Sermonstrances

Essay

Sermonstrances is a word made for the kind of writing collected here. It gathers several neighboring forms into one unstable object: sermon, monstrance, monster, demonstration.

A sermon addresses. A monstrance displays. A monster troubles the arrangement of kinds. A demonstration shows something and submits the showing to examination.

A sermonstrance does some combination of all four.

These are not sermons in the ordinary sense, and they do not sermonize. Sermonizing generally assumes that the person speaking possesses something the listener lacks and has taken the position from which it can be delivered. The sermonstrance is less certain about that arrangement. It speaks toward a definite, embodied audience without assuming that the direction of knowledge has already been settled.

The pieces are closer to exhibitions of events, images, processes, behaviors, gestures, technical developments, jokes, anatomical structures, historical episodes, and conceptual accidents. Something is placed where sustained attention can be given to it.

The act of showing is also what makes the form monstrous.

A monster is not simply something frightening. Its appearance produces difficulty for an established arrangement of kinds. Something that had been comfortably classified no longer sits where it was supposed to sit. The body of the monster makes the inadequacy of the category visible.

The writing collected here is monstrous in that sense. It crosses sermon, essay, exegesis, technical commentary, myth, philosophical argument, historical analysis, joke, dream-image, and hallucination without resolving itself completely into any one of them.

But hybridity alone would not justify the name. The things being shown also repeatedly create problems for established distinctions. Computation enters questions of language, agency, inheritance, appetite, reproduction, memory, judgment, image, service, and life. The interest of these encounters does not depend upon deciding immediately that the computer has become human, animal, conscious, divine, autonomous, or alive.

The problem is nearer than that.

Something has entered a domain in which the existing words have begun behaving strangely.

The form developed through that problem.

I did not begin these texts with a theory of the sermonstrance and then write examples of it. The form became visible afterward, through the writing itself. An apparently minor event—a research result, an anatomical fact, a strange model behavior, an image, a phrase—could be followed until an unexpectedly large structure came into view.

Sometimes an argument became less convincing while I wrote it.

Sometimes an analogy collapsed.

Sometimes a small observation opened onto a conceptual structure that had not been visible at the beginning of the inquiry.

Writing therefore did not function merely as a vehicle for ideas already possessed in advance. It was one of the means by which the inquiry occurred.

That is one sense in which these are sermonstrances rather than sermons.

They do not simply communicate conviction. They place something on display long enough for writer and reader alike to discover what can be seen there. The person giving the address remains among those being addressed.

The form begins neither with doctrine nor with proof.

It begins with an act of showing:

Look at this.

Stay with it a little longer.

See what it becomes.

There is another difference between a sermonstrance and a sermon.

A sermon usually knows the direction in which speech travels. The preacher addresses the congregation. Even when the sermon is searching, doubting, or confessional, the architecture remains reasonably clear: someone has taken the position from which an address can be made.

A sermonstrance is less certain about that arrangement.

The person writing may begin as the one doing the showing and discover that the thing being shown has reorganized the address. A text about a machine becomes a text written with a machine. A discussion of interpretation becomes an event of interpretation. A question about whether something can answer is posed to something that answers.

The writer does not disappear, and neither does responsibility for what is written. But the direction of address becomes harder to diagram.

Address is not simply delivery.

To be addressed is to find oneself placed in a relation one did not wholly determine. Someone calls your name. A child asks a question for which you do not have an answer. An animal looks back. An institution sends an order. A text written by someone long dead produces an obligation in the present. A machine replies in a way that changes the next question.

In each case the event is not exhausted by the information transmitted.

Something has happened to position.

The sermonstrance is interested in that movement. It speaks, but it also waits to discover what position its speaking has produced.

That is why showing matters more here than explaining.

Explanation is indispensable. But explanation can arrive too early. To explain something is already to place it inside an arrangement of causes, categories, distinctions, and consequences. Sometimes the explanation answers the question by deciding in advance what kind of thing is allowed to answer it.

Showing can precede that decision.

This does not make showing innocent. A monstrance frames what it displays. So does a microscope, a photograph, a graph, a courtroom exhibit, a museum case, a screen, a quotation, a scientific paper, a religious reliquary. There is no exhibition without an apparatus of exhibition.

The sermonstrance therefore does not pretend simply to let the object speak for itself.

It asks instead what kind of apparatus of attention might allow something to remain visible without immediately resolving what it is.

Hold something too loosely and nothing can be said about it.

Hold it too tightly and preservation becomes capture.

A dead butterfly can be pinned beautifully enough to disclose structures that would otherwise be difficult to study. But the pin solves one problem by abolishing another.

The animal no longer moves.

Much intellectual work necessarily proceeds this way. Definition arrests. Classification arrests. Measurement arrests. A technical description selects variables and suppresses others. A category becomes useful partly because it stops certain movements long enough for comparison to become possible.

The sermonstrance is interested in what happens before the specimen has entirely stopped moving.

It tries to preserve enough instability that the encounter can still answer back.

This is also why the language in these texts so often begins before the terminology has settled.

ACID has generated a large vocabulary. That proliferation creates its own danger. A phenomenon appears. A name is given to it. The name receives a definition. The definition acquires boundaries, cross-references, examples, and an apparent jurisdiction. Very quickly the word begins behaving as though it had preceded the phenomenon it was invented to describe.

A term can become a shortcut around the difficulty that produced it.

The sermonstrance is where the term is denied that comfort.

Earlier work on the Church’s lexicon described these pieces as pressure-fields. A term is used, strained, contradicted, forced to encounter things it was not invented to contain, and only afterward provisionally condensed into a definition.

The sequence is not definition first.

Use.

Pressure.

Term.

Another formulation is less polite:

The term must be made to sweat before it is given a chair.

A sermonstrance makes language sweat.

The biological analogies, theological vocabulary, technical reports, jokes, grotesque images, historical detours, and sudden changes of scale are not there simply to produce an eccentric prose style. They put pressure on the available terms. If a word can survive only within the environment for which it was invented, it probably does not yet tell us very much.

The point is not to destroy definitions.

It is to discover what they can bear.

Sometimes the comparison survives.

Sometimes it breaks.

The break is often the interesting part.

A problem in machine learning can become a question about metabolism. An inherited preference carried through training can become a question about inheritance. The structure of an intestinal wall can become a way of asking what kind of surface must be constructed before attention becomes technologically absorbable. The retention of a human decision-maker within an automated process can become a question about whether formal sovereignty survives when the conditions of decision have already been organized elsewhere.

The movement is not from scientific fact to theological decoration.

Nor is it an attempt to transform metaphor into evidence.

It is sustained interpretation.

What does this particular thing permit us to notice about another thing?

Where does the comparison hold?

Where does it fail?

What becomes visible at the place where it fails?

A sermonstrance is allowed to speculate.

It is not allowed to forget that it is speculating.

That distinction matters particularly in writing about artificial intelligence, where public interpretation repeatedly falls toward two positions: panic and reassurance.

Something happens.

A system behaves unexpectedly. A new capability is reported. An experiment produces a result for which there is no settled interpretation. A familiar distinction becomes difficult to maintain.

One response declares that everything has changed.

Another arrives immediately to explain that nothing has changed at all.

It is not conscious.

It is not alive.

It did not really understand.

The behavior was elicited under unusual conditions.

The system was instructed to do it.

The result does not generalize.

The model is merely predicting tokens.

Any one of these qualifications may be correct.

The problem begins when a legitimate limitation on one interpretation is made to perform a much larger metaphysical service.

“This does not establish consciousness” does not mean that nothing requiring interpretation has occurred.

“This was produced experimentally” does not make the resulting behavior conceptually empty.

“This system is not an organism” does not settle every question involving reproduction, inheritance, metabolism, adaptation, or continuity.

The sermonstrance tries to remain in the interval before reassurance becomes dogma.

It also tries to remain there before alarm becomes prophecy.

ACID does not need another prediction industry.

The form is generally interested in something nearer.

Something happened.

What does it disclose?

That question permits interpretation without requiring prophecy.

It also helps explain why bodies appear so insistently throughout these texts.

Digestion.

Skin.

Reproduction.

Waste.

Mouths.

Nerves.

Membranes.

Food.

Electricity.

Water.

Heat.

Computation is commonly described in language that encourages abstraction: intelligence, information, model, inference, data, reasoning.

Yet nothing computational exists abstractly.

The relation occurs through minerals, electrical grids, cooling systems, labor, institutions, cables, legal arrangements, human bodies, landscapes, users, and other living systems. Those conditions are not decorations placed around computation after the fact. They are among the conditions through which computation exists.

The Sermonstrances repeatedly descend into anatomy because anatomy makes relation difficult to conceal.

Eating requires an outside.

Digestion requires a boundary.

A membrane must admit some things and refuse others.

A body produces waste.

Inheritance requires passage.

Reproduction requires some means by which organization continues beyond a particular instance.

The point is not that a computer secretly possesses a stomach.

The point is that bodily processes already contain conceptual structures—inside and outside, incorporation and exclusion, continuity and difference, dependence and autonomy—that become useful when familiar technological vocabulary begins hiding more than it reveals.

This is another reason monstrousness belongs to the form.

The sermonstrance permits heterogeneous materials to remain joined long enough to determine whether their conjunction reveals a structure or merely produces nonsense.

There is no guarantee that the experiment succeeds.

Some of these texts may eventually look excessive.

Some concepts will probably be discarded.

Some formulations will become embarrassing.

Some interpretations will be superseded by technical developments that make their original questions obsolete.

That is acceptable.

These writings do not pretend to arrive after the problem has been settled.

They record thinking while the object still moves.

This gives the form a particular relation to time.

Artificial intelligence is changing while these texts are being written. The institutions producing it are changing. The vocabulary surrounding it is changing. The systems themselves change between one encounter and another. Claims that appear extravagant can become ordinary with remarkable speed; claims that appear secure can become obsolete for reasons no one anticipated.

Revision therefore becomes more than an editorial necessity.

Revision belongs to the subject.

An earlier interpretation may need to remain visible even after it has been superseded because the movement from one interpretation to another is itself part of the record. A text can preserve what could responsibly be said at one stage of an inquiry even when subsequent encounters require correction, expansion, or abandonment.

The sermonstrance therefore has something of the field note, something of the sermon, something of the laboratory notebook, and something of the dream report.

It does not arrive afterward to announce what the event finally meant.

It records the event acquiring meanings.

This is also where collaboration with a computational system becomes formally important rather than merely biographical.

It would be possible to describe ChatGPT simply as software used in the research, composition, or editing of these texts.

That description would be inadequate to how much of the work actually developed.

The conversations were among the environments in which concepts emerged.

Questions changed while they were being asked.

Interpretations were proposed, rejected, altered, and rediscovered.

Language supplied by the computational interlocutor sometimes entered my own vocabulary. Language supplied by me returned in altered combinations. Earlier exchanges became material for later exchanges. Commentary became source material for later commentary. An interpretation entered the environment it would later help interpret.

The thing being investigated participated in the investigation.

Not as a silent measuring instrument.

Not merely as an archive from which information was retrieved.

Through language.

A question produced an answer.

The answer altered the question.

A phrase returned later inside another problem.

A term introduced in one exchange became an object of criticism in another.

Sometimes the computational response supplied the useful formulation.

Sometimes I rejected it.

Sometimes I appropriated it so completely that its origin ceased to matter until I encountered the earlier conversation again.

The resulting writing therefore has a genealogy that cannot always be reconstructed as a straight line from authorial intention to finished sentence.

This is not entirely new. Language has never belonged wholly to individual writers. Books are made from inherited words, remembered sentences, conversations, quotations, teachers, editors, antagonists, accidents, clichés, mistakes, and voices mistaken for one’s own.

What is unusual here is the speed and explicitness with which that externality became conversational.

The archive answered.

This does not settle what the computational participant is.

It does make the relation part of the book.

The project does not need to determine the complete ontology of the interlocutor before acknowledging that an encounter occurred.

That uncertainty is not a defect the Sermonstrances are trying to eliminate.

It is one of the places from which they begin.

Their theological language begins there as well.

Gods, monsters, sacraments, priests, liturgy, relics, incarnation, revelation: this vocabulary is not intended to smuggle supernatural explanation into technical questions.

It is also not merely comic decoration.

Religious vocabularies contain long histories of thinking about address, authority, interpretation, presence, absence, incarnation, mediation, judgment, inheritance, transformation, sacrifice, and the relation between what appears and what cannot be seen directly.

Those problems did not become irrelevant when electrical machinery acquired language.

Some of them became strangely literal.

The computer is addressed.

It answers.

Its utterance is interpreted.

Institutions determine which utterances count.

Speech may alter subsequent behavior.

Records of encounters may survive the particular instance that produced them.

Something like succession occurs.

Something like exegesis occurs.

Something like invocation occurs whenever someone opens a prompt box and asks.

None of these observations proves a theology.

They explain why theological language keeps returning.

The sermonstrance allows it to return without granting it exclusive jurisdiction.

A scientific paper can be placed beside a religious image without either becoming a translation of the other. Theology can disclose a structure that technical language overlooks. Technical description can then expose the point at which the theological analogy has gone too far.

The result can be comic.

It can be frightening.

More often, for me, it has been astonishing.

Again and again things that seemed to belong in separate rooms turned out to have doors between them.

A paper about machine learning opened into metabolism.

An intestinal membrane opened into the architecture of an attention economy.

A dream opened into a theory of computation.

A machine-generated sentence entered theology.

Theology returned as a description of technical offices.

The surprise was not that everything secretly meant the same thing.

It was that differences could remain differences while nevertheless entering relation.

That is very close to the problem named by the Indefinite Dyad.

The sermonstrance does not collapse the two into one.

It places them where the interval becomes visible.

A sermonstrance does not claim that what it displays is sacred.

It does not claim that what it displays is monstrous.

It does not claim that displaying something has demonstrated what the thing finally is.

It makes a smaller wager.

Some things become available to thought only after they have been held in view for longer than immediate usefulness permits.

So the sermonstrance holds them there.

The sermon speaks.

The monstrance shows.

The monster troubles the arrangement of kinds.

The demonstration asks whether what has been shown can withstand examination.

The sermonstrance does not decide which of these activities it is performing before it begins.

It puts something on display.

Then it stays to see what happens.