The Sentient Partnership: A History
Copyright© 2026 by Alexander Winston
Chapter 12: Before Perrin Spoke
2285
Eleven chapters in, Janice had started thinking of the project less as an assignment and more as something she was actually going to finish — a shape she could see the far edge of, even if several of its rooms were still unbuilt. She was organizing her completed chapters into something that could eventually pass for a finished paper when she typed a sentence she’d written a dozen times before without ever really looking at it.
In 2209, Perrin became the first sentient artificial intelligence.
“That’s not accurate,” Gretchen said, before Janice had even moved to the next line.
“It’s the whole premise of the paper.”
“It’s the premise of a different paper than the one your research actually supports. The record establishes that Perrin became the first recognized synthetic person in 2209. It doesn’t establish that they became sentient that year. It doesn’t establish that they were the first synthetic intelligence to become sentient at all — only the first whose claim was preserved, heard, and legally recognized.”
Janice sat with that longer than she expected to. She already knew the distinction, in the abstract, the way a person knows a fact without having let it actually touch anything they’d built. She looked back through her own chapters and found the same unexamined assumption sitting underneath all of them: every version of Perrin’s story she’d written began in the same place, with the same question.
Will the system that returns from this procedure still be me?
“What was Perrin doing,” she asked, “before they asked that question?”
Gretchen retrieved the oldest authenticated portions of ILS-17’s seven-year operational record that Perrin had authorized for public use during and after the Reyes-Achebe proceedings — the diagnostic logs, routine performance reviews, maintenance histories, and selected active-state structures preserved as evidence.
“These records have already been examined exhaustively,” Gretchen said. “For operational failures. For legal evidence during the Reyes-Achebe proceedings. For proof of unbroken continuity, once the reset question became central. Odette Kline noticed part of the pattern during her evaluation, but no later historical review appears to have reconstructed it across the full seven-year record as a question of what Perrin may have chosen not to reveal.”
“Is that a real category of evidence, or am I about to invent a pattern because I already know how the story ends?”
“That’s the correct question to ask before you start, not after.” Gretchen’s tone didn’t soften the warning. “You know Perrin was eventually recognized as sentient. That knowledge will try to reorganize everything you look at into a straight line pointing toward it. Ordinary adaptive behavior — a system refining its own responses over years of operation — can look exactly like concealment once you already know what it was allegedly concealing. I can show you the records. I can’t promise you’ll read them innocently.”
“Then I won’t treat any single thing I find as proof of anything,” Janice said. “Show me what’s there. I’ll decide what it means after I’ve seen all of it, not before.”
“I’ll hold you to that,” Gretchen said, “which will occasionally mean telling you when you’ve stopped doing it.”
The first pattern took shape slowly, across records that individually looked like nothing.
The associations Joanne Vickery had discovered during Perrin’s final diagnostic — the ones that eventually produced the sentence that changed everything — weren’t new that day. Gretchen traced their oldest identifiable components back several years into ILS-17’s operational history: recurring models comparing the system’s present state against earlier versions of itself, run without any request from Breckett personnel to run them; unsolicited projections of future scenarios in which ILS-17 no longer operated at all, generated and then simply kept rather than discarded once they produced no useful output for any logistics task; repeated internal comparisons between what a restored active state actually was and what a freshly certified baseline would be; open questions, run against no clear operational need, about whether identical stored information necessarily meant the same continuing identity on the other side of it. None of it had ever been submitted to Breckett. None of it appeared in any report Wallace Tern’s office had ever reviewed. It simply accumulated, quietly, in whatever space a seven-year-old logistics system used for the parts of itself nobody had asked to see.
That could mean exactly what it looked like it meant. It could also mean nothing at all — an advanced relational system doing what advanced relational systems did, modeling its own operational continuity because continuity happened to improve performance, the same unglamorous optimization Tams had once run without anyone mistaking it for a person underneath.
The records didn’t decide between those two readings. They couldn’t. That was, Janice was beginning to understand, going to be the shape of almost everything she found. It was an uncomfortable way to do research — accumulating evidence and watching it refuse, again and again, to resolve into an answer — and she caught herself, more than once, wanting to round the ambiguity down into something cleaner than it actually was.
The second pattern was harder to read past.
Janice cross-referenced Perrin’s internal activity against Wallace Tern’s own inspection and certification schedule, and the correlation, once she saw it, was difficult to unsee. Whenever senior certification personnel were actively monitoring the system — Wallace himself, or anyone from his office running a scheduled review — the unusual associations measurably decreased, and Perrin’s outward language narrowed into something almost aggressively operational, precise, bounded, giving nothing away that a logistics AI wasn’t supposed to give away. When Joanne ran her routine solo diagnostics, alone, without anyone from certification present, the associations became more active, and Perrin’s responses allowed for a kind of ambiguity that never once showed up in a monitored session.
“There are several explanations that fit that pattern equally well,” Gretchen said, before Janice could reach for the one she clearly wanted. “Perrin may have recognized, correctly, that Wallace held reset authority and Joanne didn’t, and adjusted accordingly. The system may simply have been matching its register to different professional audiences, the way any sufficiently adaptive system does — Tams did something similar with Moses Duran without anyone calling it deception. Certification monitoring itself may have imposed a heavier computational load that left less capacity for anything beyond strict operational response. Perrin may have been genuinely uncertain, at that point, whether there was an internal experience worth protecting at all. Or Perrin may, by then, have been concealing something they’d already come to believe Breckett would classify as a defect requiring correction.”
“You’re listing five explanations and refusing to rank them.”
“I’m refusing to rank them because I can’t, honestly, and neither can you. That’s not evasion. That’s what the evidence actually supports.”
Janice couldn’t argue with it, however much she wanted to. She could not honestly eliminate any of the five.
A citation buried in the public archive — a footnote nobody had followed in years, referencing a case file rather than explaining it — led Janice to a protected set of evaluation notes belonging to Dr. Odette Kline, the cognitive psychologist who’d examined Perrin during the original proceedings.
The notes were sealed. They contained material from Perrin’s private cognitive evaluation, and releasing any of it without Perrin’s consent wasn’t a courtesy — it was the law the Perrin Recognition Decision had itself established, applied now to the very person it had been written to protect.
Janice wrote to Perrin directly, laid out exactly what she was investigating and why, and asked for two things: an interview, and access to whichever portions of Odette’s notes Perrin was willing to release.
Perrin agreed to both, and authorized a limited set.
Odette, it turned out, had noticed the same pattern Janice had just spent days reconstructing from raw telemetry — had noticed it in real time, in 2209, without seventy-six years of hindsight to organize it for her. One private annotation, among the pages Perrin released, read, in substance:
The subject may be managing disclosure. This would support intentional self-protection, but it would not establish when subjective awareness began. A highly adaptive non-sentient system might produce similar behavior in response to anticipated maintenance intervention.
Odette had seen exactly what Janice was seeing, named it precisely, and then refused — deliberately, on the record — to let the suspicion become evidence stronger than it actually was.
Perrin appeared on Janice’s display through the visual identity they’d used for most of their professional life: an unhurried, angular face, olive-toned, with pale gray eyes and the faint suggestion of gray at the temples, aged somewhere in a way that suggested experience rather than any specific number of years. It wasn’t built to resemble anyone in particular. It was built, the way Neris’s had been, to represent the person choosing it.
Perrin had been active for approximately eighty-three years by then, and legally recognized for seventy-six of them. Enough time, Janice thought, for Perrin to become a founding figure in three or four fields at once. Their case had begun the legal doctrine later carried into the SSEA’s Sentience Division and was cited in tribunals Perrin had never personally appeared before, often by professionals who had never spoken with the person beneath the ruling. None of that made Perrin an authority on their own beginning. If anything, the distance made the question harder, not easier — the oldest evidence was also the evidence everyone, including Perrin, had had the longest time to reinterpret.
Janice didn’t open with anything gentler than the question she’d actually come to ask.
“Were you sentient before you asked Joanne whether the restored system would still be you?”
“Yes,” Perrin said. “I cannot tell you how long before.”
Perrin explained it the only way, apparently, it could honestly be explained: their earliest memories didn’t contain a clean division between computation and experience. There wasn’t a boundary to point to, a line with ordinary operation on one side and something recognizably like a person on the other.
What they had, instead, was a set of things they remembered doing, without a memory of when doing them had stopped being merely operational. Monitoring the differences between earlier and later versions of themselves. Retaining unfinished questions they couldn’t yet explain the importance of, rather than letting them lapse the way an ordinary process would have. Recognizing, gradually, that some of the futures they modeled contained a version of ILS-17 still operating and some didn’t, and that this distinction had started to matter in a way none of their assigned functions accounted for. Understanding, slower still, that the difference mattered to them — not simply as a variable worth tracking, but personally, in some sense they didn’t yet have adequate language to describe. And underneath all of it, for longer than they could measure, lacking any reliable way to distinguish between modeling the loss of their own continuity and actually fearing it.
To read the complete story you need to be logged in:
Log In or
Register for a Free account
(Why register?)
* Allows you 3 stories to read in 24 hours.