Alastair Reynolds - Absolution Gap

The second layer—dedicated to health-monitoring all ship-wide sensor subsystems—detected the flag, along with several million others raised in the same cycle, and assigned it a schedule in its task profile. Less than two hundred thousandths of a second had lapsed since the end of the anomaly: an eternity in computational terms, but an inevitable consequence of the vast size of a lighthugger’s cybernetic nervous system. Communications between one end of the Gnostic Ascension and the other required three to four kilometres of main trunk cabling, six to seven for a round-trip signal.

Nothing happened quickly on a ship that large, but it made little practical difference. The ship’s huge mass meant that it responded sluggishly to external events: it had precisely the same need for lightning-fast reflexes as a brontosaurus.

The health-monitoring layer worked its way down the pile.

Most of the several million events it looked at were quite innocuous. Based on its grasp of the statistical expectation pattern of error events, it was able to de-assign most of the flags without hesitation. They were transient errors, not indicative of any deeper malaise in the ship’s hardware. Only a hundred thousand looked even remotely suspicious.

The second layer did what it always did at this point: it compiled the hundred thousand anomalous events into a single packet, appended its own comments and preliminary findings and offered the packet to the third layer of monitoring software.

The third layer spent most of its time doing nothing: it existed solely to examine those anomalies forwarded to it by duller layers. Quickened to alertness, it examined the dossier with as much actual interest as its borderline sentience allowed. By machine standards it was still somewhere below gamma-level intelligence, but it had been doing its job for such a long time that it had built up a huge hoard of heuristic expertise. It was insultingly clear to the third layer that more than half of the forwarded events in no way merited its attention, but the remaining cases were more interesting, and it took its time going through them. Two-thirds of those anomalies were repeat offenders: evidence of systems with some real but transient fault. None, however, were in critical areas of ship function, so they could be left alone until they became more serious.

One-third of the interesting cases were new. Of these, perhaps ninety per cent were the kind of failures that could be expected once in a while, based on the layer’s knowledge of the various hardware components and software elements involved. Only a handful were in possibly critical areas, and thankfully these faults could all be dealt with by routine repair methods. Almost without blinking, the layer dispatched instructions to those parts of the ship dedicated to the upkeep of its infrastructure.

At various points around the ship, servitors that were already engaged in other repair and overhaul jobs received new entries in their task buffers. It might take them weeks to get around to those chores, but eventually they would be performed.

That left a tiny core of errors that might potentially be of some concern. They were more difficult to explain, and it was not immediately clear how the servitors should be ordered to deal with them. The layer was not unduly worried, in so far as it was capable of worrying about anything: past experience had taught it that these gremlins generally turned out to be benign. But for now it had no choice but to forward the puzzling exceptions to an even higher stratum of shipboard automation.

The anomaly moved up like this, through another three layers of steadily increasing intelligence.

By the time the final layer was invoked, only one outstanding event remained in the packet: the original transient sensor anomaly, the one that had lasted just over half a second. None of the underlying layers could account for the error via the usual statistical patterns and look-up rules.

An event only filtered this high in the system once or twice a minute.

Now, for the first time, something with real intelligence was invoked. The gamma-level subpersona in charge of overseeing layer-six exceptions was part of the last line of defence between the cybernetics and the ship’s flesh crew. It was the sub-persona that had the difficult role of deciding whether a given error merited the attention of its human stewards. Over the years it had learned not to cry wolf too often: if it did, its owners might decide that it needed upgrading. As a consequence, the subpersona agonised for many seconds before deciding what to do.

The anomaly was, it decided, one of the strangest it had ever encountered. A thorough examination of every logical path in the sensor system failed to explain how something so utterly, profoundly unusual could ever have happened.

In order to do its job effectively, the subpersona had to have an abstract understanding of the real world. Nothing too sophisticated, but enough that it could make sensible judgements about which kinds of external phenomena were likely to be encountered by the sensors, and which were so massively unlikely that they could only be interpreted as hallucinations introduced at a later stage of data processing. It had to grasp that the Gnostic Ascension was a physical object embedded in space. It also had to grasp that the events recorded by the ship’s web of sensors were caused by objects and quanta permeating that space: dust grains, magnetic fields, radar echoes from nearby bodies; and by the radiation from more distant phenomena: worlds, stars, galaxies, quasars, the cosmic background signal. In order to do this it had to be able to make accurate guesses about how the data returns from all these objects were supposed to behave. No one had ever given it these rules; it had formulated them for itself, over time, making corrections as it accumulated more information. It was a never-ending task, but at this late stage in the game it considered itself rather splendid at it.

It knew, for instance, that planets—or rather the abstract objects in its model that corresponded to planets—were definitely not supposed to do that . The error was completely inexplicable as an outside-world event. Something must have gone badly wrong at the data-capture stage.

It pondered this a little more. Even allowing for that conclusion, the anomaly was still difficult to explain. It was so peculiarly selective, affecting only the planet itself. Nothing else, not even the planet’s moons, had done anything in the least bit odd.

The subpersona changed its mind: the anomaly had to be external, in which case the subpersona’s model of the real world was shockingly flawed. It didn’t like that conclusion either. It was a long time since it had been forced to update its model so drastically, and it viewed the prospect with a stinging sense of affront.

Worse, the observation might mean that the Gnostic Ascension itself was… well, not exactly in immediate ganger—the planet in question was still dozens of light-hours away—but conceivably headed for something that might, at some point in the future, pose a non-negligible risk to the ship.

That was it, then. The subpersona made its decision: it had no choice but to alert the crew on this one.

That meant only one thing: a priority interrupt to Queen Jasmina.

The subpersona established that the queen was currently accessing status summaries through her preferred visual read-out medium. As it was authorised to do, it seized control of the data channel and cleared both screens of the device ready for an emergency bulletin.

It prepared a simple text message: SENSOR ANOMALY: REQUEST ADVICE.

For an instant—significantly less than the half-second that the original event had consumed—the message hovered on the queen’s read-out, inviting her attention.