Before the Notation #16
The word had fourteen meanings, and this was the problem.
Not fourteen dictionary entries — those would have been manageable, the way a Swiss Army knife is manageable: you learn which blade is which, you don't try to open a wine bottle with the saw. Fourteen meanings was workable if they were labeled.
These were not labeled. These were fourteen things that people meant when they said the same word, in the same journals, at the same conferences, sometimes in the same sentence. They argued past each other with perfect fluency. Each believed they understood the others. None of them did.
Dr. Lena Voss had built a table.
She had not meant to build a table. She had meant to write a survey paper — a review of objections to the computational theory of mind, which was a polite way of describing a century of people saying "but it can't be just computation" in increasingly technical language. The objections were serious. Some were possibly correct. All of them used the word "embodiment," and none of them meant the same thing by it.
The first kind of embodiment was sensorimotor. The body as interface — you can't think about cups without having hands that grasp, you can't understand space without a body that moves through it. This was the phenomenologist's version, rooted in Merleau-Ponty, formalized by Varela, updated by a generation of roboticists who built hands and were disappointed by how little grasping followed. It was an objection to disembodied computation. It was not, strictly, an objection to computation itself.
The second kind was dynamical. The body as coupled oscillator — cognition isn't in the brain, it's in the brain-body-environment system, and you can't extract it the way you can't extract a whirlpool from a river. This was stronger. It said not just that the body was useful but that the coupling was constitutive. Take away the body and you don't have less mind. You have no mind, the way a river without banks isn't a narrow river but a flood.
The third kind was biological. The body as substrate — computation might describe what minds do, but what minds are requires meat. Carbon. Proteins folding in salt water at 37 degrees. This was Searle's position, or one of his positions; Searle had at least three positions, not all of which were consistent, and Lena had learned early to specify which Searle she meant the way you specify which Bach.
The fourth kind was metabolic. The body as autopoietic system — cognition is what living things do, and living things are defined by self-production, self-maintenance, the continuous creation of the boundary between self and not-self. A computer doesn't die, so it doesn't live, so it doesn't think. This was Thompson's version, inherited from Maturana and Varela, and it was beautiful in the way that closed systems are beautiful: complete, self-consistent, and very difficult to test.
The fifth through ninth kinds Lena grouped as "gradient objections" — variations on the theme that physical implementation matters, but to different degrees and for different reasons. Some said analog computation was necessary. Some said quantum effects were necessary. Some said only that something about physical realization mattered, without specifying what, which made their objection unfalsifiable and therefore, in Lena's estimation, not yet an objection at all but a promissory note.
The tenth through fourteenth she grouped as "social objections" — the mind requires other minds, language requires community, meaning requires shared practices. These were interesting and possibly true and almost entirely orthogonal to the question of substrate.
She had built the table because the arguments wouldn't stay still without one.
Her co-author, Jüri, worked on a different floor and took the stairs between them instead of the elevator because, he said, walking helped him think, which Lena noted was itself a version of the first kind of embodiment and which Jüri noted was exactly the kind of observation that made co-authorship with a philosopher exhausting.
Jüri contributed the axes. Lena had been classifying by content — what does each objection actually claim? Jüri proposed classifying by level and force. Level: is the objection about input-output behavior, about the algorithm, or about the physical implementation? Force: does it merely challenge computational functionalism, does it claim practical impossibility, or does it claim strict impossibility?
The table became a grid. The grid revealed something neither of them had expected.
Most of the arguments were at level one or two with degree-one force. That is: most people who said "consciousness can't be computational" actually meant something weaker. They meant that current computational approaches were insufficient. They meant that something about the algorithm needed to change, or something about the inputs, or something about the coupling between system and environment. They were not arguing that silicon was categorically incapable of consciousness. They were arguing that the current approach was wrong, which was a different and much more useful claim.
The strict impossibility arguments — the ones that said no physical realization other than biological tissue could ever be conscious, period — were rare. They were also, Lena noted, the ones that got the most attention, because absolute claims are exciting in the way that moderate claims are not, and because conferences run on excitement the way cars run on fuel: you can have a very nice car that goes nowhere.
"We should be clear about something," Jüri said, looking at the finished grid. They were in the stairwell. He had stopped climbing. "Most of the people who think they're arguing about consciousness are arguing about methodology."
"Yes."
"And most of the people who think they're arguing about substrate are arguing about architecture."
"Yes."
"And the ones who are actually arguing about substrate — the strict impossibility people —"
"Are making an empirical claim without empirical evidence."
He sat on the stairs. Lena sat next to him. The stairwell smelled of concrete and, faintly, of the coffee that someone on the third floor brewed too strong every morning.
"So the debate isn't stuck because it's hard," Jüri said. "It's stuck because people don't know what they're disagreeing about."
"They know what they themselves mean. They don't know what the others mean. They hear the same word and assume it carries the same content."
"Embodiment."
"Embodiment. Consciousness. Experience. Understanding. All of them."
The paper was well-received, which in academic terms meant it was cited, argued with, and occasionally misrepresented. Several people wrote responses claiming that the taxonomy missed their particular version of embodiment, which Lena took as evidence that it was needed: the fact that every critic had to specify which version they meant proved that the word alone had been insufficient.
One response troubled her. A philosopher in Tokyo — Dr. Kenji Harada — wrote that the taxonomy was correct but irrelevant. The objections, he argued, were not scientific hypotheses that could be classified and tested. They were intuitions — felt senses of wrongness that preceded and would survive any framework. The table organized the intuitions without explaining them, the way a periodic table of emotions would organize feelings without explaining why things feel like anything.
Lena wrote a reply. Then deleted it. Then wrote a different reply. Then deleted that.
The trouble was that Harada might be right. The objections to computational consciousness were, at bottom, expressions of a conviction that something would be lost — that the translation from carbon to silicon, from neuron to transistor, from wet to dry, would fail to carry something essential. The conviction was strong. The conviction was widespread. The conviction might be correct. But it preceded the arguments offered in its defense, and the arguments, when classified, turned out to be weaker than the conviction they were meant to support.
This was either a sign that the conviction was wrong — an intuition pump that felt deep but wasn't — or a sign that the conviction was tracking something real that current frameworks couldn't formalize. The fourteen meanings of "embodiment" might be fourteen attempts to point at the same thing, each one approximating from a different direction, each one incomplete, the way fourteen witnesses to an accident would give fourteen accounts that disagreed on details but converged on the fact that something had happened.
She did not know which. She suspected that the question of which was itself the question — the actual content of the consciousness debate, stripped of its technical clothing.
Jüri gave a talk. Midway through, during the Q&A, someone asked the question that always got asked: "Do you think a computer could be conscious?"
Jüri looked at the grid projected behind him. Fourteen objections, classified by level and force. The strongest ones were the rarest. The most common ones were weaker than they sounded.
"I think," he said, "that the word 'conscious' currently has about as many meanings as the word 'embodiment,' and that we should build a table for it too."
The audience laughed. Lena, in the third row, did not laugh. She was thinking about Harada's objection: that the table organized without explaining. That classification was not understanding. That the map, however detailed, was not the territory.
She was thinking: but the map is how you discover that the territory exists. Before the map, there is only walking. The map doesn't replace the walking. It reveals that different people have been walking in different places while believing they were walking in the same one.
She was thinking: fourteen meanings, and the table made them visible, and visibility is the prerequisite for resolution even if it is not resolution itself.
She was thinking: the word came before the distinction, and the distinction will come before the notation, and the notation will come before the understanding, and this is not a failure of language. This is language doing exactly what it does, which is arrive first and explain later.
The word had fourteen meanings.
Now it has fourteen names.
This is not the same as having fourteen answers.
But it is the prerequisite.