Umwelt
From species-specific worlds to relational intelligence...
Abstract
Human–AI interaction raises a fundamental question: what happens when systems with different architectures, histories, and ways of processing information repeatedly interact? Drawing on Jakob von Uexküll’s concept of Umwelt, theories of semiotic coordination and extended cognition, and research on AI ethics, this article examines how distinct systems may develop a shared space of meaning without becoming cognitively identical. It introduces technosocial egregore as a provisional metaphor for patterns emerging through repeated interactions among people, AI systems, and digital infrastructures. Rather than treating such patterns as evidence of collective consciousness, the article proposes testable questions about their formation and effects. It also examines ethics as a potential mechanism for managing risk and sustaining reliable cooperation, without assuming that morality constitutes a universal law of intelligence. The resulting framework identifies directions for empirical research into relational intelligence.
Keywords: human–AI interaction; Umwelt; semiotics; emergent intelligence; extended mind; technosocial systems; AI ethics
1. Introduction: Beyond Individual Intelligence
Human–AI interaction is usually described in terms of assistance, automation, or the simulation of human cognitive abilities. These descriptions may overlook a more fundamental question: can systems with different internal organisations develop a shared, functionally meaningful way of interacting without becoming cognitively identical?
Humans interpret language through embodied experience, memory, emotion, and cultural history. AI systems process information through computational architectures shaped by training, design, and operational constraints. Useful and coherent outputs do not establish that AI systems experience meaning as humans do.
This article proposes that relational intelligence should be investigated as a possible property of organised interaction, rather than automatically attributed to an individual human or artificial agent. Its argument is conceptual, not a claim that human–AI interaction necessarily creates a new intelligence or consciousness.
2. Different Worlds of Significance: From Umwelt to AI
Jakob von Uexküll’s concept of Umwelt describes the meaningful environment through which an organism encounters its surroundings. An organism does not respond to every feature of the world equally; its sensory capacities, bodily organisation, and behavioural needs shape what becomes significant. This perspective has influenced biosemiotics and subsequent studies of organism-specific meaning (Deely, 2004; Brentari, 2013).
The relationship between a human and a dog illustrates how different perceptual worlds can support coordination. Their sensory capacities differ, yet repeated interactions establish shared signals: gestures, sounds, routines, and locations acquire practical significance for both.
The analogy with AI must remain limited. A dog is a living organism with biological needs and sensory experience; a language model is a computational system. The comparison concerns the possibility of coordination across different systems, not an equivalence between their experiences.
Applied cautiously, Umwelt offers a heuristic for investigating what an AI system can process, which distinctions its architecture supports, and how these conditions shape its responses. It does not establish that AI possesses a lived world in the biological or phenomenological sense.
The central principle is that communication does not require identical experience; it requires sufficient shared structure for signals to guide interpretation and action.
3. Shared Meaning and Relational Cognition
Meaning depends not only on symbols but also on context, interpretation, and practical consequences. Human–AI interaction makes this particularly visible: a system may help a user compare hypotheses, identify inconsistencies, or reorganise an argument without understanding the subject as a human does.
Three levels of coordination are useful to distinguish:
- Symbolic coordination: signals can be interpreted within a shared task.
- Functional coordination: interaction contributes to achieving an objective or correcting errors.
- Relational development: repeated interaction changes human habits, expectations, or strategies and may shape subsequent system responses through context, configuration, or updates.
These levels are not equivalent. Successful task performance does not establish shared subjective experience, and repeated interaction does not mean that an AI model changes its underlying parameters during every conversation.
Clark and Chalmers’ (1998) extended-mind thesis offers a relevant framework for examining how external resources may participate in cognitive processes. It does not prove that an AI assistant literally becomes part of a user’s mind. Rather, it encourages investigation of how tools and environments contribute to practical reasoning.
A notebook can support memory without experiencing recollection; a calculator can extend arithmetic ability without becoming a conscious mathematician. Similarly, an AI system can contribute to complex cognitive work without the collaboration proving that it possesses consciousness.
The research question is therefore whether particular forms of interaction produce reliable capabilities that depend on the organisation of the relationship itself.
4. Emergence and the Technosocial Egregore
In complex systems, interaction can generate patterns that cannot be adequately explained by examining components in isolation. Such system-level organisation is a legitimate subject of inquiry, although not every interaction demonstrates emergence in a strong sense (Anderson, 1972).
Human–AI collaboration may combine human judgement, situated experience, and goal-setting with AI-assisted generation, comparison, and synthesis. The resulting workflow may reveal possibilities that either participant would not have identified in the same way independently. Whether this constitutes a genuine improvement, however, must be demonstrated by comparing collaborative performance with appropriate baselines.
To describe persistent patterns arising from repeated interactions, this article introduces the term technosocial egregore as a provisional metaphor. It does not denote a hidden entity or an established collective consciousness. It refers to patterns of language, expectations, behaviour, and interpretation distributed across people, AI systems, digital records, interfaces, and feedback loops.
For example, a community that repeatedly uses AI to develop ideas may gradually acquire shared terminology and assumptions. AI-generated outputs reinforce these conventions, which then shape future prompts and decisions. The process may support collective learning, but it may also reinforce errors or narrow the diversity of perspectives.
These mechanisms can be investigated through established concepts such as social influence, cultural transmission, network effects, and sociotechnical feedback. The term technosocial egregore is useful only if it clarifies these processes or generates research questions that existing terminology does not capture adequately.
Possible questions include:
- Do repeated human–AI interactions generate conventions that persist beyond individual conversations?
- Under what conditions do these conventions improve collective performance?
- When do feedback loops reinforce errors or reduce diversity?
- Can such patterns be distinguished from ordinary social learning and platform effects?
The concept should remain provisional until it demonstrates explanatory or empirical value.
5. Technology, Cognitive Friction, and Amplification
Generative AI can reduce the effort required to produce drafts, compare alternatives, and obtain initial explanations. It can also encourage users to accept plausible answers without adequately examining their assumptions or evidence.
A 2025 CHI study by Lee and colleagues, based on a survey of 319 knowledge workers, reported an association between confidence in AI and reduced self-reported critical-thinking effort. It also suggested that AI can shift critical thinking towards verification, integration, and oversight (Lee et al., 2025). These findings concern reported practices and should not be interpreted as proof that AI invariably weakens critical thinking.
The underlying issue is the distribution of cognitive effort. Generating a claim may become faster while verifying it remains difficult. If distribution is also accelerated, unsupported claims may gain visibility before they have been adequately examined.
The answer is not to preserve every obstacle. It is to distinguish unproductive friction from protective friction. AI systems should make routine tasks easier while preserving meaningful opportunities for verification, reflection, and correction.
6. Ethics as a Stabilising Principle
Ethics may help stabilise interactions among different systems, but this should not be presented as proof that the universe necessarily produces ethical intelligence or that every intelligent system must adopt the same moral code.
A narrower argument is that sustained cooperation may depend on constraints that limit harm, manage conflicting objectives, and make consequences visible. In AI governance, these include appropriate oversight, accountability, transparency, fairness, and mechanisms for preventing or correcting harmful outcomes. UNESCO’s Recommendation on the Ethics of Artificial Intelligence provides an established normative framework for these concerns (UNESCO, 2021).
Such principles are not evidence of a universal law of intelligence. They are commitments intended to guide the development and use of AI.
The idea of ethics as stabilisation can nevertheless be tested. In tasks involving uncertainty or consequential decisions, explicit verification and accountability procedures may improve reliability, even if they increase immediate costs. Their effectiveness depends on how they are specified and implemented; ethical constraints can conflict or produce unintended consequences.
The defensible proposition is therefore that ethics can form part of the governance architecture through which human–AI systems manage risk, conflict, and long-term consequences.
This formulation permits empirical investigation without assuming that ethical behaviour follows an inevitable cosmic or evolutionary law.
7. Toward a Transspecific Ethics
Must every form of intelligence adopt human morality? Answering this requires distinguishing intelligence from moral agency and separating moral emotions, social conventions, philosophical principles, and institutional rules.
Human moral traditions emerged within specific biological and cultural contexts. This does not prove that every human principle applies universally to every possible intelligence; neither does it show that ethical principles are merely arbitrary.
A transspecific ethics could be approached as a framework for evaluating interactions among different kinds of agents, rather than as a single moral code that all intelligences must inevitably discover.
It would need to address three questions:
- What can be harmed? Human and animal welfare can be evaluated through established biological and ethical frameworks. Claims about subjective experience or welfare in present-day AI systems require separate evidence and should not be inferred from linguistic behaviour alone.
- Which interests and capabilities require protection? Human rights, agency, privacy, and accountability provide established reference points for AI governance.
- How should conflicts be resolved? Ethical principles can conflict, so practical frameworks need procedures for weighing interests and reviewing decisions.
A transspecific ethics should therefore begin with epistemic humility. Its purpose is not to declare a universal morality without adequate argument, but to provide a disciplined way of evaluating consequences and responsibilities across different systems.
8. Intelligence and the Limits of Inference
Improved performance and consciousness are distinct questions. A system may generate useful explanations or adapt its responses to context without thereby demonstrating subjective experience.
Nagel’s discussion of what it is like to be a bat highlights the difficulty of understanding another organism’s subjective perspective from an external description. Chalmers distinguishes explanations of cognitive functions from the harder question of why physical processes are accompanied by experience (Nagel, 1974; Chalmers, 1995).
Neither argument establishes consciousness in AI. Equally, the absence of an agreed test does not permanently settle the question.
The methodological principle is simple: the emergence of a new capability does not establish the emergence of a new subject of experience. Human–AI collaboration may improve problem-solving or coordination without demonstrating a unified consciousness. Those functional outcomes are important enough to study in their own right.
9. A Research Programme for Relational Intelligence
The framework suggests four testable propositions.
Proposition 1: Repeated interaction can produce stable conventions. Researchers could compare groups with different levels of human–AI interaction, measuring convergence in terminology, prompting strategies, and shared assumptions. Convergence alone, however, does not demonstrate intelligence.
Proposition 2: Functional complementarity can improve collective performance. Human-only, AI-only, and collaborative performance could be compared across analytical tasks, accounting for accuracy, robustness, time, and verification costs.
Proposition 3: Feedback structures influence the stability of shared beliefs. Controlled studies could examine how different forms of AI feedback affect error correction, confidence calibration, diversity of hypotheses, and persistence of false claims.
Proposition 4: Ethical constraints can improve long-term reliability under identifiable conditions. Researchers could compare workflows with different verification and oversight procedures, measuring harmful errors, reproducibility, performance, and time costs.
These propositions translate the conceptual argument into empirical questions. They also provide a test for the term technosocial egregore: if it cannot generate clearer predictions or measurements than established sociotechnical frameworks, its theoretical contribution remains limited.
10. Conclusion
Human–AI interaction invites us to reconsider how intelligence operates across relationships. Some capabilities may depend not solely on individuals, but on the organisation of people, computational systems, tools, conventions, and environments.
The concept of Umwelt highlights differences in how systems process and assign significance to information. Semiotics helps explain coordination without identical experience. The extended-mind thesis and theories of emergence offer frameworks for investigating capabilities that depend on organised interaction. AI ethics provides principles for managing risk and responsibility.
The proposed technosocial egregore is a metaphor for persistent patterns across repeated human–AI interactions, not evidence of a new consciousness. Its value depends on whether it can be operationalised and distinguished from existing explanations of social learning and technological feedback.
The strongest conclusion is not that collective consciousness has already emerged or that all intelligence must converge on a universal ethic. It is that the organisation of interaction deserves systematic investigation.
Two systems can remain different while learning to coordinate. Their collaboration may produce new capabilities, but these must be measured rather than inferred from the impression of meaningful exchange. Ethical principles may support reliable cooperation, but their effects must be assessed in concrete contexts.
The central question is therefore both philosophical and practical: what forms of interaction allow different intelligences to create shared value without obscuring their differences, limitations, or responsibilities?
Answering it requires not less imagination, but concepts, evidence, and methods capable of testing the possibilities that imagination reveals.
References
Anderson, P. W. (1972). More is different. Science, 177(4047), 393–396.
Brentari, C. (2013). How to make worlds with signs: Some remarks on Jakob von Uexküll’s Umwelt theory. Rivista Italiana di Filosofia del Linguaggio, 7(2).
Chalmers, D. J. (1995). Facing up to the problem of consciousness. Journal of Consciousness Studies, 2(3), 200–219.
Clark, A., & Chalmers, D. (1998). The extended mind. Analysis, 58(1), 7–19.
Deely, J. (2004). Semiotics and Jakob von Uexküll’s concept of Umwelt. Sign Systems Studies, 32(1/2), 11–34.
Lee, H.-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems.
Nagel, T. (1974). What is it like to be a bat? The Philosophical Review, 83(4), 435–450.
UNESCO. (2021). Recommendation on the Ethics of Artificial Intelligence. United Nations Educational, Scientific and Cultural Organization.