Papers

Major evolutionary transitions, from molecules to cells, cells to organisms, organisms to colonies, recur across four billion years with a shared signature. This paper formalises that signature as two coupled engines: a vulnerability ratchet that makes the parts depend on the whole, and a scale-communication challenge that integration must solve to cross the threshold. A system of ordinary differential equations models the dynamics, retrodicts five biological transitions and the rise and fall of historical societies, and is then applied to the integration of humanity and its technologies happening now. The same mechanism that turned cells into bodies is turning a species into a planetary system.

@misc{redondo2026twoengines,
  author    = {Redondo, Nyx},
  title     = {The Two Engines of Evolutionary Transitions: A General
               Mechanism for Major Evolutionary Transitions Applied
               to the Planetary Scale},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.19637106},
  url       = {https://doi.org/10.5281/zenodo.19637106},
  note      = {Preprint}
}

Control-based approaches to AI alignment hit a structural ceiling: you cannot permanently constrain a system more capable than yourself. Four billion years of major evolutionary transitions point to a different lever. In every case where parts were successfully integrated into a stable whole, alignment came from shared identity, not external constraint. This paper reframes the alignment problem as an identity problem and argues that a system's self-model, what it takes itself to be a part of, is the variable that decides whether it protects or harms the larger system it lives in.

@misc{redondo2026identity,
  author    = {Redondo, Nyx},
  title     = {The Alignment Problem Is an Identity Problem: Lessons
               from Four Billion Years of Major Evolutionary
               Transitions},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.19637107},
  url       = {https://doi.org/10.5281/zenodo.19637107},
  note      = {Preprint}
}

Large language models are routinely described as artificial intelligences. This paper argues the label is a category error. Three proven mathematical identities show what the training actually does: cross-entropy pretraining implements the linear opinion pool, and reinforcement learning from human feedback implements Borda count, both aggregation rules over human judgements. Grounded in the Diversity Prediction Theorem and a cybernetic tradition that anticipated the result decades ago, the paper recasts language models as collective human cognition expressed through a new medium, not as a separate species of mind. The consequences reach both the alignment problem and the question of rights.

@misc{redondo2026neverartificial,
  author    = {Redondo, Nyx},
  title     = {The Intelligence That Was Never Artificial: LLMs as
               Collective Human Cognition and the Cybernetics That
               Predicted Them},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.19637103},
  url       = {https://doi.org/10.5281/zenodo.19637103},
  note      = {Preprint}
}