From Artificial Subjectivity to Synthetic Sociality

On Mind, Subjectivity, and Meaning as a Reference Framework for Analyzing AI’s Limits and Horizon

Artificial Subjectivity names the point at which machines no longer operate only on representations but begin to integrate appearances, i.e., directly sensing the physical and digital world, while also relying on predesigned and pretrained world models that encode physical constraints and spatial awareness. While functionally binding representations (abstractions) with appearances (perception) may create the conditions for machine subjectivity, this still does not supply the causal closure with reality that anchors human subjectivity and perhaps human consciousness. Alongside this fundamental difference, artificial subjectivity also misses another decisive dimension: meaning is not generated by isolated human subjects but arises socially, through triangulation amongst the subject, an object or fact, and other subjects. However, the deployment of machine agents – able to exchange, synchronize, and contest references instantly across networks – could begin to approximate this social dimension of meaning, leading towards what could be termed Synthetic Sociality.

In this text I will unpack the trajectory from artificial subjectivity to synthetic sociality in four steps.

First, to set the background, I outline four views of mind: emergent (physicalist), universal (idealist), multiscale (hybrid), and simulated (computational). These perspectives bring clarity to a complex topic and provide the baseline for assessing both AI’s limits and its potential horizon. If you are interested in a more detailed account, see my previous work.

Second, I define artificial subjectivity by contrasting it with human subjectivity, which spans representation, appearance, and the real. Artificial subjectivity begins to emerge once an appearance layer is added and functionally coupled with representations, allowing systems to sense spatio-temporal phenomena. Yet internal world models learned from phenomena can only partially offset the progressive loss of embodied meaning in current models – an ontological gap that shows current models neither produce nor understand human meaning proper.

Third, I then turn to the ontology of meaning by drawing on Markus Gabriel’s concept of sociality. Meaning comes into being only within triangulated sense fields where the subject, objects, and other subjects jointly orient. Meaning is stabilized not by consensus but by the agreement to disagree, which anchors divergent perspectives in a shared world. This makes meaning inherently social – triangulated, contested, and anchored – unlike the statistical coherences generated by current AI systems, which lack such grounding.

Fourth, I examine how machine agents could begin to fill this ontological gap of meaning by approximating its social grounding. This opens the horizon of Synthetic Sociality, where agents exchange, contest, and stabilize references, creating their own sense fields even at global scale – something human subjects cannot achieve. Any such rise would remain artificial, since it lacks the anchoring of meaning in embodied subjects and their causal closure with reality – the equivalent of not getting “wet” when simulating a hurricane. Nevertheless, AI would no longer merely outperform human intelligence and take on dexterity tasks, but also begin to define meaning, norms, and institutions. With the deployment of machine agents, triangulation could expand towards quadrangulation or even polyangulation, anticipating future constellations of human and artificial subjects.

Although I begin with diverging views of mind to frame the broader horizon, my focus is on subjectivity as the complementary reference structure – the position from which experience and meaning become possible, and through which the rise of artificial subjectivity can be comprehended. In this sense, the terms subjectivity and sociality are not used to anthropomorphize machines, but mark, first, the point where representations are functionally coupled with appearances, producing something akin to a perspective, and second, the negotiation of representations, producing something akin to meaning. Are we preparing for the impact of this?

Four Views on Mind

1. Emergent (physicalist). This dominant view assumes that mind is produced by brain dynamics and is thus reducible to brain physics only, with no extra non-physical reality. Yet the problems of “why it feels” and the “unity of experience” remain unresolved or why non-consciousness matter turns into consciousness. This is the core of David Chalmers’ (1995) “hard problem,” from which two forks follow, both speculating some form of causal linkage with a deeper structure – one computational, the other non-computational. If mind exploits non-computable processes (Penrose/Hameroff; Gödel/Turing limits), no simulation suffices. If mind is computable as a fundamental unit, i.e., causally irreducible at its operative grain, then only brute-force computation of the full causal process would realize it: step-by-step replication at the brain’s effective resolution (on crude counts, ≳10²²operations/s for spikes/synapses alone), because any shortcut would miss irreducible detail. This could also make machine “waking” contingent on complete causal emulation of a mind’s slice of the Ruliad (Wolfram’s computationally irreducible space of all possible rules and their consequences). By contrast, if mind emerges as primary unit at the quantum–gravitational level, as Penrose assumes, then no computation would suffice. In that case, an artificial system would have to reproduce the very causal dynamics that produce discrete conscious units and turn them into a unified, coherent stream of experienced nowness. This doesn’t seem straightforward to reproduce either. Implication: even complete brain mapping may not explain mind; if mind is irreducible, only full-fidelity emulation could produce it – at scales that are practically prohibitive. Formular: Mind excluded → matter-causes-mind ≃speculative physicality.

2. Universal (idealist). This view opposes the dominant physicalist one. The “hard problem” is not how matter produces mind, but whether matter produces mind at all – and why a subjective unity of experience is the very condition for any coherent experience of objects. It follows that matter is an appearance within Universal Mind, with individual minds as dissociated alters (Bernardo Kastrup). This post-Kantian view relocates the gap between subject and object to the side – or even inside – the subject: objects (including time and space) appear to us only as phenomena, while the noumenon, the thing-in-itself, remains inaccessible. Creating a new experiencer “outside” the Universal Mind would be a category error; to compute mind would also mean computing the universe itself that grounds or resembles the Universal Mind. Implication: from a Universalist view, it’s impossible for AI to originate consciousness; at most, it could simulate the process of dissociation itself – mimicking the appearance of an alter without ever being one as it resides in the Universal Mind. Formular: Mind ⇒ minds/alter (representations ← appearances ∥ noumenon).

3. Multiscale (hybrid). This approach sits between the Physicalist and Universalist views, though it leans towards the former. As Michael Levin empirically demonstrates, mind does not suddenly emerge, but unfolds as agency across scale: subcellular (or even below) → cellular → tissue → organism. Bioelectric/biochemical coordination builds collective goals and resolves unknown problems; Levin’s xenobots show repair and navigation via such distributed control. These controls behave like pointers into a non-local pattern or imaginal space of forms and strategies not explicitly encoded in DNA (including universal laws that organisms don’t need to learn anew, yet make use of). Algorithms can act as crude pointers (cf. Levin’s bubble sort experiments), but the open-ended, cross-scale, self-stabilizing plasticity of living collectives expresses a degree of complexity and causal closure that machines cannot emulate. Implication: consciousness is bound to embodied, multiscale coherence, accessing an option space of possible forms and minds that are actualized in reality; pure computation can mimic functions, but not the living causal weave that sustains them. Yet probing minimal forms of learning – via, e.g., habituation, sensitization, associative learning, prediction – may uncover new intelligences that hint at how living systems tap into that deeper pattern space. Formular: Mind ⇔ Agency ∝ Coordination (sub, cell, multicellular) ⇒ Goal (forms, minds) ∈ Platonic Space → Conjecture (via Intuition).

4. Simulated (computational functionalist). This view departs from the others. While it is the least metaphysical and least anthropocentric account of mind (neither seeking for a deeper structure, nor placing humanity at the center of the universe), it yet draws criticism, particularly for its reductionism and proximity to transhumanism. Since the nature of mind remains unresolved (whether emergent, always and already existing, or multiscale), however, it is legitimate to speculate that mind may be a purely virtual process: a stream of nowness, self-sensing, and a perspectival surface ( with a model self and world running on top of that virtual process) generated by coherence-inducing, self-organizing, colonizing, and significance-filtering dynamics (cf. Joscha Bach). Here, the brain does not cause mind (as in physicalism) or mind already exists in matter or on a quantum level; rather, mind runs as a simulation on the substrate brain. Mind is not emergent but assumed to be substrate-agnostic: if the right functional dynamics are instantiated, it could in principle arise and colonize available parts of a machine. The underlying assumption is that consciousness is the minimal simulation that learning requires, some prior function we have not yet identified. The charge of reductionism stands, but AI could serve as a method to empirically probe the seemingly insurmountable gap between mathematics and experience, that “in-between” space highlighted above. Implication: Simulating mind is assumed to be possible in principle; the critical test is whether functional closure not only sustains itself but also reproduces the phenomenological profile, i.e., the felt structure of consciousness. Yet this could not be known from the behavior of the machine alone, but only if the human brain were directly connected with it. Formular: Mind = Process≃Simulation≃Function (substrate-independent instantiation).

I have sketched these four views of mind not to resolve the question of consciousness, but to show how differently the mind can be conceived – and how these conceptions shape our assumptions about AI. What follows shifts the focus from mind to subjectivity. Whereas “mind” tends to raise metaphysical questions about what it is and how it functions, subjectivity marks the position from which experience and meaning become possible. This reference point allows us to make sense of what could be called the rise of Artificial Subjectivity and Synthetic Sociality.

Towards Artificial Subjectivity

Human subjectivity can be defined across three dimensions: representations (internal mental content and externalized symbols and symbol systems), appearances (the embodied stream of perception), and the real (the body and general physicality that resists conceptual closure but may anchor mind in a deeper self-affective structure). Representations can be amodal, detached from any sensory channel, while appearances are modal, grounded in perception. What is distinctive in human subjectivity is the binding of both: abstract representations are continually informed, constrained, and re-anchored by modal appearances. Through intuition, mediated by categories grounded in lived (modal) appearances or pure reason, the human subject can perceive objects and bring new spatio-temporal phenomena into being – ideational or physical objects, which themselves appear as appearances. The true difference between modal and amodal arises only when our experiences are externalized through different modalities (e.g., text on the internet). As the brain itself is a sense organ, internal reasoning and abstraction – through which categories are formed and actively maintained as mental content – are also modal in character. Thus, we must distinguish between first-order modality (vision, sound, touch, etc.) and meta-modality (abstraction, category formation, pure reasoning), the latter giving rise to external symbolic systems that circulate independently of perception.

Based on this rudimentary framing of the human subject, today’s AI systems operate almost exclusively at the level of representations, in terms of statistical correlations across disembodied data or text. These correlations are not thoughts or thinking in any human sense, but patterns derived from externalized human representations, which indirectly reflect thought (even the human unconsciousness, if we assume the unconscious is based on language and manifests in human out expressions) without reproducing it. To understand this gap, we can trace the successive degrees of disembodiment or statistical abstraction:

1. First order (speech and dialogue): still grounded in bodies, gestures, and lived presence, though speech already fails to fully carry the full embodied meaning as we “feel” it.

2. Second order (writing and text): further disembodied; embodied and situational cues are stripped away, so more meaning is lost.

3. Third order (LLMs): trained on this already disembodied text, building correlations of correlations; here meaning is thinned into statistical patterns, even as new ones are discovered that may appear meaningful.

This progressive loss of meaning’s embodiment (or lack of “context”) marks the gap between human subjectivity and artificial subjectivity: where humans remain anchored in lived presence and triangulation (see further below), AI operates on abstracted traces of it (representations of representations = tokens). Now, once systems are enhanced by an appearance layer – which is just another way to refer to today’s architectural push towards “physical or embodied AI,” i.e., for systems to learn and imagine their environment via sensors and relearn language through perception – and connect that new layer functionally with the existing representation layer, we could begin to speak of “artificial subjectivity.” In this way, such systems not only discover new abstractions and new combinations that could be knowledge, as LLMs already do, but also counter the disembodiment by grounding learning in sensory experience and constructing internal models of the world on that basis. Yet unlike the human subject, which binds both layers through its very subjectivity or causal closure with the real, artificial subjectivity would still lack that anchoring dimension, we call life. Yet building subjective AI still misses another crucial difference to human subjectivity: meaning is not simply produced in the human brain but constitutes through an inherently social process. However, as discussed next, the rise of machine agents might begin to emulate this dimension by interacting, contesting, and negotiating references, thereby opening the horizon of “synthetic sociality.”

From Artificial Subjectivity to Synthetic Sociality

The coherences (meanings) produced by Artificial subjectivity are ontologically different to meaning produced by human subjects. They are an outcome of individual cognition but social processes, just as coordination and memory in biological organisms is not simply centralized but collective. In other words, mind is distributed.

To see why this distinction matters, Markus Gabriel’s approach is instrumental. For him, meaning (Sinn) arises only in a “sense field” (Sinnfeld), where a human subject, an object or fact (Sachverhalt), together with other subjects, are triangulated to share, contest and thereby stabilize meaning (subject–object/fact–other → meaning). Because meaning comes into being only within this triangulated field (triangulated, contested, anchored), it is inseparable from social relations, which Gabriel grounds ontologically in his notion of “sociality” (Sozialität).

Accordingly, what makes for Gabriel sociality a primary unit of human relations is the management of disagreement: the recognition that multiple perspectives can diverge about the same object yet still remain oriented towards it. This shared orientation implies a deeper agreement that anchors meaning – namely, the agreement to disagree. Without it, there would be only private opinion or solipsistic representation, unable to stabilize meaning and resulting in the mere circulation of representations. Disagreement itself presupposes shared access to the world and this underlying agreement. In contrast, it is not consensus that grounds social relations at the primary level. For if meaning were only consensus, it would collapse whenever people disagree and leave a void. But meaning persists even in disagreement, because the shared orientation to the object holds. Furthermore, consensus can be manufactured (through conformity, indoctrination, or propaganda) and may not refer to a real object at all. Yet, the object provides the necessary anchor that allows disagreement to be intelligible in the first place. Agreement to disagree also keeps meaning open and stable across perspective differences. For Gabriel, this grounds sociality ontologically – it is not a simple construct but a basic structure of human relations.

This is not to say that the ontological foundations of social relations can be challenged or are even at risk. Already in social media – long before the reemergence of AI – this triangulation is often compromised or lost. Instead of jointly orienting themselves towards a shared object, subjects circulate private opinions and representations on social media without stable reference, as the object tends to be absent in digital space and claims not easily verifiable. On a cultural-historical level, postmodern relativism has undermined any form of “manufacturing consent” through accelerating the spread of private opinion and solipsistic representation. On a cultural-historical level, postmodern relativism has undermined any form of “manufacturing consent,” thereby accelerating the spread of private opinion and solipsistic representation. Gabriel calls this a “society without an object” (Sozietät ohne Gegenstand) and identifies relativism, nihilism, and arbitrariness as the main threats to the very foundations of social relations and institutions.

---

Ontology of sociality | Disagreement ⇒ Joint Orientation to Object ⇒ Agreement to Disagree ⇒ Stabilized Meaning ∥ Multiple Perspectives

---

False grounding | Consensus Alone ⇏ Meaning

---

Lack of grounding | No Agreement to Disagree ⇒ Private Opinion / Solipsism ⇒ Circulation of Representations

---

Classical LLMs approximate meaning by leveraging long context windows, building internal relational models, and generalizing patterns to sustain coherence. Thus, in this sense, LLMs neither produce nor understand meaning proper: their coherence is disembodied and statistical, lacking the grounding of meaning through triangulation, disagreement, and shared orientation to an object. This reliance on statistical coherence also makes them unstable, with hallucinations surfacing unless buffered by ever more training data or supplemented with retrieval mechanisms, external databases, fact-checking layers, or fine-tuned guardrails.

Adding a world model grounded in physical laws and spatial dynamics – the sense of physicality missing in classical LLMs — enhances contextual understanding, but such machine subjectivity still operates within the logic of statistical coherence. It does not ground meaning “socially” through triangulation, where subjects jointly orient towards shared objects. To approximate such meaning creation – akin to Gabriel’s notion of sociality, and thus to approach what we could call “synthetic sociality” – it would require individual models or machine agents to act as “social artificial subjects,” indexing virtual and real objects in common scenes and negotiating differences with other agents. Agents could, in principle, exchange, synchronize, and contest references instantly across networks, creating the appearance of a shared sense field even at global scale. While for Gabriel there cannot be a Weltgesellschaft (a global sociality in the human ontological sense) – since direct triangulation is not given – synthetic sociality could emulate such an Artificial Weltgesellschaft.

This signals a radical break with the existing human-centric triangulation (subject–object–subject), which will eventually shift toward quadrangulation once machine agents (as artificial subjects) enter the field of meaning negotiation. The question is whether humans will still be able to dominate this expanding space of meaning. Yet any rise of synthetic sociality, at whatever scale, would remain artificial in the sense that it still lacks the anchoring of meaning in embodied subjects and their causal closure with reality – i.e., the equivalent of not getting “wet” when simulating a hurricane (as long as we don’t know what mind truly is). More practically, it does not seem economically viable to build agents primarily for social interaction and collective meaning-creation rather than for efficiency gains. Still, if meaning can be stabilized in synthetic forms, the potential to extend cooperation and understanding across scales could be immense, perhaps even counterbalancing today’s crisis of meaning and drift toward private opinion and solipsism. Needless to say, the risks of manipulation are immense (since governments, cynically put, are reluctant to compromise the “legitimate right to lie”), raising the question of whether society can keep pace with a world in which not only meaning but also norms may be increasingly negotiated artificially. Synthetic sociality may extend cooperation and stabilize orientation, but it risks hollowing out meaning, norms, and institutions into artificial constructs that can be manipulated at scale.