For a semiotic of opacity: the role of biosemiotics between AI and animal communication

1 Introduction

In recent years, the idea that artificial intelligence could serve as a bridge between humans and nonhuman animals has sparked renewed excitement in both scientific and media circles. The notion of “translating” the vocalizations of dolphins, sperm whales, birds, or primates into language-like forms that humans can interpret is no longer presented as mere speculation (Hagiwara et al. 2024; Rutz et al. 2023). It is now framed as a concrete promise, made plausible by the massive computational power of neural networks and generative models. Projects such as CETI (Cetacean Translation Initiative), DolphinGemma, and NatureLM aim to collect vast amounts of bioacoustic data, segment it, map it, and ultimately convert it into linguistic structures that can be analyzed and translated. This momentum is driven by a strong assumption: that animal communication is rule-governed and therefore potentially decodable using statistical tools and syntactic models.

However, Thomas A. Sebeok (1965, 1972, 1977), the founder of zoosemiotics (in an exchange with Rulon Wells), already clarified the conceptual misunderstanding underlying this hypothesis. On the one hand, Sebeok drew a sharp distinction between human language and animal communication, emphasizing that the latter is not based on arbitrary symbolic systems but on bodily, indexical, and iconic signals oriented toward action rather than the transmission of propositional content.^[1] On the other hand, he showed how any attempt to “translate” animal communication into human linguistic categories involves an epistemological projection: a forced equivalence that conflates sign with code, semiosis with computation, and embodied meaning with disembodied syntax.

This issue becomes even more pressing when viewed through a biosemiotic lens, especially in light of Jakob von Uexküll’s (1909) theory of the Umwelt, later developed by Jesper Hoffmeyer (2004) and Kalevi Kull (2010). Every organism lives within its own meaningful world, selectively constructed according to its perceptual, affective, relational, and ecological capacities. Interpreting signs cannot be separated from this situated horizon: an animal sign only has meaning within its Umwelt and loses its intelligibility when extracted from the bodily and environmental context in which it is produced. Artificial intelligence, no matter how powerful, does not possess an Umwelt^[2] (Emmeche 2001: 676). It can correlate data but cannot interpret experiences grounded in the lifeworld of animals or their ecological existence. It can model patterns, but it cannot inhabit worlds. Its engagement with animal communication is thus inevitably extra-semiotic, unable to grasp the situated, affective, and embodied value of living signs.

In this framework, the concept of opacity acquires a crucial theoretical role. Every act of translation is, as is well known, a form of betrayal: in the attempt to make what is other transparent, one ends up flattening the plurality of possible meanings into a symbolic and computational linearity. AI, by re-encoding animal vocal signs into textual or symbolic strings (Robinson et al. 2025), performs a semiotic manipulation that reduces animal otherness to a computable object. The result is an impoverishment of animal semiosis, which is represented as language only insofar as it is deformed according to algorithmic criteria.

This article aims to approach opacity not as an obstacle, but as an epistemological and semiotic resource. Far from treating it as a failure of access, opacity is here understood as an expression of the semiotic freedom of living beings: the capacity to generate meaning beyond any attempt at formalization. Accordingly, this paper has two interconnected aims: (1) to critically assess recent AI-based projects attempting to “translate” animal communication – such as CETI, NatureLM, and ISPA – by highlighting how they often reduce semiosis to formalizable and referential codes; and (2) to propose a biosemiotic reconceptualization of interspecies communication through the notion of semiotic opacity. Drawing on core concepts such as Umwelt, semiotic freedom, and biotranslation, I argue that opacity is not a communicative failure but a constitutive condition of meaning in living systems. Rather than seeking full translatability, I suggest reframing the role of AI within a relational interface (between humans, animals, and machines) based on attentiveness, interpretive partiality, and ecological responsiveness. Biosemiotics thus emerges as a critical framework for resisting the epistemological flattening of animal alterity and for theorizing mediation not as linguistic equivalence, but as an ethics of difference.

2 Translating the untranslatable: AI and the semiotic misreading of animal communication

Within the expanding field of so-called conservation AI, a growing number of ambitious projects aim to “translate” (hidden under the term “decode”) animal communication through artificial intelligence. The promise is bold: to develop algorithmic interfaces capable of making animal vocalizations understandable, segmentable, and “speakable” according to linguistic models compatible with human interpretation (Andreas et al. 2022). This vision is embodied in initiatives such as the Cetacean Translation Initiative (CETI), Google’s DolphinGemma, NatureLM-audio, and the proposal for an Inter-Species Phonetic Alphabet (ISPA). Although they differ in methods and outcomes, these projects share a common goal: to assign structure, meaning, and translatability to nonhuman communicative systems using machine-learning technologies.

The CETI project, among the most sophisticated and visible, focuses on studying the communication system of sperm whales (Physeter macrocephalus), considered organisms with exceptional social, neuroanatomical, and vocal complexity (Andreas et al. 2022). The research plan involves collecting vast quantities of multimodal data (bioacoustic signals, behavioral, environmental, and biological information) through underwater sensors, drones, acoustic tags, and robotic systems (Bermant et al. 2019). The methodological core lies in deep learning models and unsupervised learning techniques adapted from natural language processing (NLP), used to identify recurring patterns in sperm whale clicks (so-called “codas”) and to derive a phonological, morphological, and syntactic structure analogous to that of human languages (Beguš et al. 2025; Goldwasser et al. 2023).

Similarly, NatureLM-audio, developed by the Earth Species Project, offers a generalist approach to bioacoustics through audio-language foundation models. Trained on large acoustic datasets (from birds, cetaceans, bats, and insects), NatureLM combines audio and text to generate shared representations that, according to its developers, can classify species, predict behaviors, and generate linguistic descriptions of animal vocalizations (Robinson et al. 2025). The innovation of this model lies in its ambition for cross-taxonomic generalization, that is, the ability to transfer learning across domains (from birdsong to cetacean sounds) and across tasks (from classification to captioning). Its architecture fits within the paradigm of large audio-language models (LALMs), inspired by analogous advances in the human domain (such as AudioPaLM and SpeechGPT).

A third project, the previously mentioned ISPA, also developed by the Earth Species Project, aims to create a standardized transcription system for animal sounds in textual form. The idea is to overcome the limitations of continuous representations (such as spectrograms) by converting them into discrete tokens, inspired by the International Phonetic Alphabet (IPA), thus applying NLP paradigms to animal acoustic signals (Hagiwara et al. 2024). Although formally elegant, this approach entails a radical reification of living sound, transforming an animal’s song into an alphanumeric sequence to be processed by a linguistic model, stripped of its context (the so-called “cocktail party problem”) and its ecological function, in order to extract patterns.

Starting from this brief overview, we can now explore how a semiotic methodology may be employed to analyze and reveal the epistemological limitations of projects that aim to “translate” other species.

The underlying logic shared by all these initiatives is one of segmentation, abstraction, and symbolic modeling: signals are collected, broken down into minimal units, labeled, correlated with observable behavioral contexts, and finally embedded in predictive models. This process follows a well-defined epistemic strategy, namely treating animal communication as a foreign language yet to be deciphered, as explicitly stated in CHAT (Cetacean Hearing Augmentation Telemetry), a subproject of DolphinGemma aimed at establishing communication with dolphins. Here, nonhuman vocalization is framed like an ancient language, awaiting reconstruction of its grammar and lexicon.

The analogy with human linguistics is clear. CETI, for instance, seeks to identify a sperm whale phonetic inventory, define a coda syntax, and eventually establish a functional semantics to be tested via playback experiments, i.e. experiments in which AI-generated signals are played back to animals to observe their behavioral reactions (Andreas et al. 2022).

Further evidence of this approach is found in Google’s launch of DolphinGemma, an AI model developed in collaboration with the Wild Dolphin Project (WDP) and the Georgia Institute of Technology, with the goal of decoding dolphin communication. DolphinGemma is based on Google’s language model architecture, specifically adapted for audio data analysis. It uses the SoundStream tokenizer to convert dolphin vocalizations into machine-readable sequences, enabling the model to identify recurring patterns and predict the next likely sound in a sequence. With approximately 400 million parameters, the model is optimized for use on mobile devices such as Google Pixel smartphones, which WDP researchers carry during field expeditions.

The WDP has compiled one of the world’s most comprehensive datasets on dolphin behavior and vocalization, collecting audio and video recordings linked to individual dolphins, their social relationships, and observed behaviors. This extensive dataset has enabled the mapping of specific sound types (such as signature whistles, burst-pulse sounds, and clicks) to corresponding behavioral contexts. DolphinGemma allows basic symbolic interaction between humans and dolphins through synthetic whistles associated with specific objects like seaweed or toys. When dolphins imitate these synthetic sounds, researchers interpret the imitation as a request for the associated object. This integration improves the accuracy of sound recognition and speeds up responses during underwater interactions, crucial for real-time communication.

Google has announced plans to release DolphinGemma as an open-source model in the summer of 2025 (Herzing and Starner 2025). Although currently trained on vocalizations of Atlantic spotted dolphins, the model can be adapted to other species, potentially expanding research on cetacean communication.

However, as several critical voices within the conservation field have pointed out, this “linguistification” of animal communication risks generating more of an anthropocentric illusion than genuine understanding (Hauser 2000; Ryan and Bossert 2024; Yovel and Rechavi 2023). Treating animal vocalizations as linguistic sequences implies a communication model centered on transparency, referentiality, and repeatability, whereas animal semiosis is often situated, opaque, expressive, and affective. Moreover, the ethical dimension is far from secondary: attempting to “talk to whales” (or imitate “whalish”^[3] language structure) through machine-generated signals introduces potential emotional and cultural impacts on the animals themselves beyond the boundaries of scientific observation (Singer and Tse 2023).

It is important to note that the tendency to reduce animal communication to codifiable systems precedes the emergence of AI. Earlier quantitative studies in bioacoustics and information theory, such as those by Ferrer-i-Cancho and Lusseau (2006, 2009) and Hernández-Fernández and Torre (2022), applied compression principles and Zipf’s law to model animal signals in terms of efficiency and brevity. While these studies offered insights into structural regularities in animal vocalizations, they often overlooked the contextual, ecological, and affective dimensions of meaning emphasized in biosemiotic theory. AI inherits these same assumptions, extending the metaphor of language-as-code into machine-readable form, thus reinforcing rather than challenging the semiotic reductionism embedded in earlier approaches. As argued by Bolshoy and Lacková (2021), this reductionism reflects a broader epistemological illusion shared by both modern linguistics and evolutionary biology, namely, the belief that communication systems can be fully explained through abstract, codifiable models. Their critique shows how such approaches, including many current AI paradigms, overlook the situated, embodied, and affective dimensions of meaning, projecting an illusory transparency onto complex semiotic processes.

Taken together, these projects mark a critical moment for rethinking the role of artificial intelligence in understanding life, and conversely, for reconsidering the role of biosemiotics in engaging with emerging technological trends. On the one hand, they offer the promise of new knowledge and tools for conservation (Bromley 2023); on the other, they reveal the inherent limitations of any attempt to reduce animal semiosis to computable language. It is precisely within this tension that the central theme of this article is situated: the opacity of animal communication not as a technical flaw, but as a sign of semiotic freedom.

3 Opacity as semiotic freedom: toward a theory of resistant meaning

Projects aiming to translate animal vocalizations into a human-comprehensible language through artificial intelligence technologies reveal a number of critical limitations when examined through a semiotic and biosemiotic lens. At their core lies an epistemological reduction: the assumption that animal communication can be formalized into a code, segmented into symbolic units, and mapped onto human linguistic structures. This assumption neglects the very nature of semiosis as a situated, relational, and embodied process.

One of the central theoretical omissions of these AI-driven frameworks is their disregard for the organism’s Umwelt (Uexküll 1909), the species-specific perceptual and meaning-constructing world in which every sign is embedded. Biosemiotics has repeatedly emphasized that signs do not exist in isolation, suspended in a vacuum. They become meaningful only within the dynamic interaction between organism and environment. An algorithmic model trained on spectral features and behavioral correlations may be able to recognize patterns, but it does not inhabit the world in which those patterns acquire meaning. It remains, in fundamental terms, blind to the relevance of the sign within the organism’s lifeworld.

Moreover, these projects often presuppose that communication is reducible to referential content. They seek mappings between vocalizations and stable meanings (objects, actions, intentions) that reflect the denotative logic of symbolic language. Yet, as we have seen, animal communication is frequently indexical or iconic, and its function is not simply to transmit information, but to modulate behavior, emotions, and social relationships. Attempts to translate these forms of semiosis into language-like structures overlook the expressive, affective, and performative complexity of animal signs.

Translation cannot be conceived as a simple transposition of content from one system to another, but rather as a complex semiotic process that is structurally unstable (Petrilli 2014: 249). Umberto Eco insisted that translation always means “saying almost the same thing” (Eco 2003: 29–32), emphasizing that every act of translation involves a loss of information, an interpretive reformulation, and a contextual adaptation. Already in The open work (1962), Eco demonstrated that every act of interpretation, and thus every translation, involves a selection of meaning from within a semantic multiplicity that can never be entirely preserved. Translation is therefore a form of negotiation between horizons, an attempt to maintain the original semiotic function while acknowledging the inevitability of transformation.

This perspective is significantly expanded in the biosemiotic paradigm developed by Kalevi Kull and Peeter Torop. Torop’s work in cultural semiotics further reinforces this approach by highlighting how translation is always historically and ideologically situated. In dialogue with Kull, he emphasizes that translation not only mediates between texts or languages, but also transforms the receiving system itself. In the case of biotranslation (Kull and Torop 2003), the receiving Umwelt is not a passive recipient, but an active interpreter that reconfigures the sign according to its own perceptual and functional logic. This makes biotranslation not a linguistic operation but a semiotic adaptation: a form of meaning-making deeply embedded in the lifeworld of organisms. With the term “biotranslation,” the authors propose to extend the notion of translation beyond the domain of human language, defining it as a “transmission between Umwelten” (Kull and Torop 2003: 320). Every organism lives within its own perceptual–semiotic world (Umwelt), and the possibility of communication or understanding between different species entails a translational process of biological signs. Kull introduces a distinction between eutranslation (conscious translation, characteristic of human beings) and biotranslation (biological translation, non-intentional), arguing that every meaningful interaction between organisms can be conceived in terms of intra- or interspecific translation. In this sense, translation is not merely a textual practice, but a structural principle of life itself, a continuous process of semiotic adaptation and transformation across heterogeneous systems.

The transposition between Umwelten, however, is never neutral, as every biological translation involves a readjustment of the sign to the perceptual and functional structure of the receiver. Roman Jakobson (1959), in his influential essay On linguistic aspects of translation, identified three forms of translation: intralingual (within the same language), interlingual (between languages), and intersemiotic (transmutation), defined as “an interpretation of verbal signs by means of signs of nonverbal sign systems” (Jakobson 1959: 233). This intersemiotic process typically moves from verbal to nonverbal (e.g., novel to film). Such transpositions highlight partial incommensurability between expressive planes, yet allow some content translatability. Project CETI, however, inverts this logic. Instead of interpreting verbal signs via nonverbal ones, it uses natural language processing to interpret nonverbal sperm whale vocalizations (codas), re-inscribing them into a verbal-semiotic framework. This is not Jakobson’s intersemiotic translation but its inverse: a retro-projection of human language structure onto a nonhuman form, treating vocalizations as translatable utterances with latent syntax. CETI thus enacts a formal recoding of whale communication based on human linguistic grammar, rather than a true transmutation. This approach overlooks that even human language translation is never fully isomorphic, involving interpretive processes across differing semantic, pragmatic, and cultural frameworks. Where emotive, phatic, or performative dimensions dominate (Jakobson 1960: 356), meaning cannot be abstracted from its embodied, situated expression. This holds even more true across species, where meaning is deeply tied to embodiment, context, and species-specific perceptual and ecological constraints. This discrepancy points to a fundamental misunderstanding of what translation entails when extended to nonhuman systems.^[5] As Eco (2003) emphasized, fidelity does not lie in literal reproduction, but in the preservation of the effect on the recipient; similarly, for Kull and Torop (2003), a translation occurs only if the transmitted sign activates a Funktionskreis within the new Umwelt, thereby generating corresponding behaviors. Translation is therefore only possible when there is a partial overlap of perceptual systems and response codes. Its precondition is not transparency, but functional compatibility.

The use of artificial intelligence to translate animal signals occupies a problematic intersection between these two perspectives. On the one hand, it employs algorithms inspired by logotranslation, that is, a eutranslational model based on symbolic correspondences and computable syntactic structures. On the other hand, it assumes that translation across radically different semiotic worlds (such as the human and the animal) can be fully achieved through statistical models. But this amounts to ignoring the embodied, affective, situated, and historically layered specificities of animal signs.

Such a strategy risks producing a simulation of translation rather than an actual translation, because, as both Eco and Kull remind us, authentic translation requires semiotic intelligence capable of grasping not only forms, but also rhythms, intonations, intentions, and the vital codes embedded within signs. Artificial intelligence, lacking both an Umwelt and embodied experience, does not interpret; it classifies. And what is classified is never the entirety of a message, but only that portion which is computationally visible. AI may detect acoustic patterns in a dolphin’s whistles, but it cannot apprehend their pragmatic and relational dimension within a shared environment.

The biosemiotic theory of biotranslation and Eco’s general semiotics converge in affirming that translation is always a relational, incomplete process, deeply conditioned by the structure of the subject and the context. Applying AI to animal translation without acknowledging this complexity risks generating an epistemic reduction, mistaking technical transduction for understanding, and statistical correlation for meaning. To overcome these limitations, artificial intelligence must not be conceived as a translator, but as a potential environment of mediation, one whose role is to amplify the conditions of listening, not to flatten semiotic^[6] (Zengiaro 2024a) diversity into the model of human language.

4 Beyond translation: biosemiotics of interspecies interfaces

The discussion developed in the previous sections has shown that current attempts to apply artificial intelligence to the “translation” of animal communication rest on fragile epistemological assumptions and a reductionist view of semiosis. These projects assume that they will be able to break down the interspecies barrier through an algorithmic interface, but the result is often in danger of becoming a simulation of communication rather than its genuine expansion.

This failure is not merely technical, but conceptual. Life, as a semiotic process, is irreducible to a closed system of correspondences between signs and meanings. Each organism inhabits its own Umwelt, a meaningful world that is specific, layered, affective, and embodied, and this cannot be understood without engaging with its unique relational dynamics. Any interface that ignores this relational dimension ends up projecting its own anthropotechnical categories onto the animal, resulting in a form of semiotic ideology, where the nonhuman is considered to be “speaking” only to the extent that it says something intelligible to us.

It is at this critical juncture that biosemiotics emerges not only as an alternative methodology, but as a necessary one. Its strength does not lie in proposing yet another algorithm, but in changing the interpretive framework altogether. Instead of focusing on a semantics of decoding, biosemiotics advocates for a pragmatics of relation. It does not ask: “What does this sign mean?”, but rather: “In what relational context does this semiotic act occur? What kind of response does it elicit? What kind of world does it help to construct?”

Moreover, ecosemiotics extends this approach by helping us understand how communication channels themselves are altered over time. It asks how environmental factors, such as rising temperatures, ocean acidification, and biodiversity loss, transform the expressive and pragmatic conditions of animal semiosis. In this light, a semiotic ethics of interspecies interaction must also be an ecological ethics, attentive not only to the signs that are produced, but to the material and environmental infrastructures that make them possible, audible, and meaningful.

This reversal is crucial for rethinking the role of AI itself. Rather than treating the machine as a tool for making animal communication transparent, we can conceive of it as a sensitive mediator, an amplifier of our capacity to listen, provided it is guided by an ecological rather than extractive logic. Applied biosemiotics can provide the criteria for designing interfaces that do not eliminate opacity but respect it; that do not translate, but accompany interaction; that do not normalize, but amplify difference.

In this sense, biosemiotics becomes a bridging science, capable of orienting a new paradigm of semiotic cohabitation between humans, non-humans, and machines. It offers theoretical tools for recognizing life as a plural, non-linear, and open communicative process, as well as ethical tools for envisioning a form of technological interaction that is not grounded in prediction and control, but in attentiveness, interpretive suspension, and responsive listening.

The outcomes of this analysis are clear. First, the idea of translating animal language through AI is methodologically misguided, as it conflates correlation with meaning. Second, semiosis is not a linear operation, but a situated event in which opacity is not an error, but a condition for the emergence of meaning. Third, biosemiotics, precisely because it is founded on the relationship between life and sign, can guide a radical reconceptualization of technological interfaces, shifting them away from the paradigm of transparency and toward a logic of encounter.

The future of interaction between AI and the nonhuman does not depend on the ideological perfection of translation, but on the quality of listening and interpretive engagement. The interface of the future will not be a transparent window, but an opaque threshold, traversable only to the extent that we accept the irreducibility of the other, be it animal, machine, or environment, and accompany it in a shared process of meaning-making.

In one of his contributions, Leone (2023: 41) expands this diagnosis into a broader reflection on the semiotics of artificial intelligence, arguing that the task of the semiotician today is not only to expose the limitations of algorithmic simulation, but to understand AI as a new form of semiotic agency. AI, he suggests, does not produce truth in the classical sense, but rather effects of credibility: it is a rhetorical machine, a symbolic operator that selects, reifies, and structures horizons of meaning based on what it can compute, thus actively shaping our perception of the world (Leone 2023: 43).

Integrating this reading into our discussion, Leone offers a compelling proposition: we should not reject AI as a deceptive simulacrum, but rather employ it as a differential revealer, a technology that, precisely through its limitations, signals what lies beyond computation: the undecidable, the ambiguous, the embodied, the relational.

In this vision, AI can become an ally in the construction of a critical metasemiosis, one capable of redirecting our attention to what remains opaque and, for that very reason, demands to be questioned. Rather than delivering a faithful translation of living beings, AI can provide an index of our projections, our biases, and the epistemic blind spots that permeate our relationship with animal alterity. It is within this zone of tension that a new ecotechnical semiotics might emerge as a practice of differential listening that coexists with alterity without absorbing it, and that accepts the untranslatable not as a failure, but as the very condition of every authentic semiotic relationship.

This scenario is further reinforced by the concept of the ontological gap: between the internal representation of reality constructed by AI and that generated by living beings, there exists a radical epistemological fracture. AI does not inhabit the world, has no biological needs, and does not produce intentional judgments. As a result, any attempt to make AI “speak” on behalf of nonhuman life risks generating a synthetic ontology, i.e. one that translates while simultaneously erasing the semiotic specificity of the other. A biosemiotic approach to embodiment, environment, intentionality, and purpose can partially bridge this divide. However, such an approach necessarily requires abandoning the notion of direct translation in favor of a paradigm of differential mediation.

What we can do, therefore, is not speak with animals through AI, but listen and design interfaces that do not translate but instead multiply the forms of interspecific resonance and attention. Rather than developing AI that “imitates” or “interprets” animal language in human terms, we might imagine interfaces that bend to the rhythm of the living, as in the notion of folding: not representations, but dynamic semiotic models that are reshaped by contact. In this sense, Lacková’s (Bennett 2023) proposal, a model of artificial semiosis inspired by the plasticity of folding, can be read as a biosemiotic path toward ecotechnical design: AI not as translators, but as facilitators of encounters, capable of hosting alterity without reducing it. The translation of the nonhuman should not aim at equivalence between codes, but rather at constructing zones of contact in which semiosis emerges as a relational event. AI can be useful only if it stops speaking on behalf of other living beings and instead begins to modulate its activity according to a logic of semiotic hospitality. Only then can we move from the dream of “decipherment” to the reality of shared listening.

5 Conclusions

This article has offered a critical comparison between current computational approaches to animal communication and the biosemiotic framework, highlighting a fundamental tension between the ambition to translate and the semiotic reality of life. In the first section, we examined the structure and objectives of major interspecies “translation” projects based on AI (CETI, CHAT, NatureLM, ISPA) emphasizing their underlying assumption that semiosis can be reduced to language and modeled using syntactic and referential coordinates. Despite their technological sophistication, these models reveal epistemological fragility when confronted with the theories of embodied, situated, and ecological meaning-making developed in biosemiotics.

In the second section, I developed a theory of semiotic opacity as a critical inversion of the “semiotic window” introduced by Stjernfelt and Kull. Far from being a defect, opacity is framed here as an originary condition of communicative life, one that preserves the openness of interpretation and the living being’s freedom to respond in non-predeterminate ways to environmental stimuli. Opacity, in this sense, is the mode through which Hoffmeyer’s notion of “semiotic freedom” is realized, and through which the creative indecidability of all meaning processes is sustained.

In the third and final section, I argued that biosemiotics should be seen today as a vital critical method for evaluating and redefining the conditions of interspecific interaction mediated by AI technologies. The goal is not to reject the machine, but to reframe its role within an ethics of relation grounded in suspension, listening, and the recognition of untranslatability. In line with this position, I concluded that life is not a structure to be deciphered, but a field of semiotic possibilities to be inhabited.

At a historical juncture where artificial intelligence increasingly seeks to extend its reach into the symbolic terrain of nonhuman communication, biosemiotics proposes an alternative epistemological stance that resists hegemony and reductionism in favor of an ecologically grounded relationality. In this framework, opacity is not a limitation to be eliminated, but a vital threshold to be protected. It is precisely this opacity that prevents technology from collapsing the richness of life into code, and that challenges science to shift from extraction to attunement, from control to listening. Seen in this light, biosemiotics not only enriches the contemporary discourse on AI and interspecies communication, but also lays the foundation for a new paradigm of shared cognition in which humans, non-humans, and machines coexist not by erasing difference through transparency, but by embracing the ethical demands of interpretive responsibility.