Mind and Iron: A tube train that can zip by at 600 m.p.h. will begin testing next year
Also, can an algorithm save our kids' mental health? And AI weapons keep getting doomier.
Hello good Mind and Ironers, and welcome back to another whirlwind edition of our weekly newsletter. This is Washington Post and LA Times long-timer Steve Zeitchik, every week bringing you all the future scoopage like the milk man brings the calcium.
This week I'm coming at you live from Toronto, where I'm meeting some sources, working on some stories (see one of them below) and checking out some movies at the Toronto International Film Festival.
A quick reminder to please sign up here for our Thursday missives on all things AI and the years ahead if you’re not already signed up.
And an equally quick reminder to please make your pledge to our human-centric tech journalism here — for a few bucks you’ll get unfettered access, special columns, podcasts and everything else coming down the road, without worrying about paywalls or your nagging conscience.
First, the future-world quote of the week:
“This is going to change people's connectivity at a regional level — but a regional level now means a thousand kilometers.”
—Sebastien Gendron, head of Canadian startup TransPod, which is on its way to building a tube-train capable of traveling 600 miles per hour
This issue we look at a provocative new machine-learning model that may be able to help predict teen suicide; AI weapons that are making the Pentagon way too excited; and what Meta could really do with its new AI system. And our Totally Scientific Apocalypse Score hits a new low for the year. It’s like the Aaron Rodgers of apocalypse scores.
Also, did you ever wonder why people in the future — or China — can board a train and get a gazillion miles away in half an hour? A wild visionary out of Canada named Sebastien Gendron (the man above) is attempting just that with a company called TransPod. We drop in on him to hear what he's up to.
Oh, and a totally random but possibly appreciated tip from here in The Six: On Monday night I saw a movie that left me stunned in my seat — Jonathan Glazer's Holocaust film "The Zone of Interest,” about a seemingly ordinary German father and mother during World War II raising their kids, splashing in the lake and planting their garden against a backdrop of a random large wall. Except that wall is the wall of Auschwitz, and the father is a Nazi mega-commander who organizes death camps.
A more effective work of dramatic minimalism I can’t imagine — the movie damns privilege, complicity and compartmentalization by what it doesn’t show. I don't know if this is the future of filmmaking, but it's certainly the most forward-looking way of understanding the past. Out in theaters from those paragons of taste A24 on Dec 8.
OK, let's get to the messy business of building the future.
IronSupplement
Everything you do — and don’t — need to know in future-world this week
AI weapons get troublingly hot; Meta goes deeper on AI; Digital guidance-counselors?
1. TEENAGE SELF-HARM STATS HAVE CLIMBED UPSETTINGLY in recent years; suicide rates alone are up 29 percent over the past decade.
Can math and computers predict who's at risk?
It's a tantalizing prospect. Right now to identify trouble signs we tend to rely on self-reporting (not always so useful with a teenager); parental observation (not always so useful with a teenager); and guidance counselors (not....always so useful with a teenager). Most predictive models, for their part, look at past self-harm attempts, which only help after there's been one and may not be that accurate anyway.
So researchers at New South Wales University in Australia tried something else.
They took existing data of nearly 3,000 Australian teenagers from the 2000's and trained an algorithm on some several hundred risk factors — from home environment to social interactions to scores on established anxiety tests — to see how a computer would do making those predictions.
This month the researchers published their results. They found that their model could accurately predict whether a teenage subject would or wouldn't try to harm themselves in 74 percent of the cases, and whether a teenage subject would or wouldn't attempt suicide in 72 percent cases. (With the predictive arts you want to be right in your yes-no prediction more than 50 percent of the time; otherwise you'd be better off flipping a coin.)
But how much better were these results than the extant previous-attempt model? Those were 65 and 63 percent accurate, respectively.
So basically the new machine-learning system can get it right in nearly 10 percent more cases. And as more risk factors are identified or discarded and the model is refined, that number should go up.
AI is already being used by companies to identify depression in adults, processing facial expressions and voices in ways some researchers say are more accurate than the going standard, a nine-question self-reported test. (Last year I wrote about one such startup, a company called Thymia out of the U.K., which shows promise.)
This Australian study would seem an equally encouraging development. If the model raises or silences an alarm that helps ten percent more teenagers, who would deny it; if machines give off signals we otherwise wouldn’t hear, why wouldn’t we tune in?
Yes to all those Panglossian vibes — but with a caveat.
First, there's a lot more data to be collected. (The researchers have admitted as much.) I'd also worry about the infallibility trap. Right now when a therapist makes a call about our kids, we might balk at or take it with a grain of salt. That makes their job hard, but it gives us an important check on their influence, asks them to explain their diagnosis.
When a machine makes a call, we don’t have the opportunity for such interrogation — just an ironclad statement of confidence. That isn't nothing, but we're also a long way from knowing exactly how to listen to it — from the right balance of credulity and wariness.
Not to mention the possibility of stigmatization. “Based on patient information, the ML algorithm could calculate a score for each person, and that could be integrated into the electronic medical records system,” one of the researchers, Ping-I Lin, suggested.
Comments like this should give us the sweats. Another voice informing us of what we might we want to be on the lookout for is great. A machine-like Big Brother slapping mental-health scores on children that become part of their record? Tread carefully.
2. AI WEAPONS ARE A SUBJECT TOO BIG TO TACKLE IN ONE BOOK, LET ALONE ONE NEWS-ROUNDUP ITEM. But I wanted to call your attention to the comments made by Palmer Luckey, the chief executive of a firm named Anduril. A VR wunderkind and early metaverse advocate (I talked to him in 2014 just before he sold his company Oculus to Facebook for $2 billion), Luckey has lately shifted his work to —yikes — AI weapons. (He took that Zuckerberg cash and used it to found Anduril.) We’re a long way from sitting in a virtual movie theater with your cousin in Tuscaloosa.
Last week Luckey announced “Fury,” a drone meant to fly alongside fighter jets/bombers and basically guide them to their targets with brutal digital efficiency. (The drone was created by a company Anduril just acquired.)
This comes on the heels of the announcement of several other Anduril products, like Lattice, an AI system. One use case for Lattice? Equipping cameras along Texas border towns with it so that guards can more easily follow and assess anyone who might try to cross, which is one reason the Department of Homeland Security has already bought the tech. No, this isn’t stone-cold “Children of Men” scary at all.
If you weren’t already feeling discomfited by some of the brightest tech minds working with massive amounts of money to come up with the most ruthlessly accurate weapons yet, just read some quotes from an interview that Luckey gave the military publication Breaking Defense in announcing Fury last week. He was positively giddy.
—“There are multiple Anduril products that are selling at already larger scale than some of our products that are on our website, we’ve been selling them for well over a year, they’re doing things that are incredibly cool.”
—”ChatGPT has probably been more helpful to Anduril with customers and politicians than any technology in the last 10 years.”
—“I think you’re seeing a lot of hawkishness on the part of venture capitalists.”
—”There’s so many people in the Pentagon and on Capitol Hill who have had that come to Jesus moment just because of the hype cycle around ChatGPT, which I am very happy to leverage in getting them excited about the future.”
All that’s missing is him asking if we’ve ever danced with the devil in the pale moonlight.
[Breaking Defense and Vice]
3. DEVOTEES OF THIS NEWSLETTER KNOW WE’RE NOT INCLINED to cover the Big Tech companies as obsessively as some of our MSM counterparts. A realm worthy of ink, to be sure. But future-forming goes well beyond what thought Musk or Zuckerberg had at the kitchen table in the morning and yacked out to investors in the afternoon.
This week yielded plenty of breathlessness about Meta’s new AI plans, after the Wall Street Journal reported Sunday that the company is working on building out a new system that would compete with Open AI’s GPT-4. A totally respectable piece of news if you’re the WSJ and writing for people who are deciding what to think about or invest in the company. But what effect will this have on all of us, who see these megaliths falling over themselves to build these AI systems and wondering, basically, ‘what’s all this jargon got to do with me.’
Here’s a quick breakdown:
Meta’s been a little behind in the AI game compared to OpenAI/Microsoft, Google and the other big cats. They had a system called LLaMA that was mainly for academics, then a system called LLaMA 2 that’s not been used as widely and is produced in partnership with Microsoft. And now, according to this story, Meta execs are buying chips and building data centers to start training a model next year that would be wholly their own and could (maybe) be as powerful as the latest GPT.
(Meta, unlike some of its competitors, has made its AI systems open-source, which is either a great act of generous democracy or a 4D-chess trick to squash competitors by making its own system more prevalent, depending on your level of Zuckerberg cynicism.)
But what will Meta actually use this AI for — or, more sharply asked, how will we use the AI model that Meta is producing? On this answers get scarcer. Zuckerberg has talked about how AI will be integrated into Meta’s social-media products. Eg: there could be dozens of different AI “personae” that FB or IG users can interact with.
That’s the kind of top-down idea that sounds good to an executive, but is mystifyingly atonal to the rest of us. Isn’t Facebook’s whole appeal that it connects us to each other? It seems more likely if Meta and AI are going to be entwined, it will be because we interact with AI beings created elsewhere — whether they’re ‘friends’ or brands or anything else— that just happen to use Facebook or Instagram as their conduit. Not because Meta’s AI system is something we must use because it’s Meta and we must not use anything that is not Meta.
Look, the company may well turn out to be a major player in AI, its system so strong it pops up everywhere in our digital lives. Or not. Which models will be powering what applications is still very much a Wild West. (I mean, few people were even covering OpenAI until this past year.) But there’s no reason to think Meta has any inherent advantage in becoming dominant beyond the generally true facts that they have money and users, and I can’t help feeling like the zealous coverage is a function not so much of Meta unlocking some secret power of AI but of observers assuming that a new wave of technological innovation will be dominated by leaders of a previous one. When both history and common sense tell us this is rarely the case.
[WSJ]
The Mind and Iron Totally Scientific Apocalypse Score
Every week we bring you the TSAS — the TOTALLY SCIENTIFIC APOCALYPSE SCORE (tm). It’s a barometer of the biggest future-world news of the week, from a sink-to-our-doom -5 or -6 to a life-is-great +5 or +6 the other way.
Here’s how the future looks this week:
A MENTAL-HEALTH TOOL THAT CAN PREDICT TEEN SELF-HARM? Good, but perhaps proceed with caution. +2
META USING AI TO CREATE NEW PERSONAE IN OUR SOCIAL MEDIA? Feels like No. 251 on the list of best AI use cases. -.5
SUCCESSFUL VISIONARIES ARE GETTING TRACTION ON AI WEAPONS, WHAT COULD GO WRONG? Besides everything. -7
The Mind and Iron Totally Scientific Apocalypse Score for this week:
-5.5
The Mind and Iron Totally Scientific Apocalypse Score for this year:
-7.0
MindF#ck
Notes from the Future’s Edge
Beam me up, TransPod?
When you think about high-speed trains shooting down a tube — and really, why would you? — you probably think about Elon Musk and his promises for a Hyperloop tunnel through Los Angeles a decade ago. And yet still here we sit, stuck in a logjam on the 405.
The idea of a pod shooting us down a low-pressure/quasi-vacuum tube — faster, greener and safer than other forms of transport — is a compelling idea in theory, with plenty of science to back it up. But it's also a massive challenge — incredibly costly, and filled with logistical and regulatory hurdles. Even traditional high-speed rail can barely get out of the station — check out what's bedeviled the L.A.-San Francisco project over the last bunch of years.
And now here comes Sebastien Gendron. Gendron is the co-founder and chief executive of TransPod, a small, punch-above-its-weight future-forward transit company based near downtown Toronto. I stopped off at its offices this week to get the skinny from Gendron (my Uber over there was a Tesla; the universe has a sense of humor).
A France-born engineer, Gendron and his inventor co-founder Ryan Janzen have pressed ahead with a pretty radical idea for transit — and gotten surprisingly far with it to date. Their idea is 54-person pods that can zip between cities in a fraction of the time it would take to drive, train or even fly once you factor in all the airport meshugass. You enter the pod (one comes along every two minutes), strap in, zoom down a tube and a short time later, presto, you're at your destination.
"We’re talking about the convenience of a subway with the speed of an airplane,” Gendron told me.
At their top speed, these craft, known as a FluxJet, go not the 80 miles-per-hour of an inter-city train or the 150 m.p.h. of an Acela or the 250 miles-per-hour of a Chinese high-speed rail but...the 600 m.p.h. of a sky-hugging airplane. (Gendron likes to call his pod an "aircraft without wings.")
A smattering of companies has sprung up around the world as Musk himself seems to have leaft the tube-train idea for others to pick up. TransPod is on this cutting edge. So is a Dutch company, Hardt, and a German project out of a Munich university. (Hyperloop is sort of but not really what they’re called, since that was Musk’s name; it’s an Aspirin/aspirin thing.)
The engineering behind these tube trains is a mix of old and new. Basically, electric engines propel the pods, and the low-pressure environments allow them to move very quickly. The key is making sure the pods don't hit the insides of the tube. That's where TransPod's innovation comes in. A couple of the other companies doing this use opposing magnets — ie, some on the pod, some on the tube — to create an attraction-repulsion effect that keeps them from bobbing up or down (which would be, um, bad).
TransPod's trick is to use attraction magnets only on the pod, with an algorithm determining the level of attraction, thus regulating the amount of space it leaves between the pod and the tube. This, the company says, makes the entire caboodle much cheaper.
(Btw, theoretically these trains should be able to exceed even 600 m.p.h., but small amounts of air leakage into the tube create shock waves, and you don’t want to be feeling those at 1,000 m.p.h.)
That’s the science-y part. Here’s the practical one. The planned first launch for TransPod is a line between Edmonton and Calgary, a 180-mile Albertan journey that takes around three hours to drive, plus the traffic variable. Travel by TransPod would take about 45 minutes, no traffic variable.
The timeline should look like this, Gendron said: Construction on a few miles of test tube near the Edmonton airport starts as soon as late 2024; people-less tests for said route begin in 2026; construction on the tube for the full 180-mile route starts by the end of the decade; and the line opens for passengers as early as 2035.
That seems….pretty ambitious. But even adding a few years here and there, we're talking a not-insane time frame. If Gendron's aims are borne out, we're only a few presidential administrations/15 more years of Jets playoff drought away from regularly whisking ticket-buying people through vacuum tunnels.
(I was about to ask if you'd get in a pod that traveled at 600 miles per hour through a tube just above the ground, and wondered if in fact I would, but then realized I already get in a tube with wings that travels 600 miles per hour much higher off the ground.)
As for optimal distances, the sweet spot is about 250-500 miles — it takes 30 miles just to accelerate and another 30 miles to decelerate, so a short trip isn’t worth it. And building a single line to traverse half the country isn’t feasible. But all the in-between distances could work. “This is going to change people's connectivity at a regional level — but a regional level now means a thousand kilometers,” Gendron said in his typical softspoken but still sorta bold manner.
TransPod has goals of uniting a whole range of cities — a Chicago-Quebec line, with stops in places like Detroit and Toronto, for instance. And then to replicate that model in various regions of North America.
“The goal,” he said, “is an intercontinental subway system.”
Which sounds like a hyperbolic tag-line until you remember:
1. It seemed impossible in the 19th-century for most people to imagine we could have breakfast in Lisbon and lunch in New York instead of enduring a weeks-long scurvy-riddled ocean voyage.
2. The early subway systems, like the London Underground, were at first laughed at or waved aside by many people too.
Oh yes, there’s the money. So far TransPod has received $550 million from Broughton Capital Group, a financier out of the U.K. Public money isn't on the table yet; that would potentially come later once the actual tubes need to be built and the costs for the Alberta project rises well into the billions.
So there’s going to need to be some serious capital if this will happen. And getting public funding when so many existing infrastructure needs abide in both the U.S. and Canada won’t be easy. And actually building the tubes on time and on-budget seems like a big ask considering how even that kitchen remodel went three months over.
Critics have raised these points and instead countered with high-speed rail, which entails new track but not new tech; they ask why spend billions on something unproven that will take decades and may not work scientifically or commercially. Indeed, even Gendron admits freight must also be a big part of this; passenger movement on its own won’t make this pricey project profitable.
But if Gendron or one of his competitors could pull this off the implications would be massive. What if Disneyland and Silicon Valley were just a 90-minute flash through a tube? What would New York feel like if reaching Boston or Philly was the same low-effort lift as getting to Westchester or Staten Island? If the Midwest were a giant hop-on hop-off extravaganza?
What would the implications be if we could truly unite places that have for so long been disparate? If cities far from our own will seem less like exotic destinations and more like an easy day trip? Gendron talks about this phenomenon as the physical equivalent of a type of shrinking that's been happening now for a quarter-century. “The Internet is already making the world more connected, and this will emphasize that even more,” he said.
Indeed, there’s a spiritual component to all of this. Distances are not only hassles — they’re barriers to cultural and social exposure. Drastically reducing them can give us a lot more of those interactions.
Conversely, cultures that grew up very distinctly — writing this from my hotel room I think of the differences between Toronto and Montreal — might start to entangle in all kinds of ways after entire generations are raised with the ability to bop from one place to another. These tube-trains don’t just change how we move around North America — they change the North America we move around in.
There's no way to know if any of this will work, at any level. But at regular intervals in the last few centuries our means of transit has dramatically transformed, from horses to locomotives to motorships to cars to airplanes, shrinking distances as they went, and it would be downright bizarre and history-blind to think this won't continue to happen.
“Naysayers have been here throughout history,” Gendron said. “The Wright Brothers after their first flight went back to get funding and the U.S. government turned them down. The builder of the Eiffel Tower had to self-finance the first sections. When it comes to radical innovation, 80 percent of people look at the glass as half-empty. But we keep evolving and progressing. Why would we think we’d move around 50 years from now the same way we do today?”
Perhaps climbing onto a pod that zips through a friction-less tube isn't the next phase of transit development. Perhaps it's something else. Pretty great, though, that in the next few years TransPod may have well built a tube to help us find out.