기계 체현과 모달리티

Since 2023, the convergence of artificial intelligence and robotics has moved beyond a generational turnover within the technology ecosystem and now signals a more fundamental event: machine agents have begun to enter the horizon of multisensory meaning-making alongside human beings. This working paper — the inaugural issue of the Modality Lab Working Paper Series — argues that this juncture, which it names the “Physical-AI era” (from July 2023, marked by the appearance of RT-2), cannot be adequately grasped as a mere technological trend. It must instead be re-read as a phenomenon of co-presence: a situation in which human and machine become two mutually dependent embodied subjects that jointly constitute a shared field of Modality.

The paper advances a single overarching claim: human-machine co-embodiment does not extend the human sensorium (as in the McLuhan tradition) but brings a new Modality into presence through the horizontal coexistence of distinct embodied subjects. To substantiate this, it integrates phenomenology (Merleau-Ponty, Ihde), Peircean semiotics, philosophy of technology (Simondon), posthumanist theory (Hayles, Barad), and science-and-technology studies (Suchman) into a four-tier communicative framework — Mode · Modal · Modality · Medium (the “M4 framework”) — and introduces co-modal coupling as the abductive logic that drives the joint sign-action between human and machine. From four diagnostic propositions the paper derives four design principles and a procedural checklist for designers, and closes by projecting the theoretical, practical, and regulatory implications for the Physical-AI era.

The argument proceeds strictly one-directionally — diagnosis → design → implication — and rests only on verified primary-source industry data (2025-2026) and canonical theoretical texts, treating industrial cases (Waymo, Agility Digit, Hyundai-Boston Dynamics Atlas, NVIDIA Cosmos, and others) as illustrative evidence subordinate to the theory.

1. Point of Departure and Method (§1 Framing)

• The founding observation: The 2023-onward coupling of AI and robotics indicates that machine agents have begun to participate, beside humans, in a horizon of multisensory meaning generation. RT-2 (Google DeepMind, July 2023) is taken as one inflection point (§1.1).
• Conceptual extension: The paper extends the concept of embodiment from human embodiment (the phenomenological tradition) to machine embodiment, deliberately distinguished from mere “implementation” so as to foreground corporeality and sensoriness (§1.2).
• Standpoint: The juncture is re-interpreted not as a technology trend but as a phenomenon of co-presence — a shift from tracking events to diagnosing a situation (§1.3).
• Three methodological principles govern the whole paper: (1) theory takes priority, with the six-axis industrial terrain used only as illustrative material; (2) interdisciplinary integration of theoretical resources (phenomenology, semiotics, media theory, STS); (3) conceptual precision — key terms are held stable across contexts rather than allowed to drift.

2. Part I — Present Coordinates (Current State)

The six-axis terrain of machine embodiment (§2). Machine embodiment is mapped across six axes, presented as evidentiary material rather than as the object of the argument:

• Consumer/home appliance axis (§2.1)
• Industrial axis (§2.2) — the largest resource commitment and clearest commercial demand
• Autonomous-driving axis (§2.3)
• Medical/healthcare axis (§2.4) — the most direct coupling with the human body
• World-model axis (§2.5)
• Humanoid axis (§2.6)

The present coordinates of Modality (§3). The paper then locates the Modality that machine embodiment produces here and now:

• Integrated multisensory-channel architecture (§3.1)
• Entry of touch, proprioception, and olfaction into machine sensing (§3.2)
• Reorganization of human-machine interaction Modality (§3.3)
• Reconstruction of spatial/environmental Modality (§3.4)
• The machine Umwelt and the pressure it places on the Modality hierarchy (§3.5)

Part I closes by carrying the analysis from phenomenon to situation — the pivot into the diagnostic core (§4).

3. Part II — Diagnosis (The Four Diagnostic Propositions)

Part II re-defines the situation: from a single-subject Modality to a joint (co-) Modality.

• Diagnostic Proposition 1: Mutually dependent co-Modality of two embodied subjects. Human and machine constitute Modality jointly and interdependently, not as one subject wielding a tool. Grounded in an extension of Merleau-Ponty's flesh (la chair) from the human body to the human-machine coupling (§5.2), a juxtaposition of Ihde's post-phenomenological embodiment relations with Simondon's theory of technical objects (§5.3), and a re-application of Uexkull's Umwelt to the machine (§5.4).
• Diagnostic Proposition 2: Reorganization of the Mode layer and the establishment of the four-tier frame. The lowest tier (Mode: the sensory channel, the “what”) is reorganized once machines sense, grounded in Peircean semiosis (§6.2), yielding the formal Mode · Modal · Modality · Medium hierarchy (§6.3).
• Diagnostic Proposition 3: Horizontal coexistence, not extension. Against McLuhan's extension-of-man model (§7.1), the paper argues for the horizontal coexistence of distinct embodied subjects, drawing on Hayles's posthumanism (the inseparability of information, matter, and body, §7.2) and Barad's intra-action / agential realism (the dissolution of the subject-object dichotomy, §7.3), extending to fleet- and swarm-level coexistence of multiple machine subjects (§7.4).
• Diagnostic Proposition 4: Human-machine co-embodiment brings a new Modality into presence. Read through Peirce's triadic categories (Firstness/Secondness/Thirdness mapped onto exteroception / interoception / meta-mediation, §8.2), with abduction (가추법) identified as the logical driver of co-modal coupling (§8.3), which is then formally introduced as a field of triadic sign-action standing on the Simondonian lineage of couplage (§8.4).

4. Part III — Design (The Four Design Principles)

From the four diagnostic propositions the paper derives four design principles for human-machine systems:

• Design Principle 1: A reciprocal structure, not a hierarchy. Reject the command-and-execution model, following the implications of Suchman's Human-Machine Reconfigurations (§9).
• Design Principle 2: Avoid pre-specifying the mode of coupling. The coupling should be allowed to form dynamically; read through the Clark-Chalmers extended-mind hypothesis, with Medium understood as the environment in which coupling dynamically takes shape (§10).
• Design Principle 3: Explicitly acknowledge asymmetry. The human's living body and the machine's fabricated body are not equivalent; concealing this (e.g., through humanoid anthropomorphism) is a design risk, and the asymmetry should instead be converted into a design resource (§11).
• Design Principle 4: Adopt PBM (Performative Body-Mapping) as a design prototype. Beginning from dancer-robot collaboration (Gemeinboeck & Saunders), the exploration of morphological affordances dismantles the subject-object dichotomy and extends to general HRI design; PBM is shown to correspond structurally to co-modal coupling, so that design answers diagnosis (§12).

Part III concludes with a procedural checklist for designers organized into four inspection bundles — one per design principle — covering hierarchy, coupling mode, asymmetry, and the co-production of the interpretant (§13).

5. Part IV — Implications and Series Position

Three-tier implications (§14)

• Theoretical: Re-definition of the Physical-AI juncture as a co-modal coupling situation.
• Practical: Mutual-alignment design principles for industrial deployment.
• Regulatory: A re-setting of the object of regulation toward the human-machine coupling, read against the EU AI Act and the Republic of Korea's Framework Act on AI (AI Basic Act, in force 22 January 2026).

Series position and roadmap (§15–§16)

• This issue leaves the series its skeletal framework and identifies the next issue, WP 2026-02, on the computational rendering of the Modal layer in latent space (§15.2).
• The series roadmap sequences future diagnoses along the Modal, Mode, and Medium axes (§15.3).
• Conclusion: What this issue demonstrates is the diagnostic precision of the Modality axis; what it leaves behind is the identity and keynote of the series as a whole (§16).

Key Concepts (Glossary)

• Machine Embodiment (기계 체현) — Embodiment predicated of the machine, foregrounding corporeality and sensoriness; deliberately distinguished from “implementation.”
• Co-Modality / Joint Modality (공동 Modality) — Modality constituted jointly and interdependently by two embodied subjects (human and machine), rather than by a single subject.
• Co-modal coupling — (term kept in the original) The abductively driven, triadic field of sign-action between human and machine; the paper's central diagnostic construct, positioned on the Simondonian lineage of couplage.
• The M4 Framework — Mode · Modal · Modality · Medium — A four-tier communicative hierarchy: Mode (the sensory channel, the “what”), Modal (the manner/state, the “how”), Modality (the topmost, total structure of experience), and Medium (the carrier/environment of coupling).
• Abduction (가추법) — Peirce's third mode of inference, identified as the logical engine of co-modal coupling.
• Machine Umwelt (기계 움벨트) — Uexkull's Umwelt re-applied to the machine as its own world of relevance and sensing.
• PBM — Performative Body-Mapping — A performative, dancer-robot-derived design approach (Gemeinboeck & Saunders) adopted as the design prototype for co-modal coupling.
• Physical-AI era / “post-2023” — The paper's temporal coordinate for the diagnosed situation (never framed as an “AGI era”).

Keywords

machine embodiment · co-modal coupling · Modality · Mode-Modal-Modality-Medium (M4) · Peircean semiotics · abduction · phenomenology of the flesh (Merleau-Ponty) · post-phenomenology (Ihde) · Simondon · Uexkull's Umwelt · agential realism (Barad) · posthumanism (Hayles) · human-machine reconfiguration (Suchman) · extended mind (Clark & Chalmers) · Performative Body-Mapping (PBM) · Physical AI · human-robot interaction (HRI)