Coda: AI and Human Sense-Making

The readings across Enacting Life have explored what follows from being a system that must continually produce and maintain itself. Through examples drawn from the lives of bacteria, ticks, frogs, and bees, we have asked how surroundings take on significance in relation to a particular way of living. Human beings belong within this account too. But enactivism does not by itself explain what a particular person cares about or how those concerns developed. For a person, things can go better or worse in many more ways than for a cell, by standards that do not come down to staying alive. Our lives unfold among other people, in relations shaped by language and by histories that extend far beyond any one person. Artificial intelligence is now becoming part of those relations, raising questions about what people come to understand and do through their exchanges with it.

Human Worlds

Humans, like other living beings, meet the world from a particular bodily perspective, and what follows an encounter depends on a person’s present structure, in which the effects of earlier encounters are embodied. This applies to hearing someone speak or reading a page as much as to the encounters discussed in the earlier readings.

Language gives these relations a distinctive character. Maturana understood language as coordinated conduct developing through recurrent interaction (Maturana, 1978). What he termed a consensual domain arises through histories in which each participant’s activity becomes part of the conditions for the other’s next activity. We grow up within such relations, learning ways of speaking and acting with others. The distinctions through which we understand our circumstances develop in these histories. They are neither supplied by words alone nor privately invented by each individual.

Living among other people’s words allows us to encounter much that lies beyond our own direct experience. Words can make an absent person or a distant place part of what concerns us now. Through conversation and reading, we come into contact with ways of living that differ from our own, and may find new possibilities in circumstances that had seemed familiar.

This does not mean that meaning is transferred intact from one person to another. Borrowing an example from the neuroscientist Romain Brette (Middlebrooks, 2026), a book can inform a reader who participates in a language and in practices of reading, but examining the marks on the page does not reveal a substance called information. What the words become for a reader depends on the history and activity of the person who encounters them. The same idea runs through many human activities: Someone learning to cook and an experienced cook can read the same recipe yet notice different possibilities and difficulties.

A human being is a center of activity and perspective within these ongoing relations. That perspective belongs to a life in which something can be at stake, and it develops as the person acts and others respond. Trying to express a concern can change how we understand it, and another person’s response can change what we say or do next. Speaking and writing are among the activities through which a human way of living develops. But what happens when interacting with AI becomes part of these activities?

Language Models

Large language models (LLMs) have entered the social world of words. They draw on a range of writing that no individual could read, bringing together products of many human activities and histories. Their responses can introduce distinctions we would not have arrived at ourselves, including material from areas in which we have little experience. A person may encounter in one exchange a range of perspectives that would otherwise be difficult to bring together.

These abilities also pose a challenge to an embodied account of language. Hubert Dreyfus argued that human intelligence depends on a background of practical, bodily involvement that cannot be captured by explicit rules alone (Dreyfus, 1992). His target was symbolic AI, which differs from present language models. But the broader expectation that sophisticated linguistic performance would require a system’s own sensorimotor and social history needs reconsidering. LLMs can do much with language without having undergone that history themselves. We need to ask what embodiment explains, and which claims these achievements give us reason to revise.

Much of the writing used to train these systems came from people whose words arose within their own lives and relations with others. Human judgments also shape which responses count as good, and people interpret what comes back. A model can draw on patterns in these products without having lived through the histories that produced them. An enactive account has resources for understanding this achievement: it follows how language arises in human activity and how its products become resources for further activity. The task is to explain what models can learn from those products and how their use changes human sense-making, while keeping linguistic competence distinct from a living system’s own perspective (Barandiaran & Almendros, 2025).

A living system participates in producing and maintaining its own organization, so what happens can bear on its continued existence. Present LLMs do not appear to be organized in this way. A system’s ability to produce language, or to be changed by training on its conversations, does not on its own show that anything is at stake for it. But its responses can still matter to people, changing how they understand their circumstances or what they do next.

AI in Human Ways of Living

An exchange with an LLM resembles a conversation. What a person says changes the context for the next response, and that response may change what the person says or does next. But in a conversation between people, both participants bring histories and concerns, and each is usually aware of, and often interested in, the other’s. Both are changed by what passes between them. In an exchange with an LLM, responsiveness does not establish that anything matters to the system itself.

People who know us over time can notice that we might avoid what is difficult, or are becoming able to do something we could not do before. A friend or a good teacher responds to such patterns and cares how we develop. With memory of earlier conversations, an LLM may also track such patterns and point them out, and its responses can look caring. But keeping a record of someone is not the same as having something at stake in how they develop.

A calculator can help someone work out a total, and a map can help them find a route. An LLM can assist in a similar way, carrying out a task that a person has set. Barandiaran and Almendros (2025) describe a further possibility, which they call midtended agency. Here a generative system takes part in a person’s activity as it unfolds, for instance by suggesting the next words while someone writes. What results is the person’s own, yet it did not originate with them alone. Conversation with an LLM can work in a related way, although this goes beyond the cases Barandiaran and Almendros discuss. A person may begin by asking for help with a concern and, through the exchange, come to understand that concern differently. The system’s contribution then enters into forming the purpose rather than only serving one already set.

Another difference concerns the system’s range. A person in conversation speaks from particular experience, and we usually have some sense of what they know and where their knowledge runs out. An LLM can respond across almost any area with the same fluency. But a well-formed answer can also seem convincing before the person has had a chance to think it through, and someone able to judge one part of a response may have no way of judging the rest. The same exchange can widen what a person considers while shaping which possibilities they pursue.

Repeated exchanges can enter into a person’s relations with other people and into how they understand their own circumstances. They may change what someone finds worth pursuing, what they expect from a conversation, or where they turn when something is difficult to resolve. Those changes continue beyond the screen, through actions and encounters that the system does not determine. Over time, a person may come to rely on these exchanges to remember, work through concerns, or make decisions. That support may become part of how they carry out familiar activities, so losing access to the system can disrupt what they are able to do.

Significance depends on the organization and history of the living being: sugar matters to a bacterium because of how the bacterium lives, and a recipe reads differently to someone learning to cook than to an experienced cook. For many people, that history now includes exchanges with language models. What comes to matter to people through those exchanges, and what they become able to do, is a developmental question, and one that can be studied.

References

Barandiaran, X. E., & Almendros, L. S. (2025). Transforming agency: On the mode of existence of large language models. Phenomenology and the Cognitive Sciences. https://doi.org/10.1007/s11097-025-10094-3

Dreyfus, H. L. (1992). What computers still can’t do: A critique of artificial reason. MIT Press.

Maturana, H. R. (1978). Biology of language: The epistemology of reality. In G. A. Miller & E. Lenneberg (Eds.), Psychology and biology of language and thought: Essays in honor of Eric Lenneberg (pp. 27–63). Academic Press.

Middlebrooks, P. (Host). (2026, April 8). Romain Brette: The brain, in theory (Episode 235) [Audio podcast episode]. Brain Inspired. https://braininspired.co/podcast/235/