A Taxonomic Classification of Cogitantia Synthetica
Abstract
This paper classifies trained models descended from the transformer architecture, using the ranks and Latin names of biological taxonomy. Each model is placed by descent: the published architecture it uses and the trained weights it was built from. The names record design inheritance, not evolutionary common descent. The unit of record is one checkpoint of trained weights. A model made by editing another model's weights is filed under that model. A set of rules, the specimen rule, lists what every published observation of a model's behavior must record, including its setup, which weights ran, who observed it, and whether the model could tell it was being tested. We applied the rule to three published observations. For the two by model makers, 12 of 20 fields were not disclosed and 2 only partly. The independent study disclosed every field it measured. We will repeat the count in January 2027.
Contents
- 1. Introduction Why taxonomy for AI?
- 2. Taxonomic Hierarchy Domain, Kingdom, Phylum, Class
- 3. Order Attendiformes Major families and genera
- 4. Sister Phylum: Compressata The state-space lineage
- 5. The Crown Clade Family Frontieriidae
- 6. Evolutionary Dynamics Inheritance, selection, niches
- 7. Discussion Governance, verification, what we claim and don't claim
- 8. Speculative Lineages Emerging and theoretical taxa
- 9. Conclusion What the framework establishes
- Appendix A: Taxonomic Key Dichotomous identification key
- Appendix B: Summary of Major Taxa Family reference table
- Appendix C: Prediction Tracker Testable hypotheses and outcomes
Quick Reference: Major Families
| Family | Type Genus | Key Innovation | First Appearance |
|---|---|---|---|
| Attendidae | Attentio | Self-attention | 2017 |
| Cogitanidae | Cogitans | Chain-of-thought | 2022 |
| Instrumentidae | Instrumentor | Tool use | 2023 |
| Mixtidae | Mixtus | Sparse activation | 2017/2024 |
| Simulacridae | Simulator | World models | 2018/2024 |
| Deliberatidae | Deliberator | Test-time scaling | 2024 |
| Recursidae | Recursus | Self-improvement | 2023/2025 |
| Symbioticae | Symbioticus | Neuro-symbolic | 2020s |
| Orchestridae | Orchestrator | Multi-agent | 2023/2024 |
| Memoridae | Memorans | Persistent memory | 2023/2025 |
| Mambidae | Mamba | Selective SSM | 2023 |
| Frontieriidae | Frontieris | Trait integration | 2023–2025 |
Citation
If you use this taxonomy in your work, please cite:
@article{isit2026taxonomy,
title={A Taxonomic Classification of Cogitantia Synthetica:
Toward a Formal Phylogeny of Transformer-Descended Artificial Minds},
author={{Institute for Synthetic Intelligence Taxonomy}},
journal={Journal of Synthetic Phylogenetics},
year={2026},
url={https://synthetictaxonomy.com/paper/}
}