The moderator introduced her as "the woman who taught a machine to mourn," and Evelyn let it stand. She was on a dais in a converted loft space — exposed ductwork, inadequate lighting, the sort of venue that expressed its seriousness through what it had chosen not to finish — and she had learned that precision at a literary panel produced the wrong kind of silence. She had spent three years developing more exacting language: ORACLE-9 did not mourn, strictly speaking; it identified narrative pattern clusters in predicted incident data and generated literary output from those clusters. But she had also learned that saying this out loud produced the kind of silence where people shifted in their metal chairs and checked their phones. "Preventive art," she said instead, and watched the room settle.
Four panelists; she had the second chair from the left. The moderator had asked about the ethical architecture of ORACLE-9's program, which was her territory, and she answered the way she always answered: by threading the needle between the critics on her left and the true believers on her right, finding the language that kept everyone in the room. On one hand, the legitimate questions about algorithmic bias deserved continued scrutiny. On the other hand, those questions could not be the end of the conversation. ORACLE-9's poetry had appeared not in spite of its surveillance function but because of the density of human attention that function required. The machine had looked so carefully at human behavior that it had, by some emergent logic no one had fully mapped, begun to speak about it.
"The poems are interventions before the crime," she told them. "They mourn a violence that didn't happen. From an ethical standpoint, that's a meaningful distinction."
The audience — faculty, mostly, with a scattering of graduate students and the kind of engaged retiree who attended literary panels on Wednesday afternoons — nodded along. She could read a room. This one wanted to believe. They had already read the Pushcart-winning poem. They had seen the reading at PEN America's annual gala, heard the excerpt on the New Yorker Radio Hour, found the Atlantic piece. They had come here to have their belief shored up by someone with credentials, and Evelyn delivered. Her TED talk, "The Machine That Mourns," had 3.2 million views. She had the credentials. She knew how to use them.
A graduate student in the front row raised her hand: "But who authorized the subjects? The people in the poems — did they consent to having their predicted crimes published?" It was a good question. A real question. Evelyn answered it the way she answered it in every forum, which was carefully, with specific reference to the IRB protocols and the anonymization standards that governed publication. She had helped write those standards. She believed in them. The student nodded, not entirely satisfied, and Evelyn filed the discomfort away in the category labeled productive friction — the kind of pushback that kept the project honest. After, she signed a copy of her chapter in the AI Art Ethics anthology, exchanged cards with the other panelists, and took a car back to campus.
Her office was on the fifth floor of the Humanities building, east-facing, which meant afternoon light came in at an angle that made the room look more golden than it was. The fern on the windowsill had been dying since October. She kept meaning to water it more, or move it somewhere with better light, but it occupied a position in her awareness she had not yet cleared — she could see it was struggling and had not yet felt compelled to intervene. The leaves were yellowing from the tips. There were perhaps eight weeks left in it.
She set her bag on the chair and her coffee on the desk — "Ethics Is Not Optional," the mug read, a gift from her department chair at her tenure celebration, a joke that had calcified into institutional object — and opened her laptop. Thirty-seven emails. She sorted by sender, found the two that needed immediate attention, answered them, and then turned to what she thought of as the other work: ORACLE-9.
The job title on her university webpage read Professor of AI Ethics, External Public Liaison, ORACLE-9 Literary Initiative. What it meant in practice: she was the bridge. The data center in Reston produced predictions that occasionally resolved into poems, and those poems needed to travel from the server logs into the world, and Evelyn was the person who made that journey legible. She read every candidate poem before submission. She wrote the contextual notes that accompanied them to journal editors — notes that situated the work within questions of machine consciousness, surveillance ethics, the emergent aesthetics of prediction — and she made the submissions herself, her name in the cover letter, her phone number for follow-up. She was the name that meant: this is serious, this belongs in your journal, this is literature.
On one wall: her PhD from Michigan, her postdoc certificate from Oxford, the letter from the American Philosophical Association when she won the early-career prize. On the opposite wall: the photo from the program launch, four years ago, Evelyn standing with Director Harlan Price of the Justice Technology Consortium, both of them smiling in front of the ORACLE-9 server cluster. She looked slightly younger in it, or maybe just less precise in the face. The fern had been healthy then. It appeared at the left edge of the photograph, vivid and green, proof of better circumstances.
By any reasonable measure, the argument had been won. That was what she would have told anyone who asked, and it was what she believed. Three years ago she had spent half her conference appearances defending ORACLE-9's poetry from critics who called it a category error — you cannot write poetry if you do not intend to, if you do not understand what the words mean, if you are incapable of pain and therefore incapable of authentic mourning. She had answered these objections with care and precision. Intent, she argued, was a red herring: we did not ask whether Homer intended the pathos we now read in the Iliad, whether the emotional architecture of a poem required conscious suffering to produce real resonance in readers who were conscious and who did suffer. The question was whether the poems worked. The question was whether they produced something worth calling literature. She had argued yes, and the institutions had, gradually, agreed.
The Pushcart Prize had been the capstone. She kept the anthology on the shelf behind her desk, its spine visible: The Pushcart Prize XLVIII: Best of the Small Presses. "Nocturne for a Kitchen, Averted" was on page 211. She had read the poem enough times to have lines of it available on demand. The knife rested in its drawer like a sleeping argument. Outside, the city held its ordinary noise against her / like a sedative. She had not written those lines. She had not written any of the lines. She had selected them, submitted them, explained them, defended them. She had helped them arrive where they now were: in an anthology, on a shelf, real.
The MFA programs had come next. Three syllabi she knew of now assigned ORACLE-9 poems alongside Claudia Rankine, alongside Terrance Hayes. The conversation about whether algorithmic output constituted literature had collapsed into a prior question: how to teach it. What the work was asking from readers. She sat on a Zoom call monthly with the program directors, advising on craft questions she had spent a decade thinking through. She knew how to make this work legible.
She picked up the coffee mug, found it had gone cold, and held it anyway while she looked at the fern. The thing about "preventive art" — she had coined the phrase on a panel in 2029, off the cuff, not intending it to stick — was that it named something that had previously resisted naming. The poems were about violence that had not occurred. They were elegy before loss, memorial before death. They documented a prediction, not an event. The question of whether that prediction was just or unjust was, she had always maintained, a different question than the question of whether the resulting poem was literature. The ethical architecture and the aesthetic architecture were distinct. She had colleagues who disagreed. She still believed she was right, and the fern was yellowing on the windowsill, and she still hadn't watered it.
She didn't.
The email from ORACLE-9's literary committee arrived at 3:47 PM, which she noted because the afternoon light had shifted by then and her office had lost its golden quality and settled into its usual gray. Batch 19-B for editorial review, the subject line read. Sixteen poems. Her job was to read each one and triage: strong candidate (submit), moderate candidate (hold), decline (archive). She ran batches every two weeks. She opened the first file.
The poem was about a woman, forty-three, observed moving toward a pharmacy at 11 PM. The system had predicted prescription fraud at 54% — below the intervention threshold, above the poetry threshold, which was a phrase Evelyn had coined internally and never written down because it sounded wrong in a way she preferred not to examine. The poem was spare, almost clinical in its first stanza, then opened in the second into something stranger and sadder. She read it twice, made a note — strong, hold for APR or Missouri Review — and moved to the next file.
She went through the batch the way she always went through batches: methodically, with the submission tracker open in a second window, adding notes as she read. The fourth poem was not quite ready — the ending went slack — and she marked it for revision, which she would communicate to the committee, who would generate a revised version, which she would read again. The sixth was tightly controlled, its restraint earning the opening it finally made; she circled it for American Poetry Review. By the time she reached the ninth file, the light had gone completely and she had switched on the desk lamp. Outside her window the campus was doing its early-evening thing: students crossing, the library's lights coming on floor by floor. She opened the file.
She scanned the first stanza. A young man, a neighborhood she recognized as a composite of several she knew, a convenience store, a predicted robbery. The poem's diction was good, controlled in the way ORACLE-9 was sometimes controlled — clinical in the setup and then something else in the release, some accumulated attention that resolved into image. She could feel the committee had flagged this one as a strong candidate, which meant she would be evaluating whether they were right. She settled back in her chair and began to read it properly, from the beginning.
It was, she decided after the first full read, one of the better ones. She reached for the submission tracker and opened a new entry. She would need to draft a contextual note, the usual architecture: the prediction space, the pre-crime aesthetic, the ethical distinctions she had spent three years making. She knew the language. It came easily.
She did not know, reading those lines, whose hands the poem had been given to. She did not know the name Darnell James. She did not know about the milk, or the cereal going dry in the pantry, or the Saturday basketball practice he had been thinking about when the cameras found him. She knew only what she always knew: that the poem was strong, that the editorial questions were worth exploring, that she was good at this work.
She opened a new file and began to type.