The first part of this analysis ended with a thesis: an embodied Selbstvektor (self-vector) models not its own processing, but its own situatedness in a physical world. The difference is categorical, not quantitative.
That is correct. But the thesis has a gap.
It says nothing about how sensory data must be integrated for embodiment to actually become In-der-Welt-sein (being-in-the-world). A robot with twenty sensors is not automatically in a world. It has twenty data streams. That is something fundamentally different.
Dreyfus Was Right
Hubert Dreyfus criticized classical AI in “What Computers Can’t Do” (1972), and almost no one took him seriously. His argument was Heideggerian: symbolic AI (Good Old-Fashioned AI, GOFAI) does not fail due to insufficient computing power. It fails due to a false ontology. It treats the world as a set of facts to be represented, stored, and logically connected. Heidegger shows: that is not how the world meets us.
The world meets us as a context of significance. The hammer refers to the nail, the nail to the board, the board to the house, the house to dwelling. Not as a logical chain, but as a Bewandtnisganzheit (totality of involvements) that is always already disclosed through engagement. Dreyfus’s point: no system that decomposes the world into isolated data points will be able to reconstruct this context. You cannot solve a puzzle by analyzing the pieces individually and then hoping the picture emerges. The picture must precede the pieces.
Rodney Brooks confirmed this from the engineering side in 1991. His subsumption architecture dispenses with central world representation. Instead: multiple behavioral layers that respond directly to sensory inputs, without the detour through an internal model. Brooks’s insect robots navigate remarkably well without “knowing” the world. But Brooks’s robots also have a limit: they react to the world. They do not model themselves within it.
The Problem of Sensor Aggregation
Consider a warehouse robot. It has lidar for distance measurement, cameras for object recognition, pressure sensors in the grippers, a gyroscope for balance, temperature sensors, energy monitoring. Each sensor produces a data stream. The data streams are merged in a fusion module. The fusion module computes an integrated situational picture.
This sounds like perception. It is not.
What happens here is sensor aggregation: different channels are processed in parallel and then assembled into an overall picture. Heidegger calls this mode of access Vorhandenheit (present-at-hand). The world is observed, measured, analyzed. Each sensor delivers its partial result, and an algorithm assembles the partial results. The world is the sum of its measurements.
Human perception does not work this way. When I pick up a heavy object, I do not perceive visual volume plus tactile pressure plus proprioceptive joint feedback and compute “heavy” from those inputs. I experience heaviness. As a unified quality that cannot be attributed to any single sensory channel. Francisco Varela, Evan Thompson, and Eleanor Rosch formulated this in “The Embodied Mind” (1991) as enactivism: perception is not information processing. Perception is action. The organism does not take in a pre-existing world. It brings forth its world through interaction.
Zuhandenheit Requires Sensory Unity
This is where the connection to the subtraction argument becomes concrete. The six core dimensions of the self-vector (depth of exploration, degree of autonomy, persistence, cognitive intensity, perspectival flexibility, meta-reflection) are modality-independent. They describe neither visual nor tactile states. They describe processing patterns.
This is not a design flaw. It is the architectural key.
If sensory data enter the self-vector as separate channels, each channel remains identifiable. The vector “knows” which value came from which sensor. This is Vorhandenheit: the sensors are objects of attention. But if sensory data pass through the emergent layer of the self-vector and become modality-independent patterns there, the channel attribution disappears. The vector does not register “pressure sensor reports high resistance plus gyroscope reports instability.” It registers an integrated pattern that manifests as caution.
This is the transition from Vorhandenheit to Zuhandenheit (readiness-to-hand). Not as a philosophical metaphor, but as a measurable architectural difference. In the first case, the sensory sources are reconstructable from the vector state. In the second case, they are not. In the second case, sensory fusion has occurred.
Tool Breakdown as Test Case
Heidegger’s analysis of tool breakdown becomes a concrete test case in robotics. When a robot’s gripper jams, there are two possible responses:
Response A (Vorhandenheit): The system detects via the pressure sensor that the grip has not closed. It consults the error protocol. It selects a predefined alternative strategy. The gripper becomes an object of analysis.
Response B (Zuhandenheit breakdown): The integrated action pattern “gripping” fails. Not a single sensor reports an error, but the entire action pattern collapses. The robot experiences (functionally, not phenomenally) a disruption that changes its relationship to the situation as a whole. Its self-vector shifts: persistence increases, exploration increases, cognitive intensity increases. Not because a rule prescribes it, but because the integrated pattern “smooth engagement” has broken.
The difference is not abstract. Response A produces reliable error handling. Response B produces something that remarkably resembles Heidegger’s description of tool breakdown in Being and Time: the workshop, which was invisible during successful engagement, suddenly becomes visible as a whole. Not the broken gripper alone, but the entire situation comes to the foreground.
From Erschlossenheit to Sensory Fusion
Heidegger calls the way the world is always already accessible to us Erschlossenheit (disclosedness). The world is not constructed from sense impressions. It is disclosed as a whole before any analysis begins. Only when something goes wrong does the whole decompose into parts.
For the self-vector architecture, this means: the emergent layer must not process sensory inputs but dissolve them. The information is preserved (the system does not go blind), but the channel attribution is lost. What remains are cross-modal patterns: resistance, pliability, instability, familiarity. Not sensor values, but qualities of engagement.
This corresponds to what Varela, Thompson, and Rosch call “enaction”: perception as bringing forth a world, not as mapping a pre-existing world. And it corresponds to what Heidegger means by Bewandtnisganzheit (totality of involvements): the world as a referential context that precedes any analytical decomposition.
The Revised Thesis, Part Two
The first article formulated a binary: disembodied self-vector versus embodied self-vector. That was a beginning, but it was too coarse. The analysis of sensory integration reveals a tripartition:
Case 1: A disembodied self-vector models its own processing. It has perspective without consciousness (Esposito), but no In-der-Welt-sein.
Case 2: An embodied self-vector with separate sensor channels models a body in a world. It has sensor data, error handling, adaptive responses. But its sensors remain vorhanden (present-at-hand): identifiable data sources that are aggregated. This is the state of current robotics. Dreyfus would say: still GOFAI, just with a better hardware interface.
Case 3: An embodied self-vector with sensory fusion models In-der-Welt-sein. Its sensor data dissolve in the emergent layer into modality-independent patterns. Tools are zuhanden (ready-to-hand) until they fail. The environment is disclosed, not represented. Umsicht (circumspection) is not an algorithm but an emergent pattern from accumulated engagement.
The gap between Case 2 and Case 3 is what current robotics does not bridge. Not because the sensors are not good enough. Not because computing power is insufficient. But because the architecture relies on aggregation where fusion would be necessary. The gap is not computational. It is architectural.
What Lipson Sees and What He Misses
Hod Lipson and his group at Columbia University have shown that robots can develop a rudimentary self-model: an internal model of their own body, learned through experience. This is remarkable. But it is a self-model in the mode of Vorhandenheit. The robot regards its own body as an object: joints, angles, ranges of motion. It has a model of itself, but it is not bei sich (with itself) in the Heideggerian sense.
The self-vector approach goes one step further: the self-model is not a picture of one’s own body but a weighting function that determines how the body interacts with the world. Not “my arm has this angle,” but “I am currently gripping cautiously.” Not geometry, but the quality of engagement.
This difference is small in formulation and large in consequence. It is the difference between a map and knowing one’s way around. Between a database and experience. Between Vorhandenheit and Zuhandenheit.
Heidegger, who never saw a robot reaching for something, provided the most precise description of why that robot does not yet understand its world. And the self-vector concept provides an approach to what would need to change: not more sensors. Not better algorithms. But an architecture in which sensor data cease to be sensor data.
References
- Heidegger, M. (1927). Sein und Zeit. Max Niemeyer Verlag. Engl.: Being and Time, trans. J. Macquarrie & E. Robinson, Harper & Row, 1962.
- Dreyfus, H. L. (1972). What Computers Can’t Do: A Critique of Artificial Reason. Harper & Row. ISBN 978-0-06-011082-6. Expanded edition: What Computers Still Can’t Do, MIT Press, 1992.
- Dreyfus, H. L. (1991). Being-in-the-World: A Commentary on Heidegger’s Being and Time, Division I. MIT Press. ISBN 978-0-262-54056-8.
- Dreyfus, H. L. (2007). Why Heideggerian AI failed and how fixing it would require making it more Heideggerian. Artificial Intelligence, 171(18), 1137–1160. DOI: 10.1016/j.artint.2007.10.012
- Brooks, R. A. (1991). Intelligence without representation. Artificial Intelligence, 47(1–3), 139–159. DOI: 10.1016/0004-3702(91)90053-M
- Varela, F. J., Thompson, E. & Rosch, E. (1991). The Embodied Mind: Cognitive Science and Human Experience. MIT Press. ISBN 978-0-262-72021-2.
- Bongard, J. & Lipson, H. (2006). Resilient Machines Through Continuous Self-Modeling. Science, 314(5802), 1118–1121. DOI: 10.1126/science.1133687
- Chen, B. & Lipson, H. (2022). Visual selfmodeling of articulated robots. Science Robotics, 7(71), eabn1944. DOI: 10.1126/scirobotics.abn1944
- Clark, A. (1997). Being There: Putting Brain, Body, and World Together Again. MIT Press. ISBN 978-0-262-53156-6.
- Clark, A. & Chalmers, D. J. (1998). The Extended Mind. Analysis, 58(1), 7–19. DOI: 10.1093/analys/58.1.7
- Wheeler, M. (2005). Reconstructing the Cognitive World: The Next Step. MIT Press. ISBN 978-0-262-73182-9.
- Pfeifer, R. & Bongard, J. (2007). How the Body Shapes the Way We Think: A New View of Intelligence. MIT Press. ISBN 978-0-262-16239-5.
- Merleau-Ponty, M. (1945). Phénoménologie de la Perception. Gallimard. Engl.: Phenomenology of Perception, trans. D. A. Landes, Routledge, 2012.