Quick question: What do an imploding submersible, the RMS Titanic and the rise of murderous computers have in common?
If you said “they all feature in the movies of James Cameron” then yes, alright, sure. You theoretically get a point, and ruin my intro.
For those who have been out of the loop, five hundred migrants died after their boat sank in the Mediterranean last week, which the news grudgingly acknowledged. Their boat - which was, typically for a boat, above the water - seems to have encountered difficulty a full seven hours before the occupants drowned, but it doesn’t seem like anyone was in too much of a hurry to intervene.
Then we lost contact with a submersible carrying five idiots to the sea floor and all hell broke loose.
You can call me hard-hearted if you like, but I had a great time with the “rescue operation” for the missing sub. Four extremely wealthy people and a geriatric submariner had apparently decided to take a submersible to visit the wreck of the Titanic, and then they went missing.
Last month, in-depth scans of the hull of the Titanic wreck and its surrounding debris field were completed, allowing people like me who are sat in a chair on dry land to view every detail of the site. I’m absolutely certain that this will be made available as a VR experience if it hasn’t already, but this wasn’t enough for the super-rich patrons of the OceanGate company, which promised to take them four kilometres below sea level to see the wreck in person.
The patrons were, in no particular order, Hamish Harding, a man who owned a Dubai-based private jet dealership and this unstintingly punchable face:
Shahzawa Dawood, a member of one of the richest families in Pakistan - which means one of the families in Pakistan that fucked over more people than any of the other families, and his nineteen year old son Suleman. Some people have said that the presence of the nineteen year old should garner some sympathy, but to them I say: Imagine how insufferable the average nineteen year old is, and now add “daddy took me to see the Titanic wreck during my gap year” money into the equation. There’s vanishingly little chance this kid wasn’t fucking awful.
Paul-Henry Nargeolet was a 77-year-old French diver who had visited the Titanic more times than anyone else alive, when he still was. I actually do have some sympathy for the loss of Mssr. Nargeolet but again: He was seventy-seven years old and in a submersible two miles below the ocean. At that age, tripping on the rug can prove fatal and he’d clearly been tempting the fates for a long, long time.
Rounding out the cast was Stockton Rush, who I don’t need to tell you was American. Somehow, Americans have never quite got the hang of giving people believable names. A friend of mine describes America as being like the uncanny valley of naming - everyone sounds like they could be a real person, but aren’t. In 1990, Ninetendo released a baseball game in which the non-English-speaking designers had to come up with some player names for Americans. Everyone laughs at the list, but I’m fairly certain that at least four of these people are running for congress:
Anyway, Stockton Rush (former third baseman for the California Grundles, according to Nintendo) was the founder of OceanRush. He charged clients $250,000 per person to sit in a tiny box for five hours and sink two miles down to look at something we already have footage of, and now he’s dead.
At a time when poor people in their hundreds risk - and lose - their lives trying to reach safety by boat, I can only shrug when the super wealthy die in a needlessly risky sub excursion, but I initially found the loss of the Titan submersible particularly funny because all coverage implied that the sub was “missing.” This led to images in my mind of five rich assholes trapped in the dark beneath the waves, arguing bitterly as their oxygen ran out and they came to realise that hundreds of millions of dollars in stock options and a selfie with the Titanic can’t buy you another lungful of air or somewhere to pee in private.
Unfortunately, this dark (in all senses) sitcom doesn’t seem to have come to pass. Evidence began to leak out by Tuesday that a former engineer at OceanGate had raised concerns about the safety of the Titan submersible’s viewing port, billed as the largest of any Titanic-visiting submersible.
Maybe they should have gone with a smaller viewport, because the engineer in question was concerned that the one installed on the sub wasn’t rated to the depths necessary to reach the wreck. Having raised these concerns, he was unceremoniously fired, in case you still had any lingering sympathies for the late OceanGate CEO Ringo Tupperware, or whatever his fucking name was.
If the viewing port in the sub failed, everyone on board was dead as close to instantly as is physically possible. According to this always-diverting graphic from XKCD.com, the sub was at least at the depth where a punctured oxygen tank doesn’t leak air because the pressure of the surrounding water would cause water to rush in rather than air to rush out.
The point at which the surface lost contact with the submersible was probably deeper than that. The sub appears to have lost contact at a depth where it was under pressure of about five thousand pound of pressure per square inch. To quantify this a little better, I measured it out:
Very roughly, each area of the submersible’s hull (and viewing portal) the size of a pencil sharpener was supporting the weight of an adult rhinocerous. Right up until it wasn’t.
I know what you’re thinking: Never mind how many rhinos per pencil sharpener it takes to kill four billionaires, didn’t you say something about killer robots?
First of all, congratulations on an entirely new sentence, but secondly, yes. I was going to tie this whole debacle in to the other big story of the moment, the rise of Artificial Intelligence.
I’ve written before about how artificial intelligence actually isn’t anywhere near as intelligent as people are making out. Which, I guess, means it is “artificial,” just not in the way the name intended. But the current trend for A.I. everything shows no signs of slowing down.
There has been an enormous boom in people attempting to make A.I. art, almost exclusively from people who have no artistic ability of their own. This has coincided with a writers’ strike in Hollywood as the people who write TV and film grow sick and tired of their dwindling wages. An ominous milestone was passed recently with Disney and Marvel’s new “Secret Invasion” series, in which the title credits were A.I. generated.
The plan, from Disney on down, is obvious. Stop paying writers and artists at all, and simply have computers do everything.
This isn’t going to work, for the same reasons as the OceanGate Titan.
For years, those of a paranoid bent have warned about the coming of the technological singularity - the point at which humans create an A.I. that is smarter than we are. At this point, like a cosmological singularity (an infinitisemal speck which explodes to create a new universe), that one A.I. will propogate smarter and smarter iterations of itself until we are faced with all-powerful computers which may or may not decide to wipe human beings from the earth.
This fear has become more mainstream with the rise of Chat GPT and its ilk, but it is so far hopelessly unfounded. The “intelligent” systems like Chat GPT are nowhere near as smart as they’re made out to be. Mostly they’re just an updated form of predictive text generation. If you could write messages quickly on an early 2000s Nokia, you’re most of the way to understanding what Chat GPT does.
The way that modern A.I. has been trained is by feeding it huge volumes of information until it starts to recognise patterns. If you show a computer with a learning alogirthm a million conversations, it will start to figure out that both parties often begin with “hello,” and then move on to small talk questions. This is how Chat GPT learned to write in a conversational style and, more broadly, it’s one of the issues behind the writers’ strike - Hollywood writers are annoyed that their scripts are being used for free as a training tool for the algorithms that are meant to replace them.
This is where the submersible comes in.
In order to train an A.I., you have to feed it information, and sometimes you have to pay for the scripts or books or other documents you’re feeding it. This sometimes costs money, in order to create an A.I. that produces scripts and books without costing anything.
“But wait!” some rich asshole who is yet to be crushed by a submersible says. “What if we save even more money by training the A.I. using books and scripts that have been generated by A.I.?! We won’t have to pay for anything!”
This is very obviously a stupid idea, and leads to what is known as model collapse. In the same way that recording a cassette tape of a cassette tape of a cassette tape leads to something that is ultimately unlistenable - or even in the same way that biological cells replicating endlessly over time eventually lead to senescence and breakdown - having an A.I. train itself will reinforce mistakes and flaws over and over again in a feedback loop until the whole endeavour collapses.
Just because this is an obviously dumb idea doesn’t mean it isn’t going to happen.
Consider: You’re a billionaire. You run a private jet company in Dubai, selling personal aircraft to other smug, punchable climate criminals. Your stepson gives so little of a shit about you that if you were missing at sea he’d spent the evening at a Blink 182 concert, but this isn’t relevant right now. You have more money than a human brain can conceptualise, and one day you decide you want to go and see the wreck of the Titanic.
Do you pay for the safest possible method, or do you try to save a little money - of which you have an endless supply - and go with the cheaper company?
Let’s try another thought experiment: Your name is something like Smurf Tomahawk Jr. and you are the CEO of a submersible company. You make a million dollars per round trip taking wealthy, punchable jet salesmen to the bottom of the ocean to look at a ruined boat. One of your engineers points out that the window in your submersible might not be strong enough to repeatedly withstand the rhino-on-a-pencil-sharpener pressures at that depth.
“That’s an odd unit of measurement,” you think to yourself, but more importantly you wonder if you should stop making trips to the ocean floor while you investigate a safety issue that could kill you and all your hateful, monied clientele.
Then you remember that the Titanic wreck is in international waters and as such your submersible is not held to any legal standard of safety, and also fuck it, there’s money to be made. You fire the engineer for making a fuss and move on with your soon to be shortened life.
The point is that rich assholes will always behave like rich assholes. Not just in the obvious ways, like taking excursions to the Titanic, but in the small, petty ways in which rich people have always stayed rich. Chief among them is not paying a penny more than the bare minimum in any situation.
The rich assholes in Hollywood (and elsewhere) are no different, and they will absolutely train their A.I. using A.I. generated content if it’s cheaper. Other industries won’t be any less penny-pinching. Far from being the start of the technological singularity, we might be living at the apex of the A.I. boom.
Cosmologists don’t just talk about singularities. Whilst it is impossible to know what happens at the end of the universe, one theory is that, with its energy spent, the cosmos will cease to expand and begin to contract in on itself, ultimately diminishing into a singularity once more before exploding again in a new Big Bang, creating a potentially endless loop. It’s called Big Crunch theory.
We’ve just witnessed the small crunch of a submersible. We might be heading not for a technological singularity, but for a Big Crunch of A.I., and it’s probably for the same reasons.