Interspecific Social Information Use

• Interspecific social information use is the process by which one species exploits the behavioral cues or signals from another to inform decisions and interactions.
• The topic spans a continuum from incidental cue exploitation to evolved signaling, highlighting transitions in communication mechanisms.
• Empirical and formal models demonstrate that heterospecific information use influences foraging strategies, predator-prey dynamics, and community coexistence.

Interspecific social information use is the use by one species of information generated by another species to guide behavior, decision-making, or ecological interaction. Across the literature represented here, the topic spans heterospecific cue exploitation, cross-species communication, social-referential cue following, and community-level effects of information-mediated behavior. It includes cases in which organisms respond to incidental by-products of another species’ phenotype or behavior, as well as cases in which selection shapes the producer side of the interaction so that a cue becomes an evolved signal. It also includes formal treatments in which the behavior of other individuals functions as probabilistic evidence about hidden environmental states, and ecological models in which heterospecific information use alters coexistence conditions among competitors (Lehmann et al., 2013, Nandi et al., 23 Nov 2025, Tao et al., 13 Nov 2025).

1. Definitions and conceptual scope

A central distinction is between signals and cues. In the terminology used by Lehmann et al., signals are “an evolved means of actively conveying information and influencing the behavior of receivers,” whereas cues are “passive, non-evolving biological and environmental traits that inherently provide the observer with information” (Lehmann et al., 2013). This distinction is consequential because the producer of information may or may not have evolved to modify what receivers perceive. If a receiver exploits a correlation already present in another species’ behavior or phenotype, the interaction is cue use. If selection subsequently favors traits that alter that information channel, the system moves toward signaling or signal-like manipulation.

The scope of the term is not uniform across subfields. One strand of the literature reserves “social information use” for cases in which information derives from the behavior, performance, or choices of other organisms. On that stricter reading, predator exploitation of prey phenotype is more precisely described as interspecific information use or interspecific communication, whereas dogs responding to human referential gestures more closely matches a canonical heterospecific social-information problem (Lehmann et al., 2013, Nandi et al., 23 Nov 2025). The coexistence model of predator communities adopts yet another operational definition: interspecific information use is the use of social information acquired from heterospecifics at the same trophic level to improve foraging efficiency (Tao et al., 13 Nov 2025).

Taken together, these usages imply a continuum from incidental information to active communication. A plausible implication is that interspecific social information use is best treated as an umbrella category whose internal divisions depend on three questions: whether the informational trait is a cue or a signal, whether the information arises from phenotype or behavior, and whether its primary effect is on immediate decision-making or on longer-term ecological structure.

2. Evolution from heterospecific cues to interspecific communication

A direct treatment of how interspecific information channels originate is provided by the digital evolution study of aposematism and mimicry in a predator-prey system (Lehmann et al., 2013). The model used Avida, in which digital organisms are self-replicating computer programs with executable genomes. Offspring inherited genomes with mutation: each offspring had a $0.25$ probability of a single instruction substitution and $0.05$ probability each of a single insertion or deletion. The ecological world was a $251 \times 251$ grid; resources replenished at $0.01$ units per cell per update up to $1$ unit per cell; all organisms required $10$ resource units to reproduce.

The prey community comprised three classes assigned at birth in fixed ratios: 50% poisonous, 25% safe, and 25% potential mimics. Predators evolved rather than being pre-specified; an organism became a predator when an attack instruction mutated into its genome and it successfully killed another organism. Predators could attack generally or execute class-specific attacks targeting non-poisonous, mimic, or poisonous prey. Mimics were edible like non-poisonous prey, but could alter their displayed phenotype to resemble either poisonous or non-poisonous prey. The main manipulation was poison efficacy, with poison levels of $0.01,\ 0.025,\ 0.05,\ 0.1,\ 0.2,\ 0.4,\ 0.75,\ 1.0$; populations evolved for 500,000 updates, and some treatments also imposed imperfect mimicry, such that the intended mimic display was perceived only 25% of the time (Lehmann et al., 2013).

The central result was sequential. First, predators evolved cue recognition only when the cost of consuming poisonous prey was sufficiently high. Below poison level 0.1, selection to avoid poisonous prey was weak and predators usually did not evolve selective predation. At and above 0.1, predators “nearly universally” evolved to avoid poisonous prey, and the behavior fixed in all trial populations for poison levels at and above 0.4. Second, active signaling through mimicry evolved under every condition that produced cue utilization. Mimics showed no strong display preference when predators had little reason to avoid poisonous prey, but at higher poison levels they preferentially displayed the poisonous phenotype. Third, this structure persisted even under imperfect and dishonest signaling: at poison levels $0.1,\ 0.2,\ 0.4$, the ratio of poisonous mimicry under 25% accuracy to 100% accuracy was $0.71,\ 0.83,\ 0.81$, whereas at $0.75$ and $0.05$0 it rose to $0.05$1 and $0.05$2 (Lehmann et al., 2013).

This sequence supports the paper’s claim that cues can serve as “stepping-stones” to more elaborate interspecific communication systems. The visible prey class initially functions as a passive informational feature. Once predators evolve to use it, selection favors prey that manipulate it, including dishonest Batesian-like mimics. The broader significance is that receiver-side information use can create sender-side selection, and that reliable interspecific information exchange need not disappear in the presence of deception if the cost of receiver error is sufficiently high.

3. Heterospecific cue interpretation, redundancy, and hidden-state inference

A field study of Indian free-ranging dogs examined a different but complementary form of interspecific social information use: human cue interpretation in a food retrieval task (Nandi et al., 23 Nov 2025). The subjects were 352 randomly selected adult free-ranging dogs from urban and semi-urban sites in West Bengal, India. The object-choice task used six cue conditions: control (no cue), negative control (one baited bowl, no cue), combined pointing and gazing, pointing-only, gazing-only, and conflicting cues in which pointing and gaze were directed at opposite bowls. The dogs chose correctly only in the combined pointing-plus-gazing condition. In that condition, preference for the correct bowl was significant, with $0.05$3, $0.05$4, $0.05$5, Cohen’s $0.05$6, 95% CI $0.05$7 to $0.05$8. Pointing-only, gazing-only, and conflicting cues did not differ significantly from chance, and the generalized linear model identified only the combined cue as a significant positive predictor of correct choice, with $0.05$9, $251 \times 251$0, $251 \times 251$1, $251 \times 251$2, corresponding to $251 \times 251$3 to $251 \times 251$4 (Nandi et al., 23 Nov 2025).

The same study showed that cue use and willingness to engage were dissociable. Demeanor affected familiarization success and approach latency: affiliative dogs were more likely to succeed overall and approached more rapidly than anxious or neutral dogs. However, demeanor did not significantly affect accuracy in the correct-choice model, and sex had no detectable effect on familiarization success, latency, test latency, or choice accuracy (Nandi et al., 23 Nov 2025). The interpretation offered in the paper is that interspecific communication in this population depends on signal redundancy and clarity. A point provides a directional vector; gaze supplies attentional or social-intent information; their conjunction produces a more coherent heterospecific cue than either modality alone. The failure of the conflicting-cue condition to bias choice toward either the pointed-at or gazed-at bowl indicates that the dogs did not resolve ambiguity by consistently privileging one channel.

A mechanistic analogue for such heterospecific inference is provided by a study of human social learning when rewards are not observable (Hawkins et al., 2022). In that framework, observers infer hidden reward states from publicly visible trajectories. The formal core is Bayesian: $251 \times 251$5 where $251 \times 251$6 is a hidden state and $251 \times 251$7 is observed behavior. In the collective sensing task, observers saw the trajectories $251 \times 251$8 of others but not their rewards, and inferred quantities such as $251 \times 251$9. Across three experiments, larger human groups outperformed heuristic models that merely moved toward the center of others or copied indiscriminately, and participants selectively copied exploiting-looking targets when they themselves were unrewarded. In Experiment 2, that selective copying effect yielded $0.01$0, $0.01$1, $0.01$2, 95% CI $0.01$3. Experiment 3 further showed that informative trajectories need not be fixed cues like stopping; in a noisier continuous landscape, exploiting often appeared as spinning or tight local motion, and the social benefit weakened as noise increased (Hawkins et al., 2022).

This human work is not direct interspecific evidence, but it clarifies a general mechanism: observers can treat visible behavior as probabilistic evidence about hidden success without requiring intentional signaling. This suggests that many heterospecific interactions may depend less on copying a visible payoff and more on inferring likely payoff from task-specific behavioral signatures.

4. Formal models of information integration and observation structure

A major theoretical question is whether apparently different social decision rules are genuinely different mechanisms or parameterizations of a shared inferential process. A Bayesian model of collective decision-making across zebrafish, Argentine ants, and sticklebacks derives the probability that an option $0.01$4 is good from non-social information $0.01$5 and social information $0.01$6: $0.01$7 with

$0.01$8

Under the approximation that observed behaviors are independent, the social term becomes

$0.01$9

where $1$0 is the reliability of behavior $1$1 as a cue that $1$2 is good. In the symmetric two-option case, this reduces to

$1$3

with an analogous expression for $1$4. Probability matching then gives a unified choice rule $1$5 as a function of private information $1$6, social reliability $1$7, and counts $1$8, modulated by $1$9 (Arganda et al., 2012).

The key comparative result is that dependence on absolute differences, Weber-like relative differences, and more complex nonlinear response surfaces can all emerge from this single framework. Zebrafish were best fit by $10$0, corresponding to a rapid saturation sometimes described as “counting up to three.” Argentine ants required a noise-extended rule with $10$1, placing them in a Weber-like regime. Sticklebacks behaved as if using an effective $10$2 rule in the tested range, consistent with $10$3 or with low-$10$4 approximations for other $10$5 values (Arganda et al., 2012). Although the empirical applications are conspecific, the formalism already allows multiple behavioral classes with distinct reliabilities $10$6, so heterospecific sources can, in principle, be represented as separate cue classes.

A related normative model asks how these decision rules change when observation is sparse rather than all-to-all (Mann, 2021). Agents make sequential binary choices with utility difference

$10$7

receive a noisy private cue $10$8, and observe only a subset of previous decisions through a random directed network with observation probability $10$9. The paper distinguishes $0.01,\ 0.025,\ 0.05,\ 0.1,\ 0.2,\ 0.4,\ 0.75,\ 1.0$0, the connectivity to which agents are adapted, from $0.01,\ 0.025,\ 0.05,\ 0.1,\ 0.2,\ 0.4,\ 0.75,\ 1.0$1, the connectivity at the time of the decision. Exact inference with partial observation is formally complex, but the main result is that, under sufficiently sparse connectivity, rationally adapted agents can use a “highly-simplified social information” strategy based mainly on

$0.01,\ 0.025,\ 0.05,\ 0.1,\ 0.2,\ 0.4,\ 0.75,\ 1.0$2

In that sparse regime, order becomes nearly irrelevant, observed decisions are treated as effectively quasi-independent, and a sigmoidal rule of the form

$0.01,\ 0.025,\ 0.05,\ 0.1,\ 0.2,\ 0.4,\ 0.75,\ 1.0$3

closely approximates the optimal response (Mann, 2021).

These models are directly relevant to interspecific social information use in a formal rather than empirical sense. They imply that heterospecific decisions can be treated as evidence if their reliability is parameterizable, but they also show what must be added for genuine interspecific analysis: species-specific cue reliabilities, asymmetric preference alignment, and asymmetric observation probabilities. The sparse-network model makes this explicit by proposing extensions such as a species-indexed observation matrix $0.01,\ 0.025,\ 0.05,\ 0.1,\ 0.2,\ 0.4,\ 0.75,\ 1.0$4 and a weighted difference statistic

$0.01,\ 0.025,\ 0.05,\ 0.1,\ 0.2,\ 0.4,\ 0.75,\ 1.0$5

when identity matters more than raw counts (Mann, 2021).

5. Ecological and community-level consequences

Interspecific social information use is not confined to immediate decision problems; it can alter coexistence conditions in ecological communities. A recent consumer-resource model shows that interspecific social information use among predators can relax the standard steady-state constraint associated with the competitive exclusion principle (CEP) (Tao et al., 13 Nov 2025). The system contains two consumers, $0.01,\ 0.025,\ 0.05,\ 0.1,\ 0.2,\ 0.4,\ 0.75,\ 1.0$6 and $0.01,\ 0.025,\ 0.05,\ 0.1,\ 0.2,\ 0.4,\ 0.75,\ 1.0$7, and one shared resource, $0.01,\ 0.025,\ 0.05,\ 0.1,\ 0.2,\ 0.4,\ 0.75,\ 1.0$8. Predation is modeled through explicit pair formation: $0.01,\ 0.025,\ 0.05,\ 0.1,\ 0.2,\ 0.4,\ 0.75,\ 1.0$9 with $0.1,\ 0.2,\ 0.4$0 and $0.1,\ 0.2,\ 0.4$1. The innovation is asymmetric information use: $0.1,\ 0.2,\ 0.4$2 uses information from $0.1,\ 0.2,\ 0.4$3, and this increases its encounter rate through a saturating Monod-type function,

$0.1,\ 0.2,\ 0.4$4

This modification changes the geometry of coexistence. Without information use, positive equilibrium requires

$0.1,\ 0.2,\ 0.4$5

so both consumer zero-growth conditions depend on the same scalar variable $0.1,\ 0.2,\ 0.4$6. With information use, the first condition becomes

$0.1,\ 0.2,\ 0.4$7

whereas the second remains $0.1,\ 0.2,\ 0.4$8. The effect is that one consumer’s growth depends not only on the shared resource but also on the abundance of a competitor that serves as an information source. In the authors’ interpretation, $0.1,\ 0.2,\ 0.4$9 becomes an “information resource” for $0.71,\ 0.83,\ 0.81$0, even though $0.71,\ 0.83,\ 0.81$1 remains the only biomass resource (Tao et al., 13 Nov 2025).

The reported consequences are substantial. Without interspecific information use, the one-resource/two-consumer system does not have stable constant-density coexistence. With information use, a finite region of parameter space supports coexistence. For abiotic resources, coexistence appears as a globally attractive stable fixed point. For biotic resources, coexistence can be either a globally stable fixed point or an unstable equilibrium surrounded by a stable limit cycle. The supplement provides explicit steady-state expressions, including

$0.71,\ 0.83,\ 0.81$2

and a closed-form expression for equilibrium $0.71,\ 0.83,\ 0.81$3 in terms of $0.71,\ 0.83,\ 0.81$4 (Tao et al., 13 Nov 2025).

The paper also presents quantitative matches to systems interpreted as possible information-mediated coexistence anomalies. For Black-legged Kittiwakes and Murres, using a biotic-resource parameterization, reported relative root mean square errors were $0.71,\ 0.83,\ 0.81$5 and $0.71,\ 0.83,\ 0.81$6, and the pooled comparison of observed versus predicted time-averaged relative abundances yielded Pearson correlations of $0.71,\ 0.83,\ 0.81$7 for the SSA and $0.71,\ 0.83,\ 0.81$8 for the ODE. For Tribolium-like and Drosophila serrata abiotic-resource cases, reported rRMSE values were $0.71,\ 0.83,\ 0.81$9 and $0.75$0, respectively (Tao et al., 13 Nov 2025). These fits are presented as suggestive evidence that heterospecific information transfer can function as a non-trophic facilitation mechanism among competitors.

6. Limitations, controversies, and unresolved issues

Several conceptual and methodological cautions recur across this literature. First, not all interspecific information use is social information use in the strict sense. Predator exploitation of prey phenotype is better described by the language of cues, signals, recognition, and communication than by the narrower social-information literature, even though the underlying logic is closely related (Lehmann et al., 2013). The scope of the term therefore remains partly discipline-dependent.

Second, proof-of-principle demonstrations and calibrated biological prediction are not the same. The digital evolution study is explicitly abstract: prey classes are assigned at birth in fixed proportions, signal space is discrete, predator discrimination is genetically encoded rather than learned within lifetimes, and the recognition metric infers avoidance from a final abundance threshold $0.75$1 chosen to reduce noise. The study therefore supports evolutionary possibility and mechanism, not quantitative prediction for any particular natural mimicry system (Lehmann et al., 2013).

Third, direct evidence for heterospecific cue interpretation does not by itself resolve developmental debates. The free-ranging dog study argues that success under the clear multimodal condition supports a species-level domestication effect and “disproves the enculturation hypothesis,” but the same paper also acknowledges that these dogs live in human-dense environments and have extensive lifetime exposure to humans. The stronger claim should therefore be treated cautiously: the data show successful use of clear human cues, but they do not cleanly isolate inherited predisposition from developmental experience (Nandi et al., 23 Nov 2025).

Fourth, several influential formal models remain empirically conspecific even when mathematically extendable to heterospecific contexts. The Bayesian collective-decision model and the sparse-network rational-choice model provide scaffolds for interspecific analysis, but neither incorporates species recognition, strategic unreliability, or the full dependencies that can arise when one species systematically conditions on another (Arganda et al., 2012, Mann, 2021). Their interspecific relevance is therefore strongest at the level of transferable inferential structure.

Fifth, the coexistence model treats information use phenomenologically. The effect is inserted into $0.75$2 as a saturating enhancement of encounter rate, while signal generation, transmission, detectability, and cost are not modeled explicitly. The empirical support is post hoc and correlational rather than experimental, and alternative explanations such as hidden niche differences, spatial structure, or temporal variation are not ruled out rigorously (Tao et al., 13 Nov 2025). Likewise, the hidden-reward social inference study provides a strong computational mechanism for using behavior as a proxy for unobservable success, but its experiments involve humans and artificial agents rather than heterospecific animal systems (Hawkins et al., 2022).

These limitations do not weaken the central theme so much as delimit its current state. The convergent result across these papers is that interspecific information use is neither a single mechanism nor a single level of analysis. It ranges from heterospecific gesture reading and cue redundancy, through evolutionary transitions from cue exploitation to signal manipulation, to Bayesian and network-constrained inference, and finally to community-level coexistence effects. A plausible implication is that future progress will depend on linking these levels more explicitly: measuring cue reliability and ambiguity in real mixed-species systems, modeling source-specific trust and observation structure, and experimentally manipulating heterospecific information channels to separate information-mediated effects from trophic or environmental confounds.