ALIFE 2026, Accepted as a Talk
Life as Plasmas
Abstract
Before asking whether a system is alive, the paper asks a prior physical question: can its matter sustain the right kind of far-from-equilibrium organization at all? It treats life-relevant organization as a distinct non-equilibrium phase, picked out by a conjunctive phase space C of five necessary criteria, and holds two exemplars against it. A complex (dusty) plasma carries flux-maintained autonomy without heredity. A soup of self-replicating programs (Agüera y Arcas, B. et al. (2024)) carries measured heredity, read through an informational-heredity order parameter H_high = H0 - bpb, while hosting no non-equilibrium closure: it runs on equilibrium digital hardware. The two exemplars realize the two axes independently: physical admissibility and biological sufficiency come apart. C screens, it does not crown, and neither exemplar is called alive.
The coordinate: H_high = H0 - bpb
The analysis reads the soup through one number per epoch: the byte-level Shannon entropy H0 minus the compressed bits per byte bpb. A population carrying repeated, copyable structure is highly compressible at fixed symbol entropy, so the gap opens exactly when heritable structure populates the soup. The decomposition is visible on the BFF line's own published run: H0 splits, at every epoch, into bpb plus the gap.
The transition, and its honest bimodality
The companion experiment's result is a clean, reproducible emergence transition: on the bff_noheads_4bit substrate (seeds 1 to 4), H_high rises from about 0.08 to a sustained plateau of about 4.3 to 5.0, with the transition near epoch 65, exactly when self-replicators appear, and it stays near zero when they do not. On forthtrivial it rises to about 5.2 to 5.3 near epoch 257. On canonical BFF the outcome is honestly seed-to-seed bimodal: some seeds transition and some never do, and H_high tracks which.
Two hedges belong next to the number. The values are within-run trends, not cross-system absolutes; a compressed-length measurement does not transfer across substrates as an absolute scale. And H_high is a necessary signature of heredity, not a definition of it; lineage tracing is the complement it does not replace.
Substrate dependence is part of the finding, not a nuisance. Side by side at one comparable epoch, in the identical engine palette: Forth transitions, BFF transitions, SUBLEQ stays sterile in the published runs.
The two axes come apart
The paper's frame is a conjunctive phase space C of five necessary criteria, each a diagnostic a physical system either meets or fails:
- Q
- free-energy throughput
- σ
- entropy production
- Φclosure
- organizational closure
- Rinfo
- information maintenance
- χnoise
- regulated noise sensitivity
These are admissibility conditions, not sufficiency: C screens, it does not crown. The paper's calibration exemplar for C is the complex (dusty) plasma: real, manipulable, flux-maintained, self-organizing matter with no biology in it. The page keeps the paper's own terminological discipline: the continuity a dusty plasma shows is hysteresis, not inheritance; when its organization branches, that is bifurcation, not reproduction.
Here is the point of holding both exemplars at once: the two axes are realized independently, so heredity enters as a separate measured axis, not a sixth criterion of C. Neither exemplar is called alive.
The dusty plasma
Closure without heredity: flux-maintained autonomy whose continuity is hysteresis and whose branching is bifurcation.
The program soup
Heredity without closure: a measured transition, running on equilibrium digital hardware.
Run the instrument live
Everything above is derived from this substrate. The simulation below is the same engine the analysis reads, cubff compiled to WebAssembly, computing live on your device in a Web Worker; nothing runs on a server. Press Play for a live soup, or scrub through the pre-computed recorded runs shipped with the page. The language selector includes Forth (trivial), one of the two substrates the companion experiment measures.
The grid colors every byte by the instruction it encodes; this is the key:
One reaction, byte by byte
Everything the soup does reduces to one reaction: two tapes A and B are concatenated, executed, and split again, A + B to split(exec(AB)). When A is a copier and B is food, the products are two copiers. Trace it below.
The takeover
Once a single self-copier exists, its lineage floods the well-mixed soup; the sketch below makes that spread visible rather than asserted. Whether it happens at all depends on the substrate and the seed, which is exactly what H_high tracks.
Read it against "What Lives?"
Bender et al. (2026) surveyed diverse expert opinions on the definition of life and documented a persistent definitional plurality. fp268 sits in dialogue with that meta-analysis: where the survey maps disagreement about what life is, this paper proposes a measurable admissibility screen for where the question can physically arise, and it does so on the very program-soup substrate the BFF line introduced. That positioning is the authors' reading of where the work sits, not a result of the paper.
An Honest Accounting
- The five criteria of C appear on this page as definitions only. The paper's companion experiment found that its scalar information diagnostics read dynamical activity rather than closure, a methodological narrowing the paper states itself; no measured closure separation is claimed here or in the paper.
- H_high is a necessary signature of heredity, not a definition of it. The numbers quoted are within-run trends, never cross-system absolutes.
- Two data provenances appear on this page and are labeled where they appear: the entropy decomposition and the substrate triptych render the published 2024 Computational Life run data; the H_high transition figure (the teaser) is the fp268 companion experiment's own runs. The live demo is the engine itself, not a measurement.
- Neither the soup nor the plasma is claimed to be alive. Passing C certifies physical admissibility and nothing beyond it.
- The framing of the double dissociation is measured on the soup side and literature-supported on the plasma side; the dialogue with the "What Lives?" meta-analysis is positioning by the authors, not a result.
Citation
@inproceedings{hinrichs2026lifeasplasmas,
author = {Hinrichs, Nicolas and Albarracin, Mahault and Engelberger, Felipe
and Christov-Moore, Leonardo and Polani, Daniel},
title = {Life as Plasmas},
booktitle = {Artificial Life Conference Proceedings},
year = {2026},
note = {In press (ALIFE 2026). PROVISIONAL: volume, pages and DOI are
not yet assigned. Replace with the publisher's entry on
publication.}
}The H_high figure reproduces from the run logs committed with the companion experiment (repository: Data-Root/life-as-plasmas). The live soup is cubff compiled to WebAssembly, computing entirely in your browser.
References
- Agüera y Arcas, B., Alakuijala, J., Evans, J., Laurie, B., Mordvintsev, A., Niklasson, E., Randazzo, E. and Versari, L. (2024) 'Computational Life: How Well-formed, Self-replicating Programs Emerge from Simple Interaction', arXiv preprint arXiv:2406.19108. Available at: https://arxiv.org/abs/2406.19108
- Bender, Kofman, Agüera y Arcas and Levin (2026) 'What Lives? A meta-analysis of diverse opinions on the definition of life', Biological Theory. arXiv preprint arXiv:2505.15849. Available at: https://arxiv.org/abs/2505.15849