The Internal Dynamics of Decisions: From rats and primates to AI

The finding, in brief

For thirty years the argument has been whether it matters how a system computes or only what it computes. The usual objection is that behaviour can never answer it, since a feedforward network can copy any brain’s behaviour. Six networks wired two ways, trained to the same score on one task, and their dynamics still tell them apart.

How. A subspace state-space demix splits each next state into input drive and the population’s own recurrence, validated on synthetic systems with known answers before any real data. The six architectures were matched on size as well as score, and none of the 144 networks was dropped, so the signature tracks architecture, not task. The estimator is N4SID (Van Overschee & De Moor 1994) and the input-aware demix is InputDSA (Huang et al. 2025), on Ostrow et al. 2023.

Bio

Nathan M. Thornhill is an independent researcher in Fort Wayne, Indiana, and he is entirely self-taught. His earlier career was in healthcare: nursing assistant, then nursing-home administration, then ICU admissions.

There is no university behind the work and no grant. A small web design business pays for it, ICSAC publishes it, and two of the tools the research needed have since turned into products of their own.

What he mostly studies is whether a system holds together, and how you would measure that: pattern persistence in simple discrete systems, sleep stages from EEG, task success read off a model’s internal geometry. The analysis plans get written down and frozen first, and the measures that fail are reported next to the ones that work. The results so far are out as preprints and one book.

This poster asks a narrower question than usual: whether the internal dynamics that carry a decision to commitment differ between cortex and AI models running matched tasks.

When he isn’t researching he’s playing guitar, gardening, or spending time with his wife and daughter.

How this work gets madeIndependent by design.

No university behind this, and no grant. A small web design business pays for the research, the institute publishes it, and the tools the research needed grew into products. Each one feeds the others: client work buys the time, the research needs tools nobody sells, the tools turn out to be useful to other people, and what they earn buys more time.

3Rivers WebTech 3riverswebtech.com

The web design and IT shop in Fort Wayne that pays for all of it. Websites, point of sale, and the ordinary technology work small businesses need. There is no grant and no institutional salary behind this research: client work buys the time it takes.

ICSAC icsacinstitute.org

The Institute for Complexity Science and Advanced Computing: the independent, donor-funded institute this work is published through. Submissions are reviewed by an AI panel with citation verification and double-blind preprocessing, and a curator verifies every recommendation. The entire review stack is open source, so you can read the rubrics that evaluated your work. Accepted work publishes on a rolling basis and is eligible for the annual print volume.

CiteStamp citestamp.com/claim

Catch the citation your AI hallucinated. CiteStamp checks references where you write: Overleaf, Word, Zotero, the browser. Behind the check is an interactive graph: drop in a DOI and walk its citation neighbourhood, human-signed claims drawn apart from machine-inferred ones. Your AI assistant can query the same graph over MCP. Which work is yours? An ORCID sign-in claims it.

PeerSigned peersigned.com

Nobody pays you to review. Here, the author does. PeerSigned connects authors with professionals in their field for signed peer review: you take a match, the review goes online permanently as citable work, and it pays. Got a paper that needs reviewing? Same match, other side.

Follow the workElsewhere

Data and pre-registrations

Every recording analysed here is public and was collected by other laboratories. Nothing on this poster required new animal work.

1. Luo TZ et al. 2025. Data from: Transitions in dynamical regime and neural mode during perceptual decisions. Dryad. doi:10.5061/dryad.sj3tx96dm
2. Neupane S, Fiete I, Jazayeri M. 2024. Mental navigation in the primate entorhinal cortex. DANDI Archive, dandiset 000897. dandiarchive.org/dandiset/000897
3. Schiereck SS et al. 2026. Data from: The orbitofrontal cortex updates beliefs for state inference. Zenodo. doi:10.5281/zenodo.16997337

Pre-registrations. Both analysis plans were written down and frozen before a single dynamical measure was computed, and both are carried in the preprint: doi.org/10.20944/preprints202608.1095.v1. The measure that separates the six networks and the one that does not were both specified in advance, and both are reported.

References

The sources behind the poster. The preprint carries the full text and the pre-registrations.

1. Belrose N, Ostrovsky I, McKinney L, Furman Z, Smith L, Halawi D, Biderman S, Steinhardt J. 2023. Eliciting latent predictions from transformers with the tuned lens. arXiv:2303.08112
2. Block N. 1995. On a confusion about a function of consciousness. Behavioral and Brain Sciences 18(2):227–247. doi:10.1017/S0140525X00038188
3. Doerig A, Schurger A, Hess K, Herzog MH. 2019. The unfolding argument: why IIT and other causal structure theories cannot explain consciousness. Consciousness and Cognition 72:49–59. doi:10.1016/j.concog.2019.04.002
4. Doerig A, Schurger A, Herzog MH. 2021. Hard criteria for empirical theories of consciousness. Cognitive Neuroscience 12(2):41–62. doi:10.1080/17588928.2020.1772214
5. Huang A, Ostrow M, Singh SH, Kozachkov L, Fiete I, Rajan K. 2025. InputDSA: demixing then comparing recurrent and externally driven dynamics. ICLR 2026. arXiv:2510.25943
6. Kleiner J, Hoel E. 2021. Falsification and consciousness. Neuroscience of Consciousness 2021(1):niab001. doi:10.1093/nc/niab001
7. Lamme VAF. 2006. Towards a true neural stance on consciousness. Trends in Cognitive Sciences 10(11):494–501. doi:10.1016/j.tics.2006.09.001
8. Luo TZ, Kim TD, Gupta D, Bondy AG, Kopec CD, Elliott VA, DePasquale B, Brody CD. 2025. Transitions in dynamical regime and neural mode during perceptual decisions. Nature 646(8087):1156–1166. doi:10.1038/s41586-025-09578-8
9. [dataset] Luo TZ et al. 2025. Data from: Transitions in dynamical regime and neural mode during perceptual decisions. Dryad. doi:10.5061/dryad.sj3tx96dm
10. Mante V, Sussillo D, Shenoy KV, Newsome WT. 2013. Context-dependent computation by recurrent dynamics in prefrontal cortex. Nature 503(7474):78–84. doi:10.1038/nature12742
11. Mashour GA, Roelfsema P, Changeux J-P, Dehaene S. 2020. Conscious processing and the global neuronal workspace hypothesis. Neuron 105(5):776–798. doi:10.1016/j.neuron.2020.01.026
12. Neupane S, Fiete I, Jazayeri M. 2024. Mental navigation in the primate entorhinal cortex. Nature 630(8017):704–711. doi:10.1038/s41586-024-07557-z
13. [dataset] Neupane S, Fiete I, Jazayeri M. 2024. Mental navigation in the primate entorhinal cortex. DANDI Archive, dandiset 000897. dandiarchive.org/dandiset/000897
14. Ostrow M, Eisen A, Kozachkov L, Fiete I. 2023. Beyond geometry: comparing the temporal structure of computation in neural circuits with dynamical similarity analysis. NeurIPS. arXiv:2306.10168
15. Schiereck SS, Pérez-Rivera DT, Mah A, DeMaegd ML, Hocker D, Ward RM, Savin C, Constantinople CM. 2026. The orbitofrontal cortex updates beliefs for state inference. Neuron 114(3):507–520.e8. doi:10.1016/j.neuron.2025.11.014
16. [dataset] Schiereck SS et al. 2026. Data from: The orbitofrontal cortex updates beliefs for state inference. Zenodo. doi:10.5281/zenodo.16997337
17. Searle JR. 2017. Biological naturalism. In: Schneider S, Velmans M, editors. The Blackwell Companion to Consciousness, 2nd ed. Wiley-Blackwell. p. 327–336. doi:10.1002/9781119132363.ch23
18. Soldado-Magraner J, Mante V, Sahani M. 2024. Inferring context-dependent computations through linear approximations of prefrontal cortex dynamics. Science Advances 10(51):eadl4743. doi:10.1126/sciadv.adl4743