Where Calculation Begins

The sentence meant to work

There is a way of speaking in which the sentence does not come before its effect but after it. First it is settled what shall be achieved; then the formulation that achieves it is sought. Counsel to the prince is built this way, and so is a position in a negotiation, and an answer in a job interview. None of it need be false. It can be true word for word — including the omission that does not lie and the courtesy that covers an intention. The difference does not lie in the truth value. It lies in the fact that truth is here a condition of success and nothing else.

This kind of speech is older than any machine. It never needed a device; it has been at work in courts, chanceries, and counting houses for centuries, and it is built out of the means of language itself. The dispute over machine-mediated writing looks for the border between human and machine in the material: in the words, in the share, in the recognizable texture. It is not there.

Two ways of speaking

In 1981 Jürgen Habermas worked out the distinction at issue. Speech can be success-oriented: the speaker pursues an end, and the person addressed figures in it as a condition of the outcome. It can be oriented toward reaching understanding: the plans of action are coordinated, in Habermas’s wording, not through egocentric calculations of success but through acts of reaching understanding. Both are ways of saying the same sentences. The words do not disclose which one is at work.

Five years later Harry Frankfurt sharpened the diagnosis. The lie is not the breach — the lie still honors the truth by evading it, since it must know what is the case in order to miss it deliberately. The breach is indifference toward what is said, with full attention to its effect. Whoever speaks that way no longer stands in any relation to the truth, not even a hostile one. It has become immaterial, because it contributes nothing to the outcome.

What can be handed over

The Sprachsubstrat — the language substrate: what a language model supplies in statistically probable formulation — calculates nothing. It pursues no interest, it has no addressee it means to move, and no success it orients itself by. It supplies the probable continuation of everything ever written. Calculation is therefore not its way of speaking.

Calculation is not the machine’s way. It is the human way that can be handed over to a machine entirely.

It can be handed over because nothing in it requires someone who means something. Whoever wants only the effect needs no conviction that is held; he needs a surface that lands. That surface is precisely what an apparatus can supply — language with no one behind it, in any quantity, at a cost approaching zero. Calculation is the part of speech in which no one figures who means anything, and therefore the only part that can be surrendered without remainder.

Confirmation from outside

Three remedies are currently on offer. The substrate detector, which scans a text for machine mediation. The certificate, which declares a work machine-free. The proof that a person stands behind an account or a name. Why the first of these fails on its own terms is set out elsewhere: The Witch Trial.

What matters here is what all three share. They make being human a property that is tested, and they hand the test to an apparatus. Their promise is relief: the reader is no longer to decide whom she believes, because a procedure has decided it beforehand. The price stands in the same line. A humanity that must be confirmed by machine has already been surrendered. This is no objection to unmediated writing — whoever works without the substrate carries the empty seal, and in this standard that is the unmarked default position, not the exception in need of proof. The objection is to the form of confirmation, not to the thing it means to confirm.

The naive point

What is possible instead is small and looks naive: an account. How much of a text’s linguistic surface came from the substrate, and how much of the ideas. No confession, no program, no claim about the human — a statement about a matter, given by the one who knows it.

The obvious objection holds: whoever gives the account because it earns her trust has calculated. Honesty that pays is success-oriented speech in the form of understanding. From the outside, honesty cannot be told apart from strategy — and that is why it is declared, not tested. No procedure separates the two cases, because the difference does not lie in the text but in the declarer’s relation to it. What remains is the statement of the one who knows, and a pledge in place of a proof.

The seal therefore certifies no one’s truthfulness. It records an account and stakes the declarer’s reputation on it. Whoever lies in it lies. The seal does not prevent that; it makes the lie nameable and locatable. Signal, not gate.

What is not calculated

The surrender is not a defeat. Calculation is labor, and labor can be handed on — apparatuses are built for it, and there is no loss in their doing it. What does not go along is the point at which someone stands behind a statement: not behind its working, but behind its holding. That point is not a computational step. It cannot be delegated to any computer, because what figures in it is exactly what does not figure in a calculation — someone who means what she says and remains answerable for it.

Calculation is the part that can be handed over. Standing behind what is said is the rest.

More on the substrate term: Substrate, not Machine. On the form in which the account is given: Idea. Why this rest cannot be accelerated either: What Cannot Be Accelerated.


Self-declaration of this note: W 5 · G 3. Position and occasion with the editor, linguistic surface and elaboration substrate-generated; selection, review, and responsibility with the editor. A note on the account gives one about itself.

Sources

  • Jürgen Habermas: Theorie des kommunikativen Handelns (1981) — Frankfurt am Main: Suhrkamp, two volumes. Passage for success-oriented versus understanding-oriented action: vol. 1, Handlungsrationalität und gesellschaftliche Rationalisierung, p. 385. Cited from the German original; the wording quoted here renders that passage.
  • Harry G. Frankfurt: On Bullshit (1986) — First in Raritan 6(2), 1986, pp. 81–100; reprinted in The Importance of What We Care About, Cambridge University Press 1988; book edition Princeton University Press 2005.