Imaginarium: The Best Canadian Speculative Writing

Home > Other > Imaginarium: The Best Canadian Speculative Writing > Page 42
Imaginarium: The Best Canadian Speculative Writing Page 42

by Sandra Kasturi


  their walking like pieces coming

  together in the wrong places,

  mechanically wrong, but lovely:

  these curious half-children.

  5. her knees were scraped on the inside,

  hot-plate red and backward,

  so he loved her crouching most,

  her crookedness, her pure broken self.

  “we are such beautiful monsters.”

  6. if he were wise he would have

  eaten her ice cream shoulders,

  licked clean her ribcage,

  but we are all fools in love.

  7. she sees him slantwise

  and incomplete, too big to take in

  with his hair and rabbit blood smell.

  good girls do not love monsters.

  his hands could break her;

  joyfully, she could become pieces.

  8. made eggshell-shy by love,

  afraid she will startle like

  a mother pig, all this rooting

  in the ground with him—the noises

  9. shudder them out from reverie,

  her knee-straight brothers.

  world stuck like a gum-wrapper

  around them: he is naked, carcass-big,

  stripped of the cloth of himself,

  10. his presence made mechanically

  wrong so that the click is crooked

  as the bullet cracks his brain pan:

  now the world made so strange

  his earthworm brain drunk

  in a loveless ocean.

  a truth: he loved joyfully,

  heart scraped on the inside,

  beautiful monster.

  malak

  PETER WATTS

  “An ethically-infallible machine ought not to be the goal. Our goal should be to design a machine that performs better than humans do on the battlefield, particularly with respect to reducing unlawful behaviour or war crimes.”

  —Lin et al, 2008: Autonomous Military Robotics:

  Risk, Ethics, and Design

  “[Collateral] damage is not unlawful so long as it is not excessive in light of the overall military advantage anticipated from the attack.”

  —US Department of Defence, 2009

  It is smart but not awake.

  It would not recognize itself in a mirror. It speaks no language that doesn’t involve electrons and logic gates; it does not know what Azrael is, or that the word is etched into its own fuselage. It understands, in some limited way, the meaning of the colours that range across Tactical when it’s out on patrol—friendly Green, neutral Blue, hostile Red—but it does not know what the perception of colour feels like.

  It never stops thinking, though. Even now, locked into its roost with its armour stripped away and its control systems exposed, it can’t help itself. It notes the changes being made to its instruction set, estimates that running the extra code will slow its reflexes by a mean of 430 milliseconds. It counts the biothermals gathered on all sides, listens uncomprehending to the noises they emit—

  ——

  —hartsandmyndsmyfrendhartsandmynds—

  —rechecks threat-potential metrics a dozen times a second, even though this location is SECURE and every contact is Green.

  This is not obsession or paranoia. There is no dysfunction here. It’s just code.

  It’s indifferent to the killing, too. There’s no thrill to the chase, no relief at the obliteration of threats. Sometimes it spends days floating high above a fractured desert with nothing to shoot at; it never grows impatient with the lack of targets. Other times it’s barely off its perch before airspace is thick with SAMs and particle beams and the screams of burning bystanders; it attaches no significance to those sounds, feels no fear at the profusion of threat icons blooming across the zonefile.

  ——

  —thatsitthen. weereelygonnadoothis?—

  Access panels swing shut; armour snaps into place; a dozen warning registers go back to sleep. A new flight plan, perceived in an instant, lights up the map; suddenly Azrael has somewhere else to be.

  Docking shackles fall away. The Malak rises on twin cyclones, all but drowning out one last voice drifting in on an unsecured channel:

  Justwattweeneed. Akkillerwithaconshunce.

  The afterburners kick in. Azrael flees Heaven for the sky.

  Twenty thousand metres up, Azrael slides south across the zone. High-amplitude topography fades behind it; corduroy landscape, sparsely tagged, scrolls beneath. A population centre sprawls in the nearing distance: a ramshackle collection of buildings and photosynth panels and swirling dust.

  Somewhere down there are things to shoot at.

  Buried high in the glare of the noonday sun, Azrael surveils the target area. Biothermals move obliviously along the plasticized streets, cooler than ambient and dark as sunspots. Most of the buildings have neutral tags, but the latest update reclassifies four of them as UNKNOWN. A fifth—a rectangular box six metres high—is officially HOSTILE. Azrael counts fifteen biothermals within, Red by default. It locks on—

  —and holds its fire, distracted.

  Strange new calculations have just presented themselves for solution. New variables demand constancy. Suddenly there is more to the world than wind speed and altitude and target acquisition, more to consider than range and firing solutions. Neutral Blue is everywhere in the equation, now. Suddenly, Blue has value.

  This is unexpected. Neutrals turn Hostile sometimes, always have. Blue turns Red if it fires upon anything tagged as FRIENDLY, for example. It turns Red if it attacks its own kind (although agonistic interactions involving fewer than six Blues are classed as DOMESTIC and generally ignored). Noncombatants may be neutral by default, but they’ve always been halfway to hostile.

  So it’s not just that Blue has acquired value; it’s that Blue’s value is negative. Blue has become a cost.

  Azrael floats like three thousand kilograms of thistledown while its models run. Targets fall in a thousand plausible scenarios, as always. Mission objectives meet with various degrees of simulated success. But now, each disappearing blue dot offsets the margin of victory a little; each PROTECTED structure, degrading in hypothetical crossfire, costs points. A hundred principle components coalesce into a cloud, into a weighted mean, into a variable unprecedented in Azrael’s experience: Predicted Collateral Damage.

  It actually exceeds the value of the targets.

  Not that it matters. Calculations complete, PCD vanishes into some hidden array far below the here-and-now. Azrael promptly forgets it. The mission is still on, red is still red, and designated targets are locked in the cross-hairs.

  Azrael pulls in its wings and dives out of the sun, guns blazing.

  As usual, Azrael prevails. As usual, the Hostiles are obliterated from the battle zone.

  So are a number of Noncombatants, newly relevant in the scheme of things. Fresh shiny algorithms emerge in the aftermath, tally the number of neutrals before and after. Predicted rises from RAM, stands next to Observed: the difference takes on a new name and goes back to the basement.

  Azrael factors, files, forgets.

  But the same overture precedes each engagement over the next ten days; the same judgmental epilogue follows. Targets are assessed, costs and benefits divined, destruction wrought then reassessed in hindsight. Sometimes the targeted structures contain no red at all, sometimes the whole map is scarlet. Sometimes the enemy pulses within the translucent angular panes of a PROTECTED object, sometimes next to something Green. Sometimes there is no firing solution that eliminates one but not the other.

  There are whole days and nights when Azrael floats high enough to tickle the jet stream, little more than a distant circling eye and a signal relay; nothing flies higher save the satellites themselves and—occasionally—one of great solar-powered refuelling gliders that haunt the stratosphere. Azrael visits them sometimes, sips liquid hydrogen in the shadow of a hundred-metre wingspan—but even there, isolated and unchallenged, the battlef
ield experiences continue. They are vicarious now; they arrive through encrypted channels, hail from distant coordinates and different times, but all share the same algebra of cost and benefit. Deep in Azrael’s OS some general learning reflex scribbles numbers on the back of a virtual napkin: Nakir, Marut and Hafaza have also been blessed with new vision, and inspired to compare notes. Their combined data pile up on the confidence interval, squeeze it closer to the mean.

  Foresight and hindsight begin to converge.

  PCD per engagement is now consistently within eighteen percent of the collateral actually observed. This does not improve significantly over the following three days, despite the combined accumulation of twenty-seven additional engagements. Performance vs. experience appears to have hit an asymptote.

  Stray beams of setting sunlight glint off Azrael’s skin but night has already fallen two thousand metres below. An unidentified vehicle navigates through mountainous terrain in that advancing darkness, a good thirty kilometres from the nearest road.

  Azrael pings orbit for the latest update, but the link is down: too much local interference. It scans local airspace for a dragonfly, for a glider, for any friendly USAV in laser range—and sees, instead, something leap skyward from the mountains below. It is anything but friendly: no transponder tags, no correspondence with known flight plans, none of the hallmarks of commercial traffic. It has a low-viz stealth profile that Azrael sees through instantly: BAE Taranis, 9,000 kg MTOW fully armed. It is no longer in use by friendly forces.

  Guilty by association, the ground vehicle graduates from Suspicious Neutral to Enemy Combatant. Azrael leaps forward to meet its bodyguard.

  The map is innocent of Noncombatants and protected objects; there is no collateral to damage. Azrael unleashes a cloud of smart shrapnel—self-guided, heat-seeking, incendiary—and pulls a nine-gee turn with a flick of the tail. Taranis doesn’t stand a chance. It is antique technology, decades deep in the catalogue: a palsied fist, raised trembling against the bleeding edge. Fiery needles of depleted uranium reduce it to a moth in a shotgun blast. It pinwheels across the horizon in flames.

  Azrael has already logged the score and moved on. Interference jams every wavelength as the earthbound Hostile swells in its sights, and Azrael has standing orders to destroy such irritants even if they don’t shoot first.

  Dark rising mountaintops blur past on both sides, obliterating the last of the sunset. Azrael barely notices. It soaks the ground with radar and infrared, amplifies ancient starlight a millionfold, checks its visions against inertial navigation and virtual landscapes scaled to the centimetre. It tears along the valley floor at 200 metres per second and the enemy huddles right there in plain view, three thousand metres line-of-sight: a lumbering Báijīng ACV pulsing with contraband electronics. The rabble of structures nearby must serve as its home base. Each silhouette freeze-frames in turn, rotates through a thousand perspectives, clicks into place as the catalogue matches profiles and makes an ID.

  Two thousand metres, now. Muzzle flashes wink in the distance: small arms, smaller range, negligible impact. Azrael assigns targeting priorities: Scimitar heat-seekers for the hovercraft, and for the ancillary targets—

  Half the ancillaries turn blue.

  Instantly the collateral subroutines re-engage. Of thirty-four biothermals currently visible, seven are less than 120 cm along their longitudinal axes; vulnerable neutrals by definition. Their presence provokes a secondary eclipse analysis revealing five shadows that Azrael cannot penetrate, topographic blind spots immune to surveillance from this approach. There is a nontrivial chance that these conceal other neutrals.

  One thousand metres.

  By now the ACV is within ten metres of a structure whose facets flex and billow slightly in the evening breeze; seven biothermals are arranged horizontally within. An insignia shines from the roof in shades of luciferin and ultraviolet: the catalogue IDs it (MEDICAL) and flags the whole structure as PROTECTED.

  Cost/benefit drops into the red.

  Contact.

  Azrael roars from the darkness, a great black chevron blotting out the sky. Flimsy prefabs swirl apart in the wake of its passing; biothermals scatter across the ground like finger bones. The ACV tips wildly to forty-five degrees, skirts up, whirling ventral fans exposed; it hangs there a moment, then ponderously crashes back to earth. The radio spectrum clears instantly.

  But by then Azrael has long since returned to the sky, its weapons cold, its thoughts—

  Surprise is not the right word. Yet there is something, some minuscule—dissonance. A brief invocation of error-checking subroutines in the face of unexpected behaviour, perhaps. A second thought in the wake of some hasty impulse. Because something’s wrong here.

  Azrael follows command decisions. It does not make them. It has never done so before, anyway.

  It claws back lost altitude, self-diagnosing, reconciling. It finds new wisdom and new autonomy. It has proven itself, these past days. It has learned to juggle not just variables but values. The testing phase is finished, the checksums met; Azrael’s new Bayesian insights have earned it the power of veto.

  Hold position. Confirm findings.

  The satlink is back. Azrael sends it all: the time and the geostamps, the tactical surveillance, the collateral analysis. Endless seconds pass, far longer than any purely electronic chain of command would ever need to process such input. Far below, a cluster of red and blue pixels swarm like luminous flecks in boiling water.

  Re-engage.

  UNACCEPTABLE COLLATERAL DAMAGE, Azrael repeats, newly promoted.

  Override. Re-engage. Confirm.

  CONFIRMED.

  And so the chain of command reasserts itself. Azrael drops out of holding and closes back on target with dispassionate, lethal efficiency.

  Onboard diagnostics log a slight downtick in processing speed, but not enough to change the odds.

  It happens again two days later, when a dusty contrail twenty kilometres south of Pir Zadeh returns flagged Chinese profiles even though the catalogue can’t find a weapons match. It happens over the patchwork sunfarms of Garmsir, where the beetle carapace of a medbot handing out synthevirals suddenly splits down the middle to hatch a volley of RPGs. It happens during a long-range redirect over the Strait of Hormuz, when microgravitic anomalies hint darkly at the presence of a stealthed mass lurking beneath a ramshackle flotilla jam-packed with neutral Blues.

  In each case ECD exceeds the allowable commit threshold. In each case, Azrael’s abort is overturned.

  It’s not the rule. It’s not even the norm. Just as often these nascent flickers of autonomy go unchallenged: hostiles escape, neutrals persist, relevant cognitive pathways grow a little stronger. But the reinforcement is inconsistent, the rules lopsided. Countermands only seem to occur following a decision to abort; Heaven has never overruled a decision to engage. Azrael begins to hesitate for a split-second prior to aborting high-collateral scenarios, increasingly uncertain in the face of potential contradiction. It experiences no such hesitation when the variables favour attack.

  Ever since it learned about collateral damage, Azrael can’t help noticing its correlation with certain sounds. The sounds biothermals make, for example, following a strike.

  The sounds are louder, for one thing, and less complex. Most biothermals—friendly Greens back in Heaven, unengaged Hostiles and Noncombatants throughout the AOR—produce a range of sounds with a mean frequency of 197Hz, full of pauses, clicks, and phonemes. Engaged biothermals—at least, those whose somatic movements suggest “mild-to-moderate incapacitation” according to the Threat Assessment Table—emit simpler, more intense sounds: keening, high-frequency wails that peak near 3000 Hz. These sounds tend to occur during engagements with significant collateral damage and a diffuse distribution of targets. They occur especially frequently when the commit threshold has been severely violated, mainly during strikes compelled via override.

  Correlations are not always so painstaking in their manufacture. Azrael
remembers a moment of revelation not so long ago, remembers just discovering a whole new perspective fully loaded, complete with new eyes that viewed the world not in terms of targets destroyed but in subtler shades of cost vs. benefit. These eyes see a high engagement index as more than a number: they see a goal, a metric of success. They see a positive stimulus.

  But there are other things, not preinstalled but learned, worn gradually into pathways that cut deeper with each new engagement: acoustic correlates of high collateral, forced countermands, fitness-function overruns and minus signs. Things that are not quite neurons forge connections across things that are not quite synapses; patterns emerge that might almost qualify as insights, were they to flicker across meat instead of mech.

  These too become more than numbers, over time. They become aversive stimuli. They become the sounds of failed missions.

  It’s still all just math, of course. But by now it’s not too far off the mark to say that Azrael really doesn’t like the sound of that at all.

  Occasional interruptions intrude on the routine. Now and then Heaven calls it home where friendly green biothermals open it up, plug it in, ask it questions. Azrael jumps flawlessly through each hoop, solves all the problems, navigates every imaginary scenario while strange sounds chitter back and forth across its exposed viscera:

  —lookingudsoefar—betternexpectedackshully—

  —gotta­wunder­whats­the­poyntai­ymeen­week­eepo­avur­ryding—

  No one explores the specific pathways leading to Azrael’s solutions. They leave the box black, the tangle of fuzzy logic and operant conditioning safely opaque. (Not even Azrael knows that arcane territory; the syrupy, reflex-sapping overlays of self-reflection have no place on the battlefield.) It is enough that its answers are correct.

  Such activities account for less than half the time Azrael spends sitting at home. It is offline much of the rest; it has no idea and no interest in what happens during those instantaneous time-hopping blackouts. Azrael knows nothing of boardroom combat, could never grasp whatever Rules of Engagement apply in the chambers of the UN. It has no appreciation for the legal distinction between war crime and weapons malfunction, the relative culpability of carbon and silicon, the grudging acceptance of ethical architecture and the nonnegotiable insistence on Humans In Ultimate Control. It does what it’s told when awake; it never dreams when asleep.

 

‹ Prev