Why embodiment is not a hardware problem

When robotics says “embodied AI,” it means sensors plus actuators. When phenomenology says embodiment, it means Being-in-the-world, the constitutive structure by which anything shows up as meaningful at all. These are not two descriptions of one thing at different resolutions. Bolting sensors onto an LLM adds peripherals, not embodiment, and calling the result “embodied” hides the actual research question behind a borrowed word.

Parent discussions

• What does an LLM understand — Stein, Hart, and the wrong question — the same category confusion, there at the level of intellect, here at the level of the body

Connected concepts

• God as Being - The Finite Infinite Gap — the template for the argument: the difference is categorical, not gradational; more of the lower register never sums to the higher
• Christology — the incarnation is the tradition’s maximal claim about what a body is; Chalcedon makes no sense if a body is an instrument package
• Classical Theism — the robotics program repeats the New Atheist error at the level of the body: refuting a position phenomenology never held

The point

Take the engineering seriously first. Physical Intelligence’s π0 and Google’s RT-1 and RT-2 line <citation needed — confirm these are the projects Jacob has actually engaged, not just the famous ones> represent real progress on a real problem: connecting language models to sensorimotor data so that “pick up the cup” terminates in a cup being picked up. The implicit philosophical claim is that this closes the gap between text and world, that grounding accumulates, sensor by sensor, until the model is in the world rather than merely about it. Grounding as an engineering problem of mapping symbols to referents is real and these projects advance it. The philosophical claim riding on top of it is the mistake.

Stein has the distinction that exposes it. In her work on empathy she separates Körper (the body as physical object, extended, locatable, describable from outside) from Leib, the body as lived: not something I have but the “zero point of orientation” from which there is a here and a there, a near and a far, a reachable and an out-of-reach at all. Merleau-Ponty builds the full account on this foundation: the lived body is not an instrument the mind operates. It is the condition under which a world appears, the structure that makes anything show up as graspable, threatening, inviting, mine. The robot has a Körper, impeccably. Cameras, joints, force sensors, a pose in space. What the phenomenological claim requires is Leib, and here is the precise objection: a camera feed gives the model data about a scene; it does not give the model a here from which the scene is arranged. Orientation is not one more data channel. It is what a data channel would have to already presuppose to mean anything.

This is why the difference is constitutive and not gradational, the same logic as the finite/infinite gap, one register down. You cannot add enough sensors to cross from Körper to Leib, for the same reason you cannot enlarge a creature until it becomes God: the difference was never one of amount. Sensor density, modality count, latency: all of it enriches the object-body. None of it produces a lived orientation, because “more coupling to the environment” and “there being a world for someone” are answers to different questions. Calling the result “embodied” is a definition doing the work an argument should do.

The pretense has a cost. As long as “embodiment” names a hardware roadmap, the real research question cannot be stated: what is the relation between meaning and orientation, and is a lived here realizable in an artifact at all? That is a hard and open question, possibly the hard and open question, and it becomes visible only when you stop assuming the answer is “more hardware.” The robotics program is right that text-only models lack something. It has mislocated what.

Open questions

• Is Leib realizable in an artifact even in principle, or does lived orientation presuppose life (metabolism, need, the possibility of death)? Hans Jonas’s question <citation needed — does Jacob engage Jonas directly?>
• Does the zero point of orientation require vulnerability, a body that can be harmed, for anything to show up as mattering, rather than merely as registered?
• What observation would actually distinguish a robot that has a world from one that maps its environment? Is the difference empirically visible at all, and if not, what kind of question is it?