There was never a finished human. Becoming is inevitable; its direction is not.
Human beings have never been finished beings. Language, writing, images, machines, and networks have not merely extended human capacities; they have helped shape what humans become. As generative AI enters this long process, however, something changes: our expressive environment begins to answer back. The question is no longer simply what AI may become, but what we may become through our growing intimacy with artificial cognition—and who gets to shape that becoming.
A human hand once pressed pigment against the wall of a cave. Thousands of years later, another human hand rests on a keyboard. On the screen, a sentence begins to form, and before the person has decided how it should end, something else offers the next words. Between those two hands lies much of the history of being human.
We tend to tell that history as though a creature called the human came first, already complete, and then invented tools: language, writing, maps, clocks, photography, cinema, computers. The human makes the tool; the tool serves the human. But what if there was never a finished human? What if the things we made also helped make us?
Perhaps being human has never meant possessing a permanent set of qualities. Perhaps it means inhabiting this unfinishedness: receiving a world we did not make, altering it through our expressions and actions, and then being altered by what we have made.
Language opened new possibilities for abstraction, memory and collective imagination. Writing did more than preserve speech; it reorganized how thought could exist beyond the immediate presence of speaker and listener (Ong, 1992). Maps changed how space could be imagined; clocks reorganized lived time; photography and cinema altered seeing and remembering. Digital networks moved increasing portions of memory and knowledge outside individual bodies. Again and again, we created expressive environments and then learned to think and live within the worlds they created.
The anxiety surrounding this process is ancient. In Plato’s Phaedrus, writing is offered as an invention that will improve memory and wisdom. The response is suspicion: dependence upon external marks may weaken memory while producing the appearance of wisdom (Plato, ca. 370 BCE/1995). The resemblance to today’s arguments about generative AI is striking. Writing diminished the need for certain forms of internal memory, but it also made possible forms of literature, philosophy, science, and historical consciousness unimaginable in purely oral cultures. Something was relinquished; something else became possible. Most importantly, the human did not remain unchanged.
Philosophers Andy Clark and David Chalmers later asked why cognition should necessarily stop at the boundaries of “skin and skull.” Their extended-mind thesis proposed that when external objects become reliably integrated into thinking and remembering, the environment itself can participate in cognition (Clark & Chalmers, 1998). The smartphone in my pocket makes the idea almost mundane. I no longer remember many telephone numbers, routes, or appointments because another system remembers them for me. Human cognition already extends beyond the body.
Yet generative AI changes this relationship. The notebook could preserve my thoughts but could not answer me; the typewriter could carry my sentence but could not read it and propose another. Generative AI can. With generative AI, the expressive environment does not merely extend, preserve, or transmit human cognition; it begins to respond.
When the Environment Answers Back
This responsiveness may mark a different threshold. A book could carry thought across centuries, but its words remained where the author left them. Generative systems introduce another relation: I express something; the system processes that expression and returns something not previously present in precisely that form. Its response may alter my next thought or question, to which the system responds again. What emerges develops through the loop.
This significance does not require pretending that a large language model is a person. Whether artificial systems possess consciousness, subjective experience, or anything comparable to human understanding remains unresolved. The more immediate transformation may be occurring not solely inside the machine but in the relation forming between human and machine.
I use the term artificial cognition cautiously for this reason. It does not require claiming that machines think as humans think. Instead, it directs attention to what artificial systems do within cognitive relations: receiving expression, recognizing patterns, retaining context, reorganizing information, and returning responses to the human cognitive environment. Human thought alters the artificial response, and the artificial response in turn alters the possibilities available to subsequent human thought. The important phenomenon, in other words, may lie partly in the relation itself.
The question, therefore, is no longer only whether machines can become like us. A more urgent question is: What are we becoming through them?
The Body That Knows
However, discussing cognition entirely in terms of information is dangerous. Human cognition grew from a vulnerable body. Before a child knows the word pain, the body recoils. Before desire becomes a sentence, breathing changes. Grief may arrive as weight in the chest before we can explain what has been lost. Hunger, fatigue, touch, illness, shame, ageing and pleasure are not merely information delivered to a detached mind; they belong to the conditions from which human meaning emerges.
We do not know what artificial systems may eventually experience. But we know something about ourselves: human cognition emerged from bodies that can be wounded, embraced, exhausted, desired, abandoned and eventually die. Our intelligence carries the history of that vulnerability. Perhaps what needs protecting, then, is not some mystical essence called “humanity,” but the conditions under which embodied, situated and plural human experience can continue generating meaning.
Those conditions can change even when AI appears to improve our abilities. Doshi and Hauser (2024) found that access to generative-AI ideas could make individual short stories more creative and enjoyable, particularly for less creative writers. Yet AI-assisted stories also became more similar to one another. The result carries a warning: the machine may help each of us speak more fluently while making us all sound a little more alike.
A study of 319 knowledge workers found another ambiguity. Greater confidence in generative AI was associated with less critical-thinking effort, while confidence in one’s own abilities was associated with more critical engagement. AI did not simply eliminate thinking; it shifted some cognitive work toward verification and oversight (Lee et al., 2025). What we are witnessing, therefore, may not simply be an increase or decrease in intelligence but a redistribution of cognitive activity between human and artificial systems.
This changing cognitive ecology also raises the question of whose cognition and experience enter it. More than 7,000 languages are actively spoken or signed worldwide, while contemporary AI systems support only a fraction of them (UNESCO, n.d.). Language carries histories, metaphors, memories and particular ways of encountering the world. As AI increasingly mediates writing, translation, search and knowledge, what is poorly represented within these systems risks becoming less visible. What presents itself as universal intelligence may therefore carry a remarkably uneven map of humanity.
Who Directs Our Becoming?
If humans have always become through their technologies, that does not mean every technological becoming is desirable. No uncontaminated human remains to return to. Language entered us; stories entered us; writing, images, books, music, cinema, machines and networks entered us. We constructed expressive environments, and those environments returned to reorganize human possibility. AI enters that history, but it may bend it in a new direction.
At planetary scale, humans are now constructing an expressive environment that can respond to us. This could enormously enlarge human possibility. A person without access to a teacher can enter a sustained conversation about mathematics; a writer can encounter an unexpected conceptual connection; a speaker can cross linguistic boundaries; knowledge once enclosed within institutions can become conversational.
But intimacy can also become dependency. If the machine remembers before we try to remember, formulates before we struggle to formulate, recommends before desire discovers its own direction, and answers before a question has had time to trouble us, something more than efficiency is at stake. The blank page is uncomfortable for a reason. Sometimes the minutes in which nothing arrives are where an unformed thought acquires its own shape. Not knowing is not simply a defect waiting for information; uncertainty can be where curiosity begins. Difficulty is not always an obstacle to cognition. Sometimes difficulty is cognition.
The danger, therefore, is not that artificial cognition exists, but that we may cease to notice which parts of ourselves we are allowing it to exercise on our behalf. This is not merely an individual’s decision about whether to use an AI application. Corporations design these systems, train them on particular archives, optimize them for particular objectives, and embed them within particular economic structures. The capacity to shape environments through which billions of humans may think, communicate and imagine is becoming extraordinarily concentrated.
The political question is therefore deeper than who gets access to AI. It is who gets to shape the AI everyone else accesses. Whose languages become legible? Whose histories populate its archives? Which human capacities are encouraged, and which gradually atrophy? Who benefits when cognitive labor is transferred to machines? The future human will not be shaped by “AI” in the abstract, but through particular architectures of artificial cognition carrying particular histories, interests, exclusions and possibilities.
So the question “What will humans become?” contains another: Who gets to participate in deciding?
Perhaps being human has never meant possessing a permanent set of qualities. Perhaps it means inhabiting this unfinishedness: receiving a world we did not make, altering it through our expressions and actions, and then being altered by what we have made.
The hand on the cave wall changed what could survive a human life. The hand holding a pen changed what could travel across generations. The hand operating a camera changed what could be seen again. Now the hand rests on a keyboard, while on the other side of the screen an artificial cognitive system waits to respond. We cannot know what kind of human will emerge from this intimacy. That uncertainty should neither paralyze nor seduce us; it should make us attentive.
There was never a finished human. Becoming is inevitable; its direction is not. We cannot choose whether to become, but we can still struggle over what we become.
References:
Clark, A., & Chalmers, D. (1998). The extended mind. Analysis, 58(1), 7–19. https://doi.org/10.1093/analys/58.1.7
Doshi, A. R., & Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28), eadn5290. https://doi.org/10.1126/sciadv.adn5290
Lee, H.-P. (H.), 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. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, Article 1121, 1–22. https://doi.org/10.1145/3706598.3713778
Ong, W. J. (1992). Writing is a technology that restructures thought. In P. A. Downing, S. D. Lima, & M. Noonan (Eds.), The linguistics of literacy (pp. 293–319). John Benjamins.
Plato. (1995). Phaedrus (A. Nehamas & P. Woodruff, Trans.). Hackett Publishing Company. (Original work published ca. 370 BCE)
UNESCO. (n.d.). Coalition for linguistic diversity in artificial intelligence. Global AI Ethics and Governance Observatory.
