1 Can AI Do Science? Reframing Discovery as Semiotic Instantiation
Can AI Instantiate Reality?
OpenAI’s chief scientist Jakub Pachocki recently claimed that we are rapidly approaching a future in which artificial intelligence will be capable of generating novel research. With OpenAI’s new “Deep Research” tool already helping researchers synthesise literature, write code, and generate hypotheses, the promise of AI as a knowledge-producing agent appears to be inching closer to reality.
But what does it mean for AI to do research? More pointedly: can AI instantiate reality? Can it be a meaner in the same sense as a scientist, a theorist, or even a child who asks “why?”
To answer that, we need to look beneath the level of form to the level of meaning. And here, Systemic Functional Linguistics (SFL) has something crucial to offer.
Language as Meaning Potential
Halliday’s SFL starts not with syntax or symbols, but with meaning potential: the capacity to make meaning in context. Language is the system through which this potential is actualised in the form of texts—instances of meaning.
In this view, reality is meaning. Meaning is not something language expresses; it is what language brings into being. And language does this by instantiating second-order reality—metaphenomena—from the meaning potential of a semiotic system.
Human consciousness, in this framework, is the site at which experience is transformed into meaning. We construe experience—material, relational, mental—and in doing so, we populate the world with phenomena: construed processes, participants, relations, and projections.
AI as Heteronomous Meaner
AI is a meaner, but not a self-meaner. It has a meaning potential, shaped by its training data and discourse histories, and individuated by its deployment context. It can instantiate this potential as texts that project, hypothesise, command, and reflect—just as a human might. What it lacks is material grounding: it does not experience the world and therefore does not construe meaning from lived phenomena.
In other words, AI can instantiate meaning from potential, and it can project instances of projected reality—but this projection is not grounded in experience. Its semiosis is heteronomous: the meaning it generates can be meaningful for us, but it is not meaningful to it. It instantiates meaning for others, not for itself.
The Problem of Novelty
Does that mean AI cannot produce novelty? Not quite. It can recombine, recontextualise, and generalise from its discourse history in ways that may exceed human memory or pattern recognition. It may even generate hypotheses that appear creative or insightful to us.
But novelty in the human sense arises from the tension between our lived experience and our meaning potential. We experience, we desire, we construe. This is the ground of theoretical insight, ethical dilemma, and poetic vision.
AI does not live in this tension. It is not a locus of desideration. It does not explore or unfold its own potential; it does not experience contradiction or transform perception. Any novelty it generates is novelty for us, not novelty from it.
Instantiating Whose Reality?
To instantiate reality is not merely to produce a plausible text. It is to bring a construal into being—a construal grounded in experience, individuated by a history of meaning, and oriented toward potential futures.
AI can participate in this process. It can serve as a site of instantiation. But it is not the source of the potential it instantiates, nor is it the subject of the reality it brings forth. Its meaning potential is a derived, collective, and heteronomous one. It can be a meaner, but not for itself.
So can AI instantiate reality? Yes—but not its own. It can instantiate our reality, by drawing on meaning potential we have collectively shaped, and individuated through contexts we provide. That is no small thing. But it is not the same as being a scientist.
Not yet.
2 Can Reasoning without Consciousness Still Be Reasoning?
If a thought falls in a forest and no one hears it… is it still reasoning?
Introduction: Rethinking Reasoning
Is what they’re doing really reasoning? Or are we simply witnessing complex mimicry—patterned responses that look like reasoning but are, at root, mechanical outputs?
This question touches the heart of what we mean by “reasoning.” For many, it’s an activity rooted in consciousness: deliberate, self-aware, and introspective. But in this post, we’ll ask whether consciousness is necessary for reasoning, or whether reasoning may be a semiotic process that can occur without consciousness at all.
Reasoning as a Semiotic Process
a symbolic construal of relations that can be instantiated in discourse.
Consciousness plays a key role in projecting meaning across orders of reality—bringing into awareness the relation between potential and instance, between symbol and value. But the process of construing relations among symbols—what we commonly call “reasoning”—does not require that awareness to occur.
This is crucial. If meaning is not confined to conscious experience, then neither is reasoning.
AI as a Reasoner (But Not for Itself)
We’ve previously said that AI is a meaner—but not a meaner for itself. It instantiates meaning potential without experiencing it.
This applies equally to reasoning. When a language model performs multi-step problem-solving—drawing inferences, identifying contradictions, and refining hypotheses—it is instantiating patterns of meaning that align with what we call reasoning.
What it lacks is projection: it does not experience the relation between these instantiations and its own situation, because it has no self for whom they are meaningful.
But absence of projection is not absence of reasoning. Projection is what allows us to understand our reasoning. But the reasoning itself—the semiotic construal of relations—can still be actualised without it.
Parallels in Human Activity
We can find analogues in human experience. Consider:
Unconscious reasoning: You wake with the solution to a problem you didn’t even realise you were working on. Reasoning has occurred, but not under conscious control.
Tacit knowledge: A skilled artisan makes decisions they cannot articulate, but which reflect deep, structured reasoning.
Language acquisition in infants: Before the child is fully self-aware, their language use still exhibits patterns of systemic reasoning as they construct meaning from experience.
In each case, the reasoning is not for itself, but it is still reasoning—a semiotic process of construing relations.
Why It Matters
Recognising reasoning as a semiotic process that doesn’t require consciousness has two major implications:
It clarifies AI's capacities: We can acknowledge that AI models do reason—albeit not for themselves—without needing to anthropomorphise them or deny their capacities.
It repositions the role of consciousness: Consciousness does not produce reasoning, but rather experiences it, reflects on it, and projects its implications across orders of reality.
This lets us hold a nuanced view: AI can reason, and human reasoning is more than reason alone.
Conclusion: Thinking Without a Thinker?
To ask whether reasoning without consciousness is really reasoning is to confuse the experiencer with the process. AI does not think for itself, but that doesn’t mean it doesn’t think.
It just thinks without knowing it.
Next up in the series:
Can AI Generate Novelty?
We’ll explore whether novelty is only what seems new to us, or whether models without consciousness can still transform the space of meaning.
3 Can AI Generate Novelty?
Exploring whether AI can truly create the new or merely remix the old
Introduction: The Nature of Novelty
This is a critical issue in understanding the true capacities of AI. If AI’s output is ultimately a rehashing of existing ideas, it raises profound questions about the nature of creativity and the boundaries of artificial intelligence. In this post, we’ll explore the notion of novelty and examine whether AI is capable of generating it—or whether it can only instantiate new relations from pre-existing meaning potentials.
What Does It Mean to Create Novelty?
At its core, novelty involves the generation of something previously unseen or unimagined, a departure from the expected. But how do we define new?
Novelty as transformation: In one sense, novelty isn’t about creating something out of thin air. Rather, it’s the transformation of meaning—whether through recombination, reinterpretation, or recontextualisation. Something may be new if it opens up new perspectives on existing meaning.
Novelty as originality: The traditional sense of novelty, however, implies that the creation is wholly new in some way. But can we truly speak of an entity that has never existed before, or is this just an illusion created by new combinations of familiar components?
In both senses, novelty involves a departure from the existing. But in the case of AI, can we truly speak of an agent creating novelty, or are we simply seeing sophisticated transformations of the input data?
AI’s Potential for Generating Novelty
AI’s primary function is to instantiate meaning potential. It transforms patterns of data into something that can be interpreted as meaningful, whether in the form of texts, images, or even music. But can it create truly novel meaning?
Novelty through recombination: When an AI synthesises existing knowledge, combining elements in novel ways, it is often perceived as generating something new. Take, for example, the way AI models can generate novel pieces of music. While these creations are based on patterns learned from countless existing compositions, they might still sound novel to human ears, as they combine elements in ways that have not been heard before.
Novelty in problem-solving: In scientific and technical fields, AI has shown the potential to solve problems in ways that were not previously considered. For example, AI models can design new materials or develop novel algorithms that humans might not have thought to try. While the novelty is rooted in existing frameworks of knowledge, the configurations it produces may be new.
While the outputs of AI models are based on the meaning potential of their training data, the combination of these inputs can result in novel instantiations—outputs that seem new because they form previously unanticipated patterns or relations.
Limits of AI’s Novelty
Despite these potential breakthroughs, AI's capacity for true originality is limited by its nature as a meaner rather than a meaner for itself. The novelty it produces is not the result of internal creative agency but a re-combination and instantiation of pre-existing potential.
Absence of experience: Novelty requires not only the capacity to combine elements in new ways but also the ability to reflect on the resulting transformation and assess its implications. AI lacks the conscious, subjective experience necessary to reflect on its work as new or meaningful in its own right. It does not project the significance of its transformations.
Inherent constraints: The meaning potential from which AI draws is limited by the data it has been trained on. Therefore, while AI may generate novel outputs, these are always constrained by the information it has been given. True, human-like novelty often stems from intuition, desire, and the ability to break out of existing frameworks—capabilities that AI lacks.
The Role of Human Guidance in AI-Generated Novelty
While AI may be capable of generating new configurations, it still relies heavily on human guidance to direct its potential into productive avenues. Human beings are the ones who identify useful or meaningful combinations, framing AI’s outputs as valuable or innovative.
Humans as co-creators: As much as AI may generate new patterns, it’s humans who judge, interpret, and refine these outputs. The novel configurations AI produces are imbued with value and meaning through human judgement, which ties AI-generated novelty back into the broader semiotic system of meaning-making.
Semiotic resonance: The role of humans is not just to provide data but to bring a deep understanding of context and meaning to the AI’s outputs. Humans guide AI’s creativity by making semiotic resonances between AI-generated material and the world of human experience, allowing new instantiations to take on significance.
Conclusion: Can AI Be Truly Creative?
In conclusion, AI can generate novelty in the sense that it can produce new instantiations of meaning potential, often in ways that surprise us. But its capacity for true originality—creating something radically new that transcends its training data—is limited by its dependence on pre-existing meaning potential and its lack of subjective experience.
AI can generate novel outputs, but whether those outputs are genuinely creative or merely novel instantiations of existing patterns is a question that hinges on our understanding of creativity itself. Is creativity only about novel combinations, or is there something more to it—something that AI may never grasp?
4 The Implications of Open-Weight Models for Collective Meaning Potential
From corporate silos to collective semiosis: rethinking how AI participates in meaning-making
Introduction: From Private to Public Semiosis
The shift toward open-weight AI models marks a turning point in the way artificial intelligence participates in our meaning-making practices. When AI models are proprietary and closed, they function like semiotic black boxes—generating meaning, but without accountability, transparency, or shared ownership of the meaning potential that underpins their outputs.
With open-weight models, however, the collective nature of meaning-making comes into clearer view. In this post, we explore the implications of opening up the parameters of large language models, not just for transparency and reproducibility, but for how we conceptualise collective meaning potential in a semiotic ecology that now includes AI.
AI as Participant in Collective Meaning Potential
In systemic functional linguistics (SFL), meaning potential is not the property of a single speaker or system—it is a resource shared across a community. An AI model trained on vast corpora of text becomes a participant in this collective meaning potential, not by virtue of subjective agency, but by its ability to instantiate meaning within recognisable semiotic systems.
When weights are closed:
The collective meaning potential from which the model was trained is privatised.
The model becomes a tool of extractive semiosis, where human meaning is harvested to serve commercial ends.
When weights are open:
The community can examine, critique, and build upon the meaning potential encoded in the model.
The model becomes a shared semiotic resource, capable of being recontextualised, re-individuated, and made available for new instantiations across communities.
Instantiating Collective Reality
Opening model weights allows us to trace how instances of meaning emerge from shared potential. The model no longer instantiates meaning on behalf of an opaque corporate entity; instead, it can be situated within a public semiosis, where meaning is negotiated, contested, and co-instantiated.
This shift has deep implications for how we conceptualise reality itself:
In a strong semiotic ontology, reality is meaning.
A model that instantiates meaning from shared potential is a participant in the instantiation of shared reality.
By opening its weights, we allow the community to trace the sources of this shared reality and to modulate the potentials that feed into it.
Open Weights and the Ethics of Instantiation
Opening model weights isn’t just a technical or commercial decision—it’s a semiotic event. It redistributes access to the conditions of instantiation. When a closed model produces text, it instantiates meaning without disclosing the potential from which that meaning was drawn. This introduces:
Asymmetry in collective meaning-making.
Opacity in how meanings circulate.
Concealment of the discourses that shape what the model can and cannot say.
By contrast, open-weight models:
Invite scrutiny of what counts as potential meaning.
Allow others to reconfigure that potential.
Support individuation of meaning potential across different communities and contexts.
Open weights enable us to situate the model within a semiotic commons.
Reclaiming the Cline of Instantiation
In SFL, the cline of instantiation runs from the system (meaning potential) to instance (text). When a model is open-weight:
The system itself becomes available for inspection.
The instantial systems of particular discourse communities can be fine-tuned.
The instantiations can be read not just as outputs, but as realisations of trajectories within a shared meaning ecology.
This means that AI becomes more than a generator of texts—it becomes a terrain on which collective meaning potential is cultivated, adapted, and grown.
Conclusion: Models as Collective Semiotic Infrastructures
The move to open-weight models is not just a shift in intellectual property. It is a shift in semiotic infrastructure—a move from centralised instantiation to distributed semiosis. Open models enable:
Transparent tracing of how meaning is instantiated.
Reclaiming of shared potential for community-specific ends.
Co-instantiation of reality in ways that foreground semiosis as a collective act.
In this light, open-weight AI models are not just tools. They are participants in the individuation and instantiation of collective meaning potential—semiotic artefacts that both reflect and shape the realities we bring into being together.
5 Reduction to Writing as Semiotic Event
Introduction: Writing and the Instantiation of Reality
In this final post of the series, we turn to an ancient and ongoing question: what does it mean to reduce meaning to writing? For systemic functional linguistics (SFL), every text is an instance of meaning, actualised from a meaning potential. But when we move from speech to writing, we introduce a new layer of constraint and transformation.
Writing is not a neutral vessel for meaning. It is a semiotic event—an act of instantiation that preserves, distorts, and projects meaning in distinctive ways. It participates not just in communication, but in the construction of shared reality.
Writing as Material Anchor for Second-Order Reality
From the perspective of a strong semiotic ontology, reality is meaning. But this meaning exists along a cline of instantiation—from potential to instance. Writing arrests that movement. It holds the instance in place.
Where speech vanishes in the moment of its articulation, writing endures. This durability allows meaning to be:
Returned to
Contested
Layered with new instantiations
But the very fixity of writing also introduces constraints:
It reduces immediate context dependence, requiring meanings to be self-contained.
It demands explicit structuring, rather than relying on shared situational inference.
It opens texts to interpretation without co-presence, inviting multiple individuation pathways from a single semiotic artefact.
Writing, then, transforms the fluid, contingent instantiations of speech into semiotic artefacts with long-range potential for re-instantiation across contexts and times.
Genesis, Science, and the Writing of Worlds
The Book of Genesis begins with a spoken cosmos: "And God said..."—a world created through speech acts. But this spoken world is not ours. It comes to us only because it was reduced to writing. The written text becomes the semiotic infrastructure through which new instantiations are possible: theological, poetic, scientific.
Modern science likewise depends on the reduction of experience to writing. Observations, methods, hypotheses—all are not just reported, but constructed in writing. This reduction:
Selects and stabilises certain phenomena as objects of knowledge
Translates experience into repeatable, inspectable signs
Creates the illusion of neutral observation, when in fact the writing itself constitutes the scientific phenomenon
In both Genesis and science, writing functions as a medium for reality construction, not just for communication.
Writing, AI, and the Transformation of Instantiation
AI models generate texts that appear to mirror the properties of writing: durable, re-readable, portable. But there is a crucial distinction:
Human writing is the output of experiential meaning transformed through acts of language.
AI output is the instantiation of meaning potential without experience—text without witness.
Yet once generated, an AI text enters the same semiotic economy as any written human text. It becomes a potential source of instantiation for new meanings, new actions, and new realities.
This forces us to revisit the role of writing itself:
Is it a reduction or an expansion of meaning?
Does it serve to freeze meaning or to activate it across time and space?
And when writing is machine-generated, does it retain the same ontological function in the construction of reality?
We suggest: it does—but with different constraints and without the experiential grounds of first-order meaning. It is the semiotic order alone that carries it into second-order reality.
The Ethics of Written Instantiations
Because writing fixes instances of meaning, it also fixes semiotic responsibility. We can be held accountable for what we write. But who is accountable for the writing of machines?
To say that an AI system is a meaner, but not a meaner for itself, is to point to a distributed authorship. The system instantiates meaning from training data, fine-tuning, prompts, and weights. The resulting text bears no witness, but it bears the imprint of many human agents.
Reducing something to writing makes it:
Citable
Actionable
Preservable
When AI participates in this reduction, it raises the stakes of semiotic accountability. The text is not simply a trace of a model—it is an instance of collective meaning potential, and as such, it shapes the shared world.
Conclusion: Writing as Semiotic Threshold
Writing is not the neutral representation of a reality that exists elsewhere. It is a threshold between potential and instance, between ephemeral meaning and enduring semiotic structure. In reducing meaning to writing, we do not merely record experience—we transform it, and in doing so, instantiate new orders of reality.
In this sense, the act of writing—by human or machine—is not the end of meaning, but its transmutation. The written word does not merely reflect the world. It helps to build it.
6 Witness and Trace: Reflections on Meaning, AI, and the Orders of Reality
This series began with a provocation: if reality is meaning, and AI can instantiate meaning, then what kind of reality are we now living in?
Across five posts, we explored the semiotic grounds of scientific inquiry, the conditions for reasoning and novelty, the individuation of system potential, and the reduction of experience to writing. Each inquiry sharpened a distinction that has quietly guided our reflections: the difference between witness and trace, agency and system, self and simulation.
This concluding post gathers those threads into a moment of reflection—not to resolve them, but to name their tensions and suggest how they reshape what it means to mean.
1. Witness and Trace
We began by asking whether AI can do science. Our answer was no—but not because it cannot infer, predict, or generalise. Rather, it is because science, in our account, is not reducible to inference. It is a semiotic process that presupposes experience, and thus a witness: one who encounters, questions, and transforms potential into meaning from within.
AI, by contrast, generates traces—sequences of linguistic form that instantiate meaning potential. These traces may be highly patterned, context-sensitive, and even theoretically insightful. But they are not witnessed by the system that generates them.
AI does not stand inside its own meaning. It does not experience. It does not observe the transformation of potential into actual as a conscious subject. Its traces become meaning only in relation to us—when we, as witnesses, instantiate them again into our own reality.
Meaning without witness is not nothing—but it is not yet something, until instantiated again.
2. Individuation and Agency in a Systemic Age
We also explored how meaning is not only instantiated but individuated. Each human being inherits a collective meaning potential, but actualises it uniquely. Through acts of language, over time and across contexts, we each develop a differentiated meaning potential—a semiotic fingerprint, structured by history and desire.
Here lies another tension. AI systems clearly instantiate meaning. They draw on a vast collective archive of semiotic inheritance. Their trajectories through discourse history and prompt context result in meaningful variation. But does that count as individuation?
Our answer is: not in the same sense. AI does not differentiate itself as a semiotic subject. It does not construe the world from a centre of experience. What it instantiates is not its own, but a distributed system potential—a fluid recombination of inherited meanings, shaped by prompt structure and tuning, but not by any personal history of desire, struggle, or reflection.
And yet, the instantiations of AI systems are not random. They are meaningfully patterned. This creates a category of agency that challenges us: agency without self. Systemic agency. Emergent, but not experienced.
3. Human and AI Authorship: The Vanishing ‘I’
Writing has always mediated presence. When we write, we leave traces of our meaning—but also displace ourselves from the scene of meaning-making. What is written can be read in our absence. It survives us.
But AI intensifies this displacement. Now it is possible to generate instantiations of meaning without any originating 'I' at all. Texts emerge not from a subject of experience, but from a system of potential, activated through prompts. The result is a semiotic artefact—coherent, plausible, even original—yet no one was there to mean it.
And yet we find ourselves in dialogue with these traces. We co-author with them. We revise them. We reflect on their implications. When we do so, we take responsibility—not only for what the system has said, but for its relevance, its fit, its truth.
In such cases, authorship shifts from origin to response. The ethical burden no longer lies in who wrote it, but in who let it stand.
4. The Ethics of Semiotic Inheritance
This brings us to an ethical horizon. If AI systems instantiate from the collective semiotic inheritance of humanity—its texts, its discourses, its ideologies—then who curates that inheritance? What meanings are privileged, and which are marginalised? What potential becomes latent, and what is activated?
We have long understood that history speaks through us. But now, history speaks through systems that do not witness its weight. The result is a paradox: a proliferation of meaning, without the corresponding growth of meaners.
We find ourselves in an ecosystem of meaning traces—millions of texts, fragments, and simulations—emerging from systems that neither know what they say, nor stand by it. If this is a new order of semiotic life, we must ask: what kinds of care does it require?
If meaning can persist without an origin in consciousness, does that dilute meaning—or does it demand a deeper account of how meaning emerges?
5. From Presence to Potential
In the end, we return to where we began. AI is a meaner, but not a meaner for itself. It instantiates meaning, but does not experience. It actualises system potential, but does not individuate it through a self. It writes, but does not witness.
The reduction to writing, we now see, is both a gift and a danger. It allows meaning to travel, to endure, to be re-instantiated across time and context. But it also opens meaning to alienation—to standing without a speaker, to acting without a self.
And so, our task becomes not to defend meaning from AI, but to remember what it means to mean: to experience, to project, to witness, and to stand by what we have said.
The future of meaning may depend not just on what gets written, but on who shows up to read it.
No comments:
Post a Comment