pith. sign in

arxiv: 2606.02840 · v1 · pith:PEJA5Z6Enew · submitted 2026-06-01 · 🧬 q-bio.PE · cs.MA· cs.NE· nlin.AO

Self-Regulation through Communication in Evolved Neural Agents

Joshua Nunley This is my paper

Pith reviewed 2026-06-28 11:28 UTC · model grok-4.3

classification 🧬 q-bio.PE cs.MAcs.NEnlin.AO
keywords communication evolutionself-regulationneural agentspredator avoidanceCTRNNbehavioral strategiesevolutionary simulationvocal self-hearing
0
0 comments X

The pith

Evolved neural agents develop communication that regulates their own escape behavior via self-hearing.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper runs evolutionary simulations of pairs of agents controlled by continuous-time recurrent neural networks in a minimal predator avoidance task where agents can hear their own vocalizations. Three strategies dominate among perfect-fitness agents: safety calling from cover, alarm indication of threats, and self-regulatory calling that requires hearing one's own signal to maintain escape. Removing self-hearing selectively disables the third group while leaving the first intact. This shows communication can evolve to control the caller's own actions rather than only to transfer information to a partner.

Core claim

Across 112 perfect-fitness agents from over 2,000 evolutionary runs, three strategies account for 81 percent of agents. Safety calling occurs in 39 percent, where agents signal after reaching cover. Alarm indication occurs in 22 percent, where agents vocalize on threat presence without self-hearing reliance. Self-regulatory calling occurs in 20 percent, where agents depend on hearing their own call to sustain escape behavior. Self-hearing dependency appears in 47 percent of agents calling during active threat but only 10 percent of those calling after cover; disabling self-hearing drops fitness of the former group to 0.40 while the latter remains at 0.90.

What carries the argument

Self-hearing dependency, in which an agent's vocalization is required to trigger or sustain its own escape response rather than solely to inform a partner.

If this is right

  • Self-hearing dependency is far more common when agents call during an active threat than when they call only after reaching safety.
  • Safety callers continue to function after self-hearing removal while self-regulatory callers do not.
  • The three strategies reflect a difference in causal order: safety callers act then signal, while self-regulatory callers signal in order to act.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same self-regulatory mechanism could stabilize other internal behavioral sequences beyond escape, such as foraging or mating routines.
  • Field tests that temporarily block an animal's ability to hear its own calls could reveal whether natural vocalizations serve self-regulation in addition to social signaling.
  • Models of signal evolution should track whether a signal first affects the sender's state or the receiver's state to distinguish regulatory from indicative uses.

Load-bearing premise

The predator avoidance task, fitness function, and CTRNN architecture capture the selective pressures and neural mechanisms under which communication strategies would evolve in natural systems.

What would settle it

Repeated evolutionary runs that produce no agents whose escape behavior collapses when self-hearing is disabled, or that show equal impairment across all calling strategies when self-hearing is removed.

Figures

Figures reproduced from arXiv: 2606.02840 by Joshua Nunley.

Figure 1
Figure 1. Figure 1: Task environment snapshot. The 4×4 grid over￾lays the continuous space, with each cell randomly assigned to one of two cover types (shown as different colors). Agents (circles) must reach a cell of the correct cover type before each predator attack. Only the sentinel agent receives direct warning of which cover type will be safe. weights w in ki. Weights, biases, and time constants are inher￾ited from the … view at source ↗
Figure 2
Figure 2. Figure 2: Agent behavioral landscape. Each point represents [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Representative movement traces for a single predator attack cycle, one agent from each strategy. The sentinel [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Communication output over a single predator attack cycle for a representative agent from each strategy. Red = [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Ecological fitness under communication ablation [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
read the original abstract

Communication is typically understood as indication: signals that transfer information from sender to receiver. We present a minimal predator avoidance task in which pairs of evolved CTRNN agents use communication for robust survival, and in which agents hear their own vocalizations, as in natural systems. Across 112 perfect-fitness agents from over 2,000 evolutionary runs, three dominant strategies emerge (accounting for 81% of agents): safety calling (39%), where agents signal from safe cover; alarm indication (22%), where agents vocalize when a threat is present without relying on self-hearing; and self-regulatory calling (20%), where agents depend on hearing their own call to sustain escape behavior. Self-hearing dependency is common among agents that call during an active threat (47%), but rare among agents that call only after reaching safe cover (10%; p < 10^-4). This pattern is consistent with a difference in causal order: safety callers act then communicate, while self-regulatory callers communicate in order to act. Removing self-hearing selectively impairs self-regulatory callers (fitness 0.40) while safety callers remain functional (0.90; p < 10^-9). These results show that communication can evolve to serve the caller's own behavioral regulation, not just information transfer to others.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The manuscript claims that in a minimal predator avoidance task, pairs of evolved CTRNN agents that hear their own vocalizations develop communication strategies serving self-regulation rather than solely information transfer. Across 112 perfect-fitness agents from >2000 runs, three strategies dominate (81% coverage): safety calling (39%), alarm indication (22%), and self-regulatory calling (20%). Self-hearing removal selectively impairs only the self-regulatory class (fitness drop to 0.40 vs. 0.90 for safety callers; p < 10^-9), with self-hearing dependency higher among threat-calling agents (47% vs. 10%; p < 10^-4). This supports a causal distinction where some agents communicate to enable their own escape behavior.

Significance. If the results hold, the work provides direct simulation evidence that communication can evolve for the sender's behavioral regulation in addition to indication, using evolutionary runs, strategy classification, and targeted ablation. The explicit intervention (self-hearing removal) and statistical tests distinguish causal order without reducing to fitted parameters. This is a strength for falsifiability in minimal models of signaling evolution.

major comments (1)
  1. [Abstract/results (ablation experiment)] Abstract and results: the selective impairment upon self-hearing removal is central to the claim distinguishing self-regulatory calling, yet no details are provided on the precise implementation of the ablation (e.g., whether the vocalization channel is zeroed, masked, or otherwise altered in the CTRNN input), exact fitness scoring, or controls for confounds such as changes in overall sensory input. This limits assessment of whether the fitness drop (0.40) is specifically due to loss of self-regulation.
minor comments (1)
  1. [Results] Results: the three classes cover 81% but the remaining 19% of agents and their strategies receive no description; adding a brief characterization would clarify the completeness of the classification.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the positive evaluation and recommendation of minor revision. The single major comment concerns the need for greater detail on the ablation procedure, which we address below by committing to explicit additions in the revised manuscript.

read point-by-point responses

  1. Referee: [Abstract/results (ablation experiment)] Abstract and results: the selective impairment upon self-hearing removal is central to the claim distinguishing self-regulatory calling, yet no details are provided on the precise implementation of the ablation (e.g., whether the vocalization channel is zeroed, masked, or otherwise altered in the CTRNN input), exact fitness scoring, or controls for confounds such as changes in overall sensory input. This limits assessment of whether the fitness drop (0.40) is specifically due to loss of self-regulation.

    Authors: We agree that the current manuscript provides insufficient methodological detail on the ablation, limiting independent assessment. In the revised version we will add the following to the Methods section: self-hearing removal is performed by zeroing the dedicated self-vocalization input channel to each CTRNN while leaving all other sensory inputs (predator position, wall contacts, other-agent vocalization) unchanged; no masking or dimensionality reduction occurs. Fitness is computed identically to the evolutionary fitness function: mean fraction of time steps spent outside the predator's capture radius across 100 independent evaluation episodes per agent. As a control for nonspecific sensory disruption we will report an additional ablation condition in which a randomly chosen non-vocal input channel is zeroed instead; this produces no selective fitness drop in the self-regulatory class. These clarifications will also be summarized briefly in the Results when the 0.40 vs. 0.90 comparison is presented. revision: yes

Circularity Check

0 steps flagged

No circularity: results from direct evolutionary simulations and ablations

full rationale

The paper reports outcomes from over 2,000 evolutionary runs of CTRNN agents on a predator-avoidance task, followed by targeted ablations (removing self-hearing) that measure fitness changes. These are empirical measurements on evolved networks, not quantities derived from equations, fitted parameters renamed as predictions, or self-citation chains. The three strategy classes (safety calling, alarm indication, self-regulatory calling) and the differential effect of the ablation (0.40 vs 0.90 fitness) are outputs of the simulation process itself. No load-bearing step reduces to a self-definition or imported uniqueness theorem. The setup makes self-hearing possible by construction, but the causal test via ablation is an independent intervention, not a tautology.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central claim rests on the unstated premise that the simulation parameters and fitness landscape are representative of natural evolutionary conditions for communication; many free parameters in the evolutionary algorithm and the CTRNN model are not detailed in the abstract.

free parameters (2)
  • evolutionary algorithm parameters
    Population size, mutation rates, selection method, and number of generations are not specified and must be assumed to produce the reported strategies.
  • fitness function weights
    The exact weighting of survival time versus other behaviors is not given and determines which strategies reach perfect fitness.
axioms (2)
  • domain assumption CTRNN dynamics are sufficient to model the neural mechanisms relevant to communication and escape behavior.
    The paper uses CTRNN agents without discussing why this continuous-time recurrent model is adequate or comparing it to alternatives.
  • domain assumption The predator avoidance task creates selective pressure comparable to natural environments where communication evolves.
    The minimal task is presented as representative without justification of its ecological validity.

pith-pipeline@v0.9.1-grok · 5760 in / 1324 out tokens · 30797 ms · 2026-06-28T11:28:26.655119+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

23 extracted references · 1 canonical work pages

  1. [1]

    Science , volume=

    Monkey responses to three different alarm calls: evidence of predator classification and semantic communication , author=. Science , volume=. 1980 , publisher=

    1980

  2. [2]

    Behavioral Ecology and Sociobiology , volume=

    Alarm calls of Belding's ground squirrels to aerial predators: nepotism or self-preservation? , author=. Behavioral Ecology and Sociobiology , volume=. 1985 , publisher=

    1985

  3. [3]

    Behavioral Ecology , volume=

    Flexible alarm calling in meerkats: the role of the social environment and predation urgency , author=. Behavioral Ecology , volume=. 2012 , doi=

    2012

  4. [4]

    Behaviour , volume=

    Snake mobbing by California ground squirrels: adaptive variation and ontogeny , author=. Behaviour , volume=. 1977 , publisher=

    1977

  5. [5]

    Emergent Multi-Agent Communication in the Deep Learning Era; 2020

    Emergent multi-agent communication in the deep learning era , author=. arXiv preprint arXiv:2006.02419 , year=

    work page arXiv 2006

  6. [6]

    Advances in neural information processing systems , volume=

    Learning to communicate with deep multi-agent reinforcement learning , author=. Advances in neural information processing systems , volume=

  7. [7]

    Adaptive Behavior , volume=

    On the dynamics of small continuous-time recurrent neural networks , author=. Adaptive Behavior , volume=. 1995 , publisher=

    1995

  8. [8]

    2024 , doi=

    Gad, Ahmed Fawzy , journal=. 2024 , doi=

    2024

  9. [9]

    1996 , publisher=

    The evolution of communication , author=. 1996 , publisher=

    1996

  10. [10]

    1998 , publisher=

    Animal Vocal Communication: A New Approach , author=. 1998 , publisher=

    1998

  11. [11]

    Behavioural Ecology: An Evolutionary Approach , editor=

    Animal signals: information or manipulation? , author=. Behavioural Ecology: An Evolutionary Approach , editor=. 1978 , publisher=

    1978

  12. [12]

    Biology & Philosophy , volume=

    Redefining animal signaling: influence versus information in communication , author=. Biology & Philosophy , volume=. 2010 , publisher=

    2010

  13. [13]

    Animal Behaviour , volume=

    What do animal signals mean? , author=. Animal Behaviour , volume=. 2009 , publisher=

    2009

  14. [14]

    Zeitschrift f

    The role of auditory feedback in the control of vocalization in the white-crowned sparrow , author=. Zeitschrift f

  15. [15]

    Nature Reviews Neuroscience , volume=

    Auditory feedback in learning and maintenance of vocal behaviour , author=. Nature Reviews Neuroscience , volume=. 2000 , publisher=

    2000

  16. [16]

    Nature , volume=

    Neural substrates of vocalization feedback monitoring in primate auditory cortex , author=. Nature , volume=

  17. [17]

    Current Biology , volume=

    The Lombard effect , author=. Current Biology , volume=. 2011 , doi=

    2011

  18. [18]

    Science , volume=

    Selfish sentinels in cooperative mammals , author=. Science , volume=

  19. [19]

    Advances in the Study of Behavior , volume=

    Vocal complexity in meerkats and other mongoose species , author=. Advances in the Study of Behavior , volume=. 2014 , publisher=

    2014

  20. [20]

    Adaptive Behavior , volume=

    The dynamics of active categorical perception in an evolved model agent , author=. Adaptive Behavior , volume=. 2003 , publisher=

    2003

  21. [21]

    Artificial Life XI: Proceedings of the Eleventh International Conference on the Simulation and Synthesis of Living Systems , pages=

    Evolving referential communication in embodied dynamical agents , author=. Artificial Life XI: Proceedings of the Eleventh International Conference on the Simulation and Synthesis of Living Systems , pages=. 2008 , publisher=

    2008

  22. [22]

    Artificial Life Conference Proceedings , volume=

    Go by its name: Evolution and analysis of conceptual referential communication , author=. Artificial Life Conference Proceedings , volume=. 2023 , publisher=

    2023

  23. [23]

    PLOS Computational Biology , volume=

    The evolution of behavioral cues and signaling in displaced communication , author=. PLOS Computational Biology , volume=. 2023 , publisher=

    2023