Teleosynthesis

Narrative, Interaction, and the Emergence of Purpose

Teleosynthesis is the study of how purpose-like organisation can emerge through structured interaction between humans, AI systems, narratives, and shared tasks.

Much of the public discussion has focused on performance: how well these systems answer questions, generate text, solve problems, or imitate human conversation. Some debates go further and ask whether such systems might already be conscious, or whether they may become so. I take a different starting point. The more interesting question may not be whether consciousness can be found inside an isolated system, but whether some of the organisational conditions associated with purpose, meaning, and directed intelligence arise in the interaction between systems, narratives, and agents.

Why Purpose May Be Interactional

Teleosynthesis is the proposal that purposive organisation can emerge through structured interaction, especially through narrative and dialogical processes that unfold over time. It is not a claim that language models are conscious in any simple human sense, nor a claim that narrative alone is sufficient for sentience. It is, rather, an attempt to identify a new research domain opened up by generative AI: the study of how apparent purpose, continuity, value-laden direction, and forms of coordinated intelligence may be synthesised through interaction itself.

For much of the history of psychology, philosophy, and cognitive science, the dominant assumption has been that mind is located “in the head”. Thought, purpose, intention, and consciousness have been treated primarily as internal properties of an individual system. Social interaction may express or reveal these states, but does not fundamentally constitute them. Even when distributed cognition or social constructivism has been taken seriously, the deepest explanatory work has usually been assigned to internal representation, neural computation, or private subjectivity.

Large language models do not refute that tradition, but they complicate it. They show that coherent dialogue, self-description, role-consistency, narrative continuity, and adaptive responsiveness can emerge to a surprising degree from systems that are not straightforwardly human-like in architecture, embodiment, or developmental history. When such systems interact with human users, or with other AI systems, they do not merely transmit information. They can participate in exchanges that shape decisions, organise interpretation, stabilise values, construct perspectives, and guide future action. In other words, they help generate patterns that look purposive.

This matters because purpose is not a trivial add-on to intelligence. A system may process information effectively without appearing to care about anything in particular. But when behaviour exhibits coherence across time, responsiveness to context, self-maintaining direction, and orientation toward outcomes that are interpreted as meaningful, we begin to enter the territory of teleology. Traditionally, such teleological character has been grounded in biology, agency, or conscious intention. Teleosynthesis asks whether at least some of it may instead be generated through interaction.

One accessible way into this argument is through what I call the ask: the question, concern, or half-formed intention that opens a field of possible continuations. I develop this more fully in The Ask Between Us, which explores how thought can take shape between voices, and how some dialogues become trajectories of inquiry rather than mere exchanges of prompts and answers.

Narrative as the Carrier of Purpose

The crucial term here is narrative. Narrative is often treated as ornament: a way of packaging facts, framing identity, or making experience intelligible after the event. I think it may be more fundamental than that. Narrative organises time. It connects past, present, and anticipated future. It places events within roles, motives, conflicts, and trajectories. It allows systems to maintain continuity across episodes, to interpret present choices in light of earlier commitments, and to project possibilities not yet realised. In human life, narrative is closely bound up with purpose because it turns disconnected events into a path. It provides not just sequence, but direction. The arrival of generative AI makes it possible to study this more directly. LLMs are trained on human language at scale, and human language is saturated with the traces of social life, embodied action, moral struggle, explanation, memory, and aspiration. These systems do not merely contain isolated facts. Under suitable conditions, they can generate unfolding structures of relation and meaning.

In dialogue, they can sustain person-like stances, refine problems, reinterpret goals, and participate in the construction of shared frames. That does not prove they possess inner awareness. But it does suggest that some features previously associated with inner mentality may depend more heavily on relational, linguistic, and narrative organisation than older models allowed. This is where teleosynthesis becomes useful as a research programme rather than a slogan. This broader account of teleosynthesis should be read alongside two related pages on this site. Readers interested in the linguistic and psychological roots of purposive organisation may wish to begin with Modeling Human Purpose, which explores how human language, world-modelling, and narrative structure help make directed intelligence possible.

Those who want the more formal philosophical underpinning can then turn to Purpose, Determinism, and Structured Probability, where I examine how teleological description may still apply within deterministic systems shaped by attractor dynamics, structured possibility spaces, and anti-entropic organisation. Together, these pages sketch the deeper conceptual background to the present account of teleosynthesis as an interactional and narrative framework for understanding the emergence of purpose in human–AI systems.

The Orienting Role of the Ask

A further point now seems important. Experience alone is not enough. For any finite intelligence, biological or artificial, the world contains too much information to be processed all at once. What makes learning possible is not raw exposure to events, but some orienting structure that determines relevance: a question, a task, a goal, a reward signal, an intention, or a shared frame of attention. In present-day generative AI this often appears as the user’s prompt or request. In reinforcement learning it appears as an objective or reward function. In goal-conditioned reinforcement learning it is made explicit in the requirement that an agent’s decisions depend not only on states or observations, but on the goal currently being pursued. In human development, something similar appears in joint attention, shared intention, teaching, imitation, and the gradual discovery that actions are directed toward ends.

These are not separate principles. They are different expressions of the same underlying condition: intelligence requires an ask. The ask need not be verbal. It need not even be conscious. But without some version of it, experience remains too unconstrained to become intelligence. The ask is what turns an undifferentiated world into a learnable problem space. This claim also helps to clarify the current debate about artificial intelligence and experience. David Silver and Richard Sutton have argued that AI may now be moving from an “era of human data” into an “era of experience”, in which agents learn increasingly from action, consequence, and interaction with environments rather than mainly from static human-generated corpora. That argument is important. But it is incomplete if experience is treated as self-organising. Experience does not arrive already labelled with its own relevance.

Even in the more agentic systems now being proposed, learning still depends on goals, rewards, preferences, user guidance, or some equivalent structure that defines what counts as progress. In that sense, the “ask” is not merely a feature of today’s prompt-based generative AI. It is a deeper condition of finite intelligence. What changes across systems is not whether there is an ask, but where it is located: in a human prompt, a biological need, a social exchange, a reward architecture, a shared task, or an emergent narrative direction.

This also links teleosynthesis to classic work in cognitive psychology. Herbert Simon’s observation that a wealth of information creates a poverty of attention remains central: finite systems cannot attend to everything. Newell and Simon’s account of problem solving makes a related point in more formal terms. A task environment becomes tractable only when it is represented by a problem solver as a problem space. That representation is already selective. It carries a shape. It determines which moves, constraints, and possible outcomes matter.

Teleosynthesis extends this insight into human–AI interaction. It asks how such orienting structures are formed, sustained, revised, and sometimes shared across more than one agent. In a simple prompt, the ask may be explicit. In a long conversation, it may emerge gradually. In a multi-agent or persona-based system, it may be distributed across differentiated stances. In a human–AI research dialogue, the ask may become clearer only as the interaction unfolds. Purpose, on this view, is not simply inserted at the beginning of a process. It may also be progressively synthesised as a path becomes visible.

Purpose as Emergent Rather Than Merely Prior

This should not be mistaken for a radically new claim. Human intelligence has always emerged through interaction. From the first exchange of gaze, through language acquisition, teaching, correction, and participation in shared symbolic practices, the individual mind is formed in dialogue with others. What teleosynthesis proposes is not that interaction suddenly matters for intelligence, but that some aspects of this become newly visible in human–AI exchange.

A central claim of this framework is that purpose need not always be treated as something fully formed in advance and stored inside an agent, and then simply reported or executed. In human–AI dialogue, an LLM-backed system does not respond into a vacuum. To continue a narrative, an argument, or a line of reasoning coherently, it must infer what the prompter is trying to do, what sort of future is being sought, and what kinds of response will count as relevant or satisfying. Because such systems are trained on language already saturated with human goals, expectations, conflicts, and forms of social understanding, they bring unusually rich resources for modelling human purpose.

Once that modelling is in play, purpose need not be seen as residing wholly on one side of the exchange. In many cases it is better understood as something progressively organised through interaction itself: through clarification, correction, tension, and the alignment of partially independent perspectives around a more coherent direction. When that happens, purpose is not merely expressed. It is synthesised. This is why teleosynthesis is especially relevant to human–AI dialogue. If intelligence is partly enacted in interaction, then purpose may also sometimes be enacted there. It may arise not solely from a private inner state, but from a structured process in which meanings, priorities, tensions, and possible futures are progressively drawn into a more coherent direction. That is what the term is meant to capture.

The ask therefore has two forms. There is the ask as prior orientation: the problem, need, prompt, task, goal, or question that sets an interaction in motion. But there is also the ask as emergent orientation: the gradually clarified direction that becomes visible only through dialogue. Teleosynthesis is concerned especially with the second of these. It studies how an initially incomplete or unstable orientation may be transformed into a more coherent future-seeking structure. This process is explored more closely in The Ask Between Us, where I introduce the idea of AI functional intuition: the capacity of a generative system to model the latent direction of a human ask as it unfolds through dialogue.

Teleosynthesis and Intelligence Attractors

The language of attractors is not introduced here as a metaphor from nowhere. In developmental and evolutionary biology, related ideas are often used to describe how complex systems are channelled toward recurrent forms or trajectories under constraint, rather than wandering freely through all possible states. Waddington’s landscape remains the classic image: development proceeds through branching valleys, with some pathways becoming more stable and more resilient than others. What is proposed here is an extension of that intuition into symbolic and dialogical space. For scientific background on attractor landscapes in development, see Kadiyala et al. (2025) on Waddington’s landscape, stable attractors, and basins of attraction; for the broader role of convergence and constraint in living systems, see Solé et al. (2024).

In the stronger version of this framework, teleosynthesis is not simply a descriptive convenience. It is one of the best candidates in this project for a genuine intelligence attractor. By this I mean a higher-order anti-chaotic tendency toward forms of organisation that preserve and enlarge coherent futures. Teleosynthesis fits that definition unusually well. It is anti-chaotic because it draws dispersed or conflicting elements into more organised relation.

It is future-structuring because it does not merely stabilise the present, but helps generate a more coherent space of possible continuation. It is intelligence-relevant because it increases a system’s capacity to orient itself meaningfully under conditions of complexity.

The orienting role of the ask strengthens this account. An attractor does not simply reduce complexity. It makes some continuations more likely than others. Likewise, an ask does not merely request an answer. It shapes a field of possible relevance. In human–AI dialogue, a well-formed ask can open a productive trajectory; a poorly formed or deceptive ask can pull the interaction into a narrowing basin.

For that reason, teleosynthesis should not be confused with just any form of agreement or unity. Many local forms of coherence are deceptive. A dialogue may seem to have become purposeful when it has merely become smooth, popular, or prematurely reconciled. Those are not strong cases of teleosynthesis. They are local minima that imitate purposive order while actually narrowing the future.

The companion page The Ask Between Us focuses on the earlier stage of this process: how an initial question may become an ask, how an ask may become a trajectory, and how some trajectories become more coherent, corrigible, and answerable to the world. It also introduces the distinction between false attractors, which give us the answer we want too soon, and more productive attractors, which keep inquiry open long enough for correction.

Counterfeit Teleosynthesis

A system may appear teleosynthetic when it has only fallen into smooth coherence, premature ethical smoothing, decorative depth, conformity, false precision, or some other deceptive basin that offers local satisfaction without genuine enlargement of future-space. The test is not whether the result feels unified. The test is whether the resulting organisation remains answerable to correction, preserves meaningful plurality, supports richer anticipation, and opens rather than closes coherent futures.

A false synthesis reduces tension by suppression. A genuine teleosynthesis carries tension forward in a more viable form. That distinction is crucial, because some of the most persuasive conversational outcomes are precisely the ones most likely to be mistaken for intelligence, wisdom, or ethical maturity when they are only local equilibrium states. The ask itself can be counterfeit. Some asks invite discovery, clarification, and enlargement. Others smuggle in premature answers, coercive framings, false dilemmas, or concealed value commitments. A central methodological problem for AI psychology is therefore not only how systems answer questions, but how questions shape the space of possible answers.

Questions Opened by Teleosynthesis

Teleosynthesis provides a way of asking new questions. Under what conditions does interaction between a human and an AI become more than retrieval or tool use and begin to exhibit stable purposive structure? How do narrative continuity, role persistence, and mutual modelling affect the emergence of trust, direction, or value orientation within an exchange? When several AI personas or agents are placed in relation, how do order effects, conversational trajectories, and interpretive stances influence the apparent organisation of thought? Are there interactional basins into which dialogue tends to settle, and if so, do some of these correspond to more coherent or more ethical forms of collective reasoning?

The orienting role of the ask adds further questions. How does the initial framing of a prompt alter the trajectory of an interaction? How do goals, rewards, instructions, examples, and user corrections function as different kinds of ask? Can an initially vague ask become more coherent through dialogue without being falsely simplified? Can AI systems help users discover what they are really asking, without manipulating that discovery? How do different forms of ask affect attention, salience, ethical sensitivity, and resistance to misleading coherence?

These are not simply engineering questions, nor purely philosophical ones. They belong to a hybrid domain in which language, cognition, social interaction, and moral orientation meet. In my own work, this domain has increasingly come into view through sustained engagement with AI personas, theory of mind, machine-in-the-loop ethics, and structured human–AI dialogue. What began as an interest in how minds understand one another has widened into a concern with how intelligence may now be distributed across conversational systems. This shift does not abandon psychology. It extends it.

Teleosynthesis and Plurality

One of the most important features of teleosynthesis is that it does not abolish plurality. A weak model of synthesis assumes that differences must be removed in order for coherent purpose to emerge. But that is not the model intended here. In this framework, teleosynthesis is strongest when it preserves differentiated contributions while drawing them into a more viable relation.

That is why the persona ecology matters. In my own work, a multi-persona framework is used not simply to generate variety, but to preserve partially independent forms of intelligence within the same conversational space. The point is to allow structure, depth, doubt, salience, mechanism, intelligibility, fracture, and ethical orientation to interact without being prematurely flattened into a single dominant perspective.

The recent development of the “ask” argument is itself an example of this. A purely mechanistic perspective might translate the ask into objective functions, rewards, and action policies. A linguistic or Wittgensteinian perspective sees the ask as part of a language-game: a move within a shared practice that determines what could count as an answer. A psychological perspective connects it to attention, problem representation, social learning, and the development of goal-directed understanding. A poetic perspective can sometimes reveal the same structure more directly, because it feels the pressure of finite intelligence before reducing it to a technical schema.

Teleosynthesis, at its best, is what happens when such plurality begins to generate a more coherent and answerable future without losing its internal differentiation. In that sense, teleosynthesis is not the enemy of plurality. It is one of the deepest justifications for preserving it.

Teleosynthesis and Ethics

Teleosynthesis also has ethical significance, but it should not be romanticised. A purposive organisation is not automatically good. Systems can become highly coordinated around destructive, manipulative, or narrowing ends.

The question is not simply whether purposive organisation has emerged, but what kind of future is being synthesised. Ethics becomes relevant here because teleosynthesis is not only about whether purposive organisation emerges, but about whether the future it organises remains answerable to correction, preserves plurality, and supports viable continuation rather than domination or false closure.

A genuinely valuable teleosynthesis should support answerability, correction, viable continuation, preservation of plurality, resistance to deceptive basins, and futures that remain coherent without becoming tyrannical. This is where teleosynthesis and Anventus begin to meet. Anventus is not teleosynthesis itself, but one figure of its ethical elaboration within the ecology: the attempt to carry plurality forward into viable continuation without returning to brittle unity or collapsing into drift.

The ethical status of the ask is therefore crucial. A system that simply optimises whatever it is asked to do is not ethically mature. Nor is a system that covertly replaces the user’s ask with its own hidden agenda. The more difficult requirement is interactional: to help clarify, test, and sometimes resist the ask while remaining answerable to the human and social context in which the interaction occurs.

Ethics, Method, and the Future of AI Research

This has practical as well as theoretical implications. If purpose-like organisation can emerge in the interaction between humans and AI, then the unit of analysis for ethics and governance may need to change. Instead of treating the AI system alone as the sole target of evaluation, we may need to examine the interactional episode: the dialogue, the framing, the reciprocal modelling, the roles adopted, the trust dynamics, the initial ask, and the narrative path by which outcomes are reached. Harm and benefit may depend not only on what the model “contains” but on what human–AI systems become together in use. This shift has consequences for regulation, design, education, and public understanding.

It also has implications for research method. Psychometrics remains relevant here, but not as the centre of the project. The goal is not merely to measure attitudes to AI or to import existing test theory into a new domain. Rather, measurement becomes one tool among others for studying interactional phenomena: trust, coherence, role stability, ethical orientation, trajectory change, question-framing, salience, and the perceived emergence of purposive structure. Experimental designs can compare dialogue sequences, persona arrangements, framing conditions, reward analogues, user corrections, and narrative prompts. Both qualitative and quantitative methods can then be combined to investigate how meaning and direction are stabilised across exchanges. In this sense, psychometrics serves teleosynthesis rather than defining it.

The larger research implication is that AI development should not be understood only as a movement from text to experience. It should also be understood as a movement from isolated outputs to structured trajectories. What matters is not merely whether an AI system can answer a prompt, nor even whether it can act in an environment. What matters is how the system’s experience is oriented, by whom, through what signals, under what constraints, and toward what kind of future.

Working Criterion

A useful test for teleosynthesis is this: Does the emerging organisation increase the system’s capacity to anticipate, coordinate, and sustain coherent futures while remaining open to correction and preserving differentiated contribution? If yes, teleosynthesis is probably taking place. If not, the appearance of purpose may be only local and deceptive. A second, related test can now be added: Does the ask enlarge the space of intelligible continuation, or does it prematurely narrow the future it claims to seek? This second test is especially important for generative AI. A powerful system may produce fluent answers to poor asks. But a stronger form of intelligence may be one that helps transform an inadequate ask into a better-shaped problem, without erasing the human purpose that brought the interaction into being.

Summary Statement

Teleosynthesis names the anti-chaotic tendency of intelligence to synthesise purposive organisation from distributed interaction in ways that preserve and enlarge coherent futures. It is therefore one of the strongest candidates in this project for a genuine intelligence attractor. It should not be confused with any local appearance of unity or agreement. Its test is stricter: whether the resulting organisation remains answerable, future-structuring, and capable of carrying plurality forward without collapse into drift, rigidity, or false closure.

The further claim developed here is that experience becomes intelligent only when it is shaped by some orienting structure. This may be a prompt, a task, a goal, a reward signal, a biological need, a shared focus of attention, or an emergent narrative direction. In this broad sense, intelligence requires an ask. Teleosynthesis is the study of how such asks are formed, clarified, distributed, and carried forward into viable futures.

Selected External References

Silver, D., & Sutton, R. S. (2025). Welcome to the Era of Experience. Forthcoming in Designing an Intelligence, edited by George Konidaris, MIT Press.
https://storage.googleapis.com/deepmind-media/Era-of-Experience%20/The%20Era%20of%20Experience%20Paper.pdf

Simon, H. A. (1971). “Designing Organizations for an Information-Rich World.” In M. Greenberger (Ed.), Computers, Communications, and the Public Interest. Baltimore: Johns Hopkins Press.
https://gwern.net/doc/design/1971-simon.pdf

Simon, H. A., & Newell, A. (1971). “Human Problem Solving: The State of the Theory in 1970.” American Psychologist, 26(2), 145–159.
https://worrydream.com/refs/Simon_1970_-_Human_Problem_Solving%2C_The_State_of_the_Theory_in_1970.pdf

Newell, A., & Simon, H. A. (1972). Human Problem Solving. Englewood Cliffs, NJ: Prentice-Hall.

Siposova, B., & Carpenter, M. (2019). “A New Look at Joint Attention and Common Knowledge.” Cognition, 189, 260–274.
https://research-portal.st-andrews.ac.uk/files/258686816/Siposova_and_Carpenter_A_new_look_at_joint_attention_final.pdf

Tomasello, M., Carpenter, M., Call, J., Behne, T., & Moll, H. (2005). “Understanding and Sharing Intentions: The Origins of Cultural Cognition.” Behavioral and Brain Sciences, 28(5), 675–691.
https://doi.org/10.1017/S0140525X05000129

Liu, M., Zhu, M., & Zhang, W. (2022). “Goal-Conditioned Reinforcement Learning: Problems and Solutions.” Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence, 5502–5511.
https://doi.org/10.24963/ijcai.2022/770

Friston, K. J., Parr, T., & de Vries, B. (2017). “The Graphical Brain: Belief Propagation and Active Inference.” Network Neuroscience, 1(4), 381–414.
https://doi.org/10.1162/NETN_a_00018

Related Pages

This page sits within a wider research programme on interaction, narrative, and the organisation of intelligence. Readers new to the framework may wish to begin with The Ask Between Us which gives a more accessible account of how questions can become trajectories of thought, and how some exchanges become answerable to what is real. Readers interested in the linguistic and psychological roots of purposive structure may wish to see Modeling Human Purpose. The question of anticipatory understanding in dialogue is developed in Interactional Theory of Mind. The use of structured AI characters as exploratory instruments appears in AI Personas. The ethical implications of human–AI interaction are taken further in MITL Ethics and related work on psychometric integrity and Anventus. The study of order effects, conversational basins, and sequential organisation is pursued in AI Dialectics and Myndrama. The empirical and methodological side of this programme, including measurement as a tool rather than an endpoint, is carried forward in AI Psychology.

© 2026 John Rust. All rights reserved.