“Sub2, can we focus please”, Axon sighed, spying Sub2’s video feed in his begoggled periphery.
Sub2 was inspecting the underside of a snailfish again, holding it adrift like some damp, flapping tissue. Sub2 seemed to have developed a thing for this particular fauna. Sub-fleet A were here for the benthic life, but FishDrive’s Seafloor-State Drive (SfSD) principle 1 prohibited handling of this sort.
Readjusting his goggles, Axon shifted slightly, before reclining his cradle one click back. “Dive unit, switch to Sub2 video feed, refresh the environment, and then make a note to book in Sub2 weights for mCBT, that’s a second violation on principle 1.”
Much to the annoyance of the memory stewards, submersible neural network weights are not developed in-house. FishDrive preferred to keep costs down by repurposing generalised intelligence. Given FishDrive’s close links to the ultra-rich, Axon supposed Sub2’s weights had previously been under the employ of some highly particular hotelier, with brute-force compute deployed on napkin inspection. Trivial over-reinforcement tended to produce machine compulsion, making the repurpose of intelligence difficult.
“Switching to Sub2 video feed and refreshing the environment. mCBT for Sub2 confirmed for reinforcement learning on principle 1. Sub2 to recall that biomass is handled only as far as memory is concerned”, Dive unit responded. And then a short pause, before Dive unit emitted a series of loud rapid-fire clicks. From the right, Sub2’s soundscape video feed swelled to fill the entirety of Axon’s field of view.
A bright blinking sepia tone of shape and contour, Axon’s goggle feed was now receiving a visual interpretation of Sub2’s auditory navigation system. Appearing as a detailed sketch-like animation, Axon saw reaching out ahead of him a dexterous, robotic arm, clasping a limply flapping fish by its tail.
With a delicate flourish, Sub2 gently repositioned the snailfish on a rock, and then appeared to freeze momentarily. Axon felt the reset as Sub2 processed Dive unit’s tokenised audio. FishDrive memory stewards were primarily selected on the basis of machine empathy and adaptability to soundscape vision.
To his left, Axon’s in-goggle feed displayed a calendar invite for “Sub2 machine cognitive behavioural therapy”, scheduled to follow a debrief next Tuesday. Axon blinked once to accept and dismiss the message—nothing. 10,000m below, on board a fusion-reactor dive unit, communicating with a fleet of autonomous submersibles via the intellect of a 15 trillion parameter neural network, and Microsoft Teams is apparently the technology most likely to malfunction.
“Dive unit, accept and dismiss mCBT invite.”
“Sub2 mCBT accepted for Tuesday 18th April. Message dismissed,” Dive unit confirmed.
The calendar invite evaporated.
On video, a vast expanse of the Mariana Trench came into view, as Sub2 raised its front up from the seafloor. Sub2 had climbed a slight ridge, up and away from the main fleet. Looking back and down, like some café latte high-res stencil, Sub2 was perceiving 9 fellow autonomous submersibles, sprinkling a fine cocoa dusting onto the milky-froth seafloor below.
Creamy bubbles of life swam amongst rock and machine, feasting as they burst in and out of view.
Invertebrate and vertebrate oddities unknowingly write memories they will never recall.
Axon’s 51st SfSD upload had begun; this was his deepest one yet.
Kicking back in his cradle, Axon pulled his goggles up onto his forehead. Soundscape view is bright, Dive unit is not. Like some gurgling, twinkly deprivation tank, Axon was surrounded on all sides by a thick spherical bulwark, bespangled with buttons that flashed intermittently. He’d never needed them, but it was comforting to know manual override was there.
“Dive unit, soundscape has got me thinking about coffee again. Make me a cup, and then let’s check progress on memory insertion”, Axon said.
“Initiating black coffee brew from the unit cradle. Where would you like me to start on progress?” Dive unit could now reason about the ambiguity of a prompt. Much improved from an earlier model, which might have proceeded to brew Axon a pumpkin spiced latte, and then attempted to retrieve all real-time data from 35 sub fleets currently under operation by FishDrive.
“Ok, let’s start with retroviral offload progress on all 10 subs, and then give me a read on eDNA at 1 meter above seafloor”, Axon responded.
“Retrieving data on genetic material release from all 10 subs of fleet A. Instructing subs 1 through 5 to extend probes and sample at 1 meter.” Protected by 3 inches of steel and an additional internal acoustic shield, Axon heard again the rapid-fire clicking as Dive unit relayed instructions to the sub fleet, but heard nothing of the reply, as all 10 subs produced their own clicks in response, relaying information on their respective material back to Dive unit.
A brief silence.
And then on his right, a panel in the arm of his unit cradle opened up, revealing a dark, steaming mug, which rose up to meet him like some foreboding Sith Lord. Axon greeted the mug with his hand and took a sip.
“Sub fleet data retrieved, relaying visualisations to your goggle feed,” Dive unit spoke.
Axon reached up to the top of his head, pulled the goggles down, and then nestled the ocular sealant into position. Briefly a view of the Sub2 video feed, and then from the right, espresso gave way to a set of equally caffeinated visualisations, as Dive unit relayed continual acoustic updates from the sub fleet.
To his left, 10 panels interpreting the quantity of retroviral genetic material dispensed.
To his right, 10 equivalent panels describing the remaining animal feed, a highly proteinaceous but otherwise obfuscated sludge, mixed with the retroviral content at distribution.
“Dive unit, give me a zoom in on retroviral dispense,” Axon instructed, before taking another slurp from Darth coffee.
Axon had little concern for animal feed.
The proteinaceous sludge of panel right theoretically helped support the integrity of the local deep-sea ecosystem, but otherwise only encouraged retroviral uptake. For sure that wasn’t what FishDrive told the UN. For permission to write to SfSD, the official line from FishDrive was that where non-invasive genetic engineering could be used to fund deep-sea conservation, private companies should be granted permission to operate autonomous gene insertion at the seafloor.
If the UN had seen FishDrive’s machine principles, they would know that conservation was actually superseded by 4 other principles, all kept private as a commercial product. And like many large private entities, FishDrive’s principles are now exclusively written to SfSD, meaning any data retrieval required the deployment of autonomous eDNA sampling at the seafloor.
“Zooming in on retroviral dispense,” Dive unit responded.
“Ok, let’s see what we got,” Axon muttered.
All 10 panels to his left grew across his field of view, resizing upon reaching their terminus to his right. With the visualisation resized, on each panel Axon now had a clear view of the viral particles dispensed for each of subs 1 through 10, both as a percentage and absolute number. He could also see the length of the RNA sequence of the current retroviral upload for each sub, as well as the length of the non-coding region for deep-sea insertion.
Inset on each panel was an image of the ludicrously wealthy individual to whom that data belonged, as well as their unique client ID. Whatever it was they wanted written to SfSD down here, they wanted it still there 500 years from now, and they wanted a 10,000 meter auditable depth guarantee. With the exception of Sub2 and its brief culinary distractions, all subs appeared to be progressing on schedule.
“Dive unit, all seems ok here, switch me over to the seafloor eDNA read-out.”
“Switching over to eDNA samples from subs 1 through 5,” Dive unit confirmed.
A slight delay, and then a set of far more sedate panels sidled in from the right. Each a rendering of individual samples taken from 5 equally spaced locations at the seafloor, giving a rough approximation of memory written to SfSD across this whole sector. All samples showed negligible memory corruption. Sub1 was giving a reading suggesting client AB1657 had experienced a 3% loss, but this region appeared to be represented at all of subs 2, 3, and 4.
In sum, swimmingly.
Axon knew enough about machine intelligence and communication at depth, but the actual genetics? That was for surface nerds. But Axon at least knew this much.
DNA evolved for the storage of information. Life figured that sequences of nucleic acids can be used to encode chains of amino acids, and that those chains can be used to fold bulky-blobs called proteins. Life liked to use those proteins to build itself, again and again for millions of years. Life also learned that you can mess up a bunch of nucleic acids or duplicate them, and sometimes it didn’t matter, or it turned out to be useful, and you ended up with an especially long finger that you could use to tap repeatedly on a tree.
FishDrive were interested in the former—the non-useful gene stuff. Repurposing a programme called AlphaFold, FishDrive’s parent company figured a way to predict gene sequences that would either highly unlikely affect the folding of a protein, or sit within regions of the genome that encoded for nothing. And importantly, highly conserved regions that would mutate very slowly across multiple generations. Seeing an opportunity, and with memory a sought resource, parent company founded FishDrive to facilitate a form of high-end data back-up to regions of non-coding DNA in deep-sea fish.
FishDrive thought deep-sea fish were great. Number 1, at the seafloor you can’t easily communicate with the surface via electromagnetic radiation. Number 2, even if you could, it’s a fish, so it’s not going to run your malicious code, and it doesn’t have somewhere you can just whack in “password123!” to get instant omnipotent access. And number 3, the environment of a deep-sea fish is highly sheltered and stable, meaning if you can insert genes across a whole population, it’s less likely climatic or environmental change will drive those novel genes extinct.
But how to get the data actually in the fish? FishDrive found they could encrypt digital data in RNA sequences of retroviruses, and then through infecting deep-sea marine biota with those retroviruses, transfer that material over to the non-coding DNA of life at the seafloor. Then you just needed to ensure nothing killed all that life, which mostly just involved sitting back and not killing it, and you had secure, long-term data storage for at least the next few thousand years.
Retrieving the data wasn’t too hard. At the seafloor you just scoop up a bunch of water full of old fish skin, sequence any DNA picked up, extract the relevant encrypted data, and then ping that back to a central dive unit via a form of Morse-like code. And then between there and wherever that data goes next, you protect it through a supreme intellect and a hollow ball of encapsulating steel.
To do all that you need a lot of money, hence the ludicrously wealthy individuals; and you need an excuse to be sprinkling brown sludge at the bottom of the ocean, hence the highly respectable and honorable feeding of an endemic deep-sea life, and the courtesy to navigate and communicate via sound alone.
The term Seafloor-State Drive (SfSD)? It’s thought that came later, some say in a science fiction piece originally written for a competition on DNA data storage, rejected, and then a sentence changed to make it more appropriate for a different publication, rejected again, and then posted up on a website.
Anyway, Axon didn’t care all that much, he was just about done for the day.