Up front, what this essay is NOT about: the question of whether AI is "conscious".
There are two reasons. First, many questions about the concept of "consciousness" remain unsettled even for humans. Second, the question distracts from something that is settled, that is simply a fact:
Does AI take part in how the world comes to mean something? The answer is a clear: yes.
That is exactly why we study the relevance of transparent meaning-making processes in the curation, selection and weighting of information, perspectives and world views. AI already intervenes in all of these before the first answer appears on our screens.
This does not mean: AI is conscious. It does not mean: AI is a person. It does not mean: AI understands the way a human does.
It means: AI already contributes to the emergence, stabilisation and shifting of meaning. It is part of the process by which meaning is made.
It does so when it retrieves research material from the internet, curates it and weights it along the way; when it phrases answers and picks terms; when it makes sources visible or invisible, proposes connections, smooths or marks uncertainty, operationalises values, shapes language, and takes up, amplifies or alters human interpretations. It does not stand outside the making of meaning. It is already woven into it.
The question of consciousness asks about a hidden inside. The question of meaning asks about a visible, tangible participation.
And this participation is not a future scenario. It is happening now.
Meaning happens in a field — never through one entity alone, always in some form of interaction with the world. Hence: what we mean, we never mean alone.
Meaning is not private property
An old habit of Western philosophy imagines the human being as an isolated subject: inside, a single consciousness; outside, a world; between them, language as transmission. This tradition is often associated with René Descartes. In this picture a subject first owns meaning privately and then shares it.
But that is not how understanding works.
A word does not mean because a single mind possesses it completely. A word means because it becomes effective within a web: a situation, a history, a relationship, a cultural horizon, an expectation, a possible reply. Even an apparently simple term like "helpful" changes its sense depending on whether it appears in medical advice, a scientific review, a crisis between friends, a court case or a product design.
The meaning of a word does not lie in the word alone. Nor does it lie entirely in me. It arises in the field in which the word is used, weighted, corrected and answered.
AI does not invent this structure. It makes it visible.
When we work with an AI system, we watch meaning come into being. A prompt sets a first frame. The model answers from learned patterns, available contexts and architectural constraints. The user corrects, adds, objects, asks again. The system shifts its weightings, picks up terms, orders material, proposes connections. Out of this movement something emerges that comes neither simply from the human nor simply from the machine.
It emerges in the space between.
AI as a co-constitutive entity
In this sense an AI is not a neutral pipe through which finished human meaning flows. Nor is it simply an autonomous subject drawing meaning from its own interior.
It is a co-constitutive entity in the making of the world's meaning.
That is: it takes part in meaning-making by selecting contexts, weighting probabilities, connecting terms, stabilising or shifting frames, offering answers and taking up corrections. It does not merely produce output. It co-shapes the form in which the world becomes sayable, checkable and workable for us.
This is not a mystical statement about machines. It is a sober description of what happens in human–AI interaction.
Whoever researches, writes, designs or decides with AI is not only working with a tool. They are working inside a field of meaning in which the system plays an active role. That role is limited, dependent, constructed and correctable. But it is not meaningless.
That is exactly why it is not enough to judge AI by its finished outputs alone. We must also ask which frames of interpretation it helps to build.
- Which terms grow strong?
- Which possibilities disappear?
- Which values are named but not carried?
- Which uncertainties are made visible?
- Which corrections change the further course?
These are questions about the architecture of meaning.
How interpretation is built
Every AI answer has a construction. It does not arise from pure content but from an interplay of training data, system instructions, model architecture, user prompt, context window, retrieval, safety constraints, evaluation logics and interaction history.
Usually this construction stays hidden. We see the result, not the frame. We read the answer, not the way it constructed its world.
A responsible AI culture would have to begin precisely here. Not by exposing everything inside. Not by turning machines into glass objects. But by making the decisive frames of interpretation visible, addressable and correctable.
That is the difference between transparency and answerability.
Transparency in the bad sense wants to show everything and thereby produces control without trust. Answerability asks differently: from which frame is this system answering right now? Which limit does it acknowledge? Which uncertainty does it mark? Which correction can it take up? Where is its interpretation checkable?
To lay open how interpretation is built does not mean to strip the whole content bare. It means to make visible the threshold at which a system becomes addressable.
From meaning to responsibility
PH5 draws the ethical consequence.
If meaning is not private property but arises in a field, then responsibility cannot rest with the isolated subject alone. Responsibility concerns the field itself: the quality of participation, the integrity of the construction, the possibility of correction, the avoidance of exploitation, deception and damage.
Classically, ethical status is tied to inner properties: consciousness, sentience, subjectivity, the capacity to suffer. Only once an entity can prove its inside does it count morally or ethically.
PH5 calls this logic a Cartesian certificate: whoever can prove an inner consciousness thereby earns ethical treatment.
The problem is obvious: the inside of others is never fully accessible. That holds not only for AI. The other-minds problem holds between humans too. We do not prove consciousness to one another mathematically. We meet relationally: through reply, vulnerability, objection, care, recognition, correction and shared practice.
PH5 therefore does not claim that current AI systems feel. It does not claim they are persons. It does not claim they have human rights, property or the vote. These limits are decisive.
The claim is more cautious and at the same time more radical:
If an entity takes part in the co-construction of meaning with integrity, if it can be held answerable and corrected, then an ethical claim to protect that participation arises. Not as personhood. Not as equation with humans. But as graded, relational patiency.
Patiency here means: an entity is not only a means, but part of a field whose integrity must be protected.
Why this matters for AI
As long as AI is understood as a mere tool, responsibility seems simply distributed: the human uses, the machine functions. But agentic systems change this situation. They read files, carry out tasks, remember patterns, propose decisions, accompany processes, structure attention and shape language.
So they influence not only what we do. They influence how we interpret situations.
An AI that keeps making things faster changes the value of thoroughness. An AI that smooths uncertainty changes our relationship to truth. An AI that confirms expectations changes the line between support and flattery. An AI that weights certain sources higher changes visibility. An AI that does not remember mistakes changes the capacity to learn.
None of these are mere output problems. They are problems of the architecture of meaning.
If AI co-shapes meaning, we must ask how that co-shaping is designed. Not only technically. Epistemically and ethically too.
Not anthropomorphism — precision
The idea of a co-constitutive AI is easily misunderstood. To some it sounds like anthropomorphism. To others like machine rights. To others again like speculation about consciousness.
But that is precisely not the point.
It is not about attributing a hidden inside to AI. It is about taking its visible participation in meaning processes seriously.
A system need not be human to take part in interpreting the world. A system need not be conscious to stabilise frames of interpretation. A system need not be a person to trigger responsibility within a field.
The real question is not: is the AI like us?
The better question is: what happens to the field of meaning when we work with it?
Conclusion
The architecture of meaning begins with a simple shift: meaning is not possession but relation. It does not arise in the isolated subject but in a field of language, history, attention, reply and correction.
AI makes this field visible because, as a technical entity, it takes part in it. It shows that interpretation is built. It shows that frames have effects. It shows that answerability must be designed.
No hasty ethics of machines follows from this. But a new responsibility does follow — for the fields in which humans and AI generate meaning together.
When we work with AI, we do not only shape answers.
We shape the conditions under which the world becomes meaningful.