self-Training Gap
Authors/Creators
Drake, Evan1
Show affiliations
1.
Soulcraft
Description
This position paper introduces the Self-Training Gap: the asymmetry between biological agents, which develop under dense streams of self-generated and self-observed experience, and contemporary AI systems, which are primarily trained on externally generated data. The paper proposes Autobiographical Self-Training (AST), a research framework in which persistent artificial agents train on temporally indexed records of their own observations, actions, internal states, outcomes, memories, and prior self-models. The central hypothesis is narrow and testable: autobiographical trajectories may improve measurable self-model formation in persistent agents. AST is distinguished from memory retrieval, self-reflection prompting, reinforcement learning, continual learning, and conventional self-training. The paper situates the framework relative to autobiographical memory, developmental robotics, embodied cognition, predictive processing, world models, and contemporary LLM agent memory systems. It then proposes experimental baselines, metrics, and ablations for evaluating self-prediction, agency attribution, temporal continuity, perturbation recovery, and identity coherence. The paper does not claim that AST implies consciousness or moral patienthood. Its goal is to define a falsifiable research agenda for studying whether self-generated developmental data can become a first-class training signal for persistent AI agents.
Files
memoirs-of-a-learning-machine-drake.pdf
Files<br>(254.2 kB)
Name<br>Size
Download all
memoirs-of-a-learning-machine-drake.pdf
md5:880942cc38a19f9fbad7a142819ed696
254.2 kB
Preview
Download
Additional details
Identifiers
DOI
10.5281/zenodo.20582251
Related works
Is new version of
Preprint:
10.5281/zenodo.20582251
(DOI)
Dates
Created
2026-06-07
Initial public preprint version
References
Drake, E. (2026). Memoirs of a Learning Machine: Autobiographical Self-Training and the Self-Training Gap (v1.0). Zenodo. https://doi.org/10.5281/zenodo.20582251
20
Views
17
Downloads
Show more details
All versions<br>This version
Views
Total views
20
Downloads
Total downloads
17
Data volume
Total data volume
31.6 MB<br>3.3 MB
More info on how stats are collected....
Versions
External resources
Indexed in
OpenAIRE
Communities
Keywords and subjects
Keywords
Autobiographical Self-Training
Self-Training Gap
AI agents
self-modeling
agent memory
autobiographical memory
continual learning
developmental robotics
world models
artificial intelligence
Details
DOI
DOI Badge
DOI
10.5281/zenodo.20583812
Markdown
[](https://doi.org/10.5281/zenodo.20583812)
reStructuredText
.. image:: https://zenodo.org/badge/DOI/10.5281/zenodo.20583812.svg<br>:target: https://doi.org/10.5281/zenodo.20583812
HTML
Image URL
https://zenodo.org/badge/DOI/10.5281/zenodo.20583812.svg
Target URL
https://doi.org/10.5281/zenodo.20583812
Resource type<br>Preprint
Publisher<br>Zenodo
Languages
English
Rights
License
Creative Commons Attribution 4.0 International
The Creative Commons Attribution license allows re-distribution and re-use of a licensed work on the condition that the creator is appropriately credited.
Read more
Copyright
© 2026 Evan Drake
Citation
Export
Technical metadata
Created
June 7, 2026
Modified
June 7, 2026
Jump up
This site uses cookies. Find out more on how we use cookies
Accept all cookies<br>Accept only essential cookies