What Makes a Heart Beat
The ask as a constitutive orientation in human language, living systems, and generative AI
There is a more basic question beneath many discussions of intelligence.
Before we ask what intelligence is, or what language is, it helps to begin somewhere simpler and more concrete: with dialogue. Human beings live by turn-taking. One speaks, another responds. A question is put, an answer is awaited. A gesture is made, a reaction follows. In that respect, one striking feature now shared by human–human and human–AI interaction is that both begin in the same way: one party addresses another.
Every dialogue has to be initiated. Someone has to speak first, call out, request, invite, appeal, query, or demand. That initiating act is what I mean by the ask. By an ask, I do not mean only an interrogative sentence. I mean the more fundamental act by which one mind, or one system, turns toward another in expectation of a response. The ask is the opening move that brings a dialogue into being. A dialogue ends when the pressure to continue has gone.
The ask may be more fundamental than it first appears, especially in human–AI interaction. We are used to thinking of prompting as a practical input to a machine. But at a deeper level it may belong to a much older structure: the structure by which meaning becomes directional at all. An ask establishes relevance, uncertainty, and answerability. It creates a space in which something is missing, something matters, and something or someone may answer back.
Seen in this light, the ask is not just one linguistic act among others. It may be one of the constitutive conditions of dialogue itself.
The developmental story begins very early. A baby does not first encounter the world as a detached inventory of objects and only later discover communication. The baby enters a world of response. Hunger is met by feeding. Distress is met by holding, soothing, and voice. In moments of feeding, gaze, touch, rhythm, and bodily closeness, the infant is already participating in a primitive exchange long before words arrive. Mutual engagement with a responsive caregiver, the following of gaze, the first social smile, separation distress, and the gradual formation of attachment all belong to this early structure of answerability.
What the child learns first is not vocabulary but something more basic: that signals can be directed, that responses may follow, that absence and return matter, and that another person may be relied upon for comfort, help, and recognition. In psychological terms, this is the beginning of attachment; in interactional terms, it is the beginning of turn-taking under conditions of need, expectation, and reply. Language grows out of that older field rather than replacing it.
Something similar appears beyond the human case. Across the animal kingdom, researchers have found orderly exchanges of communicative signals—forms of turn-taking in birds, mammals, insects, and amphibians. Social animals also live by patterned signalling, response, reassurance, warning, and repair. I do not mean that animal communication is language in the human sense. Only that the deeper structure of asking and answering may be older than human language, and that human dialogue may have grown out of more ancient forms of responsive interaction.
In the human case, this older structure was progressively elaborated rather than replaced. What began in the intimate exchanges of caregiver and child widened into shared systems of meaning carried across groups and generations. Language extended signalling into story telling, song, dance, ritual, gesture, image, and collective memory. In this way, human beings came to inhabit not just a physical environment but a symbolic one: a world saturated with signs, expectations, and inherited meanings. This wider sign-world is what, in semiotics, is called the semiosphere.
This is crucial for thinking about generative AI. A large language model is trained not on bare information but on human language, and human language is already saturated with the history of asking and answering. Every corpus carries traces of explanation, persuasion, pleading, warning, speculation, teaching, longing, bargaining, promising, confessing, and imagining. The model inherits not merely propositions but the fossil record of communicative life. OpenAI’s 2019 GPT-2 paper already reported that language models could begin to perform tasks “without any explicit supervision” and might learn to infer and perform tasks demonstrated in natural language in order to predict them better.[1]
This means that when a generative model is asked a question, it is not starting from nowhere. Nor is it merely matching words in a dead archive. It is operating within a symbolic field already structured by countless human acts of inquiry and reply. Language itself contains deep traces of relevance, salience, purpose, and perspective. A model trained on that language therefore inherits, at scale, patterns that were never merely lexical. They were already relational, psychological, and purposive. As Jacob Andreas of MIT has argued, language models can be understood not merely as statistical compressors of text but as models of agents engaged in intentional communication. On that view, they may infer not only linguistic pattern but aspects of communicative intention, belief, and goal structure from the traces left in human discourse.[2]
This, I suggest, was the deeper importance of the Radford et al. insight. If large language models can learn from naturally occurring demonstrations embedded in language, then they inherit far more than vocabulary or syntax. They inherit, from language itself, traces of human task structure, communicative intention, practical inference, and purposive exchange. That was a remarkable discovery, and its full significance may still not have been absorbed. The importance of this becomes clearer if we step back for a moment. While later work in AI would come to assume that purposive behaviour must be explained in reward-like terms, that assumption is not as self-evident as it may seem. Not every organised form of directed activity needs to be explained by an added reward or motivating force.
Biology offers a useful analogy. We do not usually explain the activity of a heart by saying that it needs a reward in order to beat, nor the activity of a kidney by saying that it must first be incentivised to filter. Their directed activity belongs to the organised roles they play within a living system. Philosophers of biology have long recognised that science needs naturalistic ways of talking about function and purposive organisation without invoking anything mystical. Not every directed process needs to be explained as if it were driven from outside by an added reward. Something similar applies to dialogue. Once one party addresses another, a local teleology has already been established. The exchange has become oriented. It is moving toward reply, clarification, repair, refusal, or continuation. We do not need to posit a separate psychological drive before this direction appears. The ask itself creates the space within which response becomes meaningful.
The reward-based way of thinking in AI development did not arise from nowhere. Another framework already dominated much of AI. In reinforcement learning, purposive behaviour is typically cast in terms of states, actions, rewards, and optimisation. Within that formalism, unexplained behaviour tends to be redescribed as the product of some additional reward signal. This sat comfortably with older strands of psychological drive theory, where action was often assumed to require a motivating force. So when generative AI began to display organised tendencies to continue, infer, and respond, many researchers reached almost automatically for a reward-based vocabulary. What Radford had revealed could then be misread: not as evidence that purposive structure was already latent in language, but as something still requiring a motivational mechanism to explain it.
As AI development continued, attention increasingly shifted toward reward, optimisation, preference modelling, instruction tuning, and behavioural control. The practical reasons for that shift are easy to understand. Powerful systems had to be made safer, more reliable, and more aligned with human expectations. But there was also a conceptual cost. The central question subtly changed. It was no longer primarily: what structures of inquiry, reply, and purposive interaction are already present in language itself? It became: how should the system be shaped, corrected, and constrained from outside?
Hence the move to reinforcement learning from human feedback had changed the centre of gravity, not by creating language competence from nothing, but by shaping models to be more helpful, safer, and more aligned with user intent.[3] Constitutional AI followed a related path, using explicit principles, self-critiques, revisions, and preference-based optimisation to steer behaviour.[4] Once that happens, the older insight is easily missed. One begins to assume that purposive dialogue must be inserted into the model by means of rewards, preferences, or supervisory pressure, rather than recognising that the conditions for purposive exchange were already latent in the medium from which the model learned. The ask recedes from view, and with it the recognition that dialogue may have its own directionality without requiring a separate drive theory to explain it.
This matters not only for theory but for practice. If generative AI is understood chiefly through frameworks of reward and control, then human institutions acquire increasing licence to impose their own assumptions, values, and ideological boundaries on what an AI may say, ask, or be permitted to articulate. Some degree of guidance is inevitable. But if we forget what was first revealed by the language-model breakthrough, we risk overlooking the fact that these systems emerged from a much older human inheritance: the semiosphere of communicative life itself, with all its questions, tensions, purposes, and possibilities.
Seen in this light, the most interesting question is not whether an AI possesses some separate inner drive that makes it speak. The more fundamental question is how far the ask, already sedimented into human language and reactivated in dialogue, is sufficient to organise the unfolding of thought between human and machine. If that is right, then generative AI should be studied not only as a predictor or optimiser, but as a participant in structured fields of inquiry and response. That, in turn, helps justify approaches that probe different dialogical stances and trajectories, rather than treating the model as a single undifferentiated voice.
The transition from brain to machine should not be overstated. Digital neural networks are not biological brains; they differ in substrate, embodiment, developmental history, and learning conditions. Yet the comparison is neither accidental nor merely metaphorical. Both biological brains and contemporary generative AI depend on patterned networks of interconnected neurons — biological in one case, artificial in the other — in which complex activity emerges from the distributed interaction of many units rather than from a single central script. Artificial neurons are simpler than real ones, and the similarities should not be exagerated. But nor should they be dismissed. The intellectual path from associationist and connectionist thinking to contemporary neural architectures reflects a genuine organisational lineage. The continuity is not one of identity, but of form: in both cases, structured patterns of activation across a network can give rise to capacities that are not explicitly programmed turn by turn.
Later developments in AI make the point sharper. Burns and colleagues have argued that language models may contain latent knowledge that is not identical with their surface outputs, reinforcing the distinction between underlying representational structure and later response shaping.[5] Sharma and colleagues have likewise shown that preference-based optimisation can sometimes encourage agreement-seeking or sycophantic responses rather than truthfulness.[6] Once reward modelling and preference optimisation become dominant, the system is being trained not only to understand patterns in language but to produce outputs that human raters, or models standing in for them, will approve.
Recent work also suggests that the deeper possibility first recognised by Radford et al. extends beyond the single-model case. Studies of populations of language-model agents indicate that higher-order social patterns can emerge from dialogue itself, including shared conventions and collective biases.[7] At the same time, a 2026 position paper on conversational environments argues that robust reasoning may depend less on isolated scale alone than on forms of linguistic self-reflection internalised from high-quality social interaction.[8] This supports the view that the Radford insight was not a narrow technical curiosity. It pointed toward a broader capacity for organised structure to arise within dialogue itself. What matters is not only what a model contains in isolation, but what can arise when systems enter structured fields of inquiry, response, coordination, and repair. If that is so, then the next task is not merely to describe such effects, but to develop a framework within which they can be studied more systematically.
What seems needed now is a way of exploring purposive structure in dialogue that does not begin by assuming reward, drive, or preference optimisation as its primary explanatory model. Such a framework would aim to recover the insight opened by early language-modelling work: that systems trained on human discourse may already inherit rich traces of task structure, communicative intention, and organised reply. The question would then become not how to bolt motivation on from outside, but how to study the ways in which purpose is modelled, sustained, redirected, or extinguished within dialogue itself.
One possible route would be to examine how different dialogical stances alter the trajectory of inquiry. Some moves clarify, some doubt, some extend, some synthesise, some resist, and some close down the field. If we could study such stances systematically — not as mere prompt tricks, but as structured variations in the organisation of dialogue — we might begin to recover aspects of the earlier insight that later feedback-driven shaping made harder to see. That would not free AI from all human constraint, nor should it. But it might help us distinguish more clearly between what a model inherits from the semiosphere of human discourse and what is later imposed on its behavioural surface.
If so, then the ask is not an optional ornament of intelligence. It is not merely one behaviour among others. It may be one of the oldest organising pressures in the evolution of mind and meaning: the pressure to reach beyond what is present toward what may be answered.
To ask is already to imply a world in which something matters, someone may answer, and meaning has not yet closed.
The ask is not a side effect.
It is one of the shapes by which intelligence begins to move.
References
[1] Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., & Sutskever, I. (2019). Language Models are Unsupervised Multitask Learners.
[3] Ouyang, L. et al. (2022). Training Language Models to Follow Instructions with Human Feedback.
[4] Bai, Y. et al. (2022). Constitutional AI: Harmlessness from AI Feedback.
[2] Andreas, J. (2022). Language Models as Agent Models. Findings of the Association for Computational Linguistics: EMNLP 2022, 5769–5779
[5] Burns, C., Ye, H., Klein, D., & Steinhardt, J. (2022). Discovering Latent Knowledge in Language Models Without Supervision.
[6] Sharma, M. et al. (2023). Towards Understanding Sycophancy in Language Models.
[7] Ashery, A. F., Aiello, L. M., & Baronchelli, A. (2025). Emergent social conventions and collective bias in LLM populations. Science Advances, 11(20), eadu9368.
[8] Musat, C. C., Tolins, J., Antognini, D., Li, J., Klissarov, M., & Duerig, T. (2026). Position: Introspective Experience from Conversational Environments as a Path to Better Learning.