Midjourney has just announced their jump from being just the "makes funny images" AI company to being the "revolutionises diagnostics and human medicine forever" AI company, as a side gig. Here's the post.
Basically, they've announced the creation of a highly advanced full-body ultrasound scanner beyond anything that's been done until now. The description of the process sounds straight out of a science fiction novel:
It starts by stepping into a shallow pool of golden light. You then begin to descend into the water. Your body passes through a ring of underwater sensors, each acting like a dolphin, using its echolocation. The sensors send ultrasonic sound waves through your body from every angle. With enough waves, and enough angles, we form an image of what's happening inside your body.
This is meant to make something equivalent to MRI into a cheap, trivial occurrence. You just go and pop in for your daily full-body scan. The one day it happens to find a cancer, no big deal - it's going to be early enough that you can just get it removed swiftly and with little consequence. In fact, you might eventually go to the spa just for fun and relaxation - the scan purely a nice side benefit:
Our spa will have hot tubs, saunas, cold plunges, and cozy rooms with pools of golden light which softly scan your body.
It should be a place you love going, whether it's by yourself, or with friends. It should be available 24/7. The scans are a side-effect. You barely think of them when going to the spa. But suddenly, you have a huge library of data about your health.
That sounds... kinda incredibly awesome? I get regular cheap checkups, and a revived public bath culture? I mean, don't mind if I do.
But then the "too good to be true" bug bites.
I feel like if there's one lesson to take from the last few years, it's that when things that were science fiction turn into reality, they have a way of becoming way more controversial and less cool than they looked in the shows and movies and books. In fiction, even space trucking can be all about the romance of the new frontier, you get to get lucky on Venus and have a ball on Mars. In reality, you'll probably just spend several atrocious days locked in a cramped cockpit, pissing in a funnel because a real toilet would cost precious weight, your muscles too atrophied to go back to Earth before serious rehabilitation, knowing that some dipshit who sells futures on plots of land on Tau Ceti-3 makes in a minute as much as you make in a year.
There are two ways in which this happens, I think. First, the real disconnect - fiction usually has a point to make and a plot to move, so very often it fails to work out all the logical consequences of its new technologies and social structures. You can get a Star Trek episode about some guy falling in love with a holodeck character once, you don't really get a full examination of how a society with realistic holodeck romantic and sexual partners wouldn't just have the one guy now, it would have to have dealt years ago with the problem of people essentially wireheading themselves into one. This happens simply because the holodeck is ultimately a plot device and the exploration of these themes happens only within the easily followed microcosm of the USS Enterprise.
The second way is that it's easy to find those technologies cool and fun when they're in the low stakes world of fiction, but suddenly when they are in the real world, it gets a lot scarier. Even if that fear may not be entirely justified, we do fear them. And now everyone has to deal with them at the same time and make political deliberations about them, even people who didn't particularly think those were cool and fun even in fiction. The world is full of sources of friction for new technology, friction that exists mainly to avoid repeats of past disasters, or to assuage the concerns of certain factions. But that friction also steers and potentially shapes what the new tech becomes. Cloning and biotech are examples of something that on paper had wild potential for both good and bad - science fiction is full of genetically enhanced super-warriors and resurrection by cloning. But in reality, it has mostly turned into a tool for biological research and farming, and not a particularly transformative one; neither the fears nor the promises have come true yet, mostly because we have simply rejected the tech almost wholesale and confined to domains where it can produce incremental innovation but not radically upend our way of life.
And even for the Midjourney Spa, I feel like the future lies somewhere on a spectrum between two paths.
In this path, the promises are overblown and Midjourney is a bit in over its head. We've seen this story before. Lots of innovative ideas, lots of cool concept art, but that all needs to be backed up by the reality of boring, repetitive, solid science, which does not abide by the laws of fancy marketing, and it is only done when it's done, and if it doesn't want to cooperate it can't be made to.
For example, maybe the models are trained on existing medical imaging outputs, among other things. Maybe they can hallucinate. The post mentions seeking out FDA approval, but assuming that they are successful in obtaining it, that is not a 100% guarantee of perfection. In the worst possible outcome, failing or struggling with FDA approval, Midjourney manages to pass this, at least in the US, as a mere "wellness" tool - something that just gives you general health information, and may at most recommend you go see a doctor just in case. In this scenario, the false negatives would be bad as they could generate a sense of false confidence, but the false positives could be far worse. If you scanned yourself every morning - well, most mornings of your life you don't have cancer. But any non-insignificant false positive rate might mean now you suspect one. And since the scan alone wouldn't be enough to go straight to therapy and surgery you would probably seek out a confirmation via specialist appointment and MRI. And that means you would then go and use up the still limited resources that everyone relies on for diagnosis anyway, the main bottleneck. You'll probably take the place of some fellow who really does have cancer, and you'll find nothing, and public health will be overall worse for it. With a low enough base rate of the disease, even a 1% false positive rate will mean a significant amount of misdiagnoses.
In this scenario, the Midjourney Spa makes things worse, not better.
Let's assume the magic Midjourney Spa echo scanner is the first perfect instrument in the history of instruments - perfect recall, perfect precision. What happens then?
Well, at first, of course, no one believes it.
No one believes it because by default, and reasonably, scepticism is the starting position. In every regulatory regime in the west, the burden of proof that some new invention accomplishes any medically useful purpose falls entirely on the inventor. This requires long and expensive trials to meet that burden (plus a long documentation trail, rigorous standards in the design of both hardware and software, and so on and so forth). The FDA demands in the US are slightly easier to clear perhaps than the ones in the EU and UK, but that doesn't make them trivial by a long shot.
But then of course there's also, beyond that, the actual trust of doctors, who tend to be a rather conservative professional class, and are used to facing many promised miracle devices/cures that turn out to be just another scam or sloppy attempt by amateurs. And that can also make or break adoption in hospitals.
Because let's be clear, in this scenario, there will be no golden light spas with relaxing baths. The sonar is a medical device and to reach the volumes described in the post, 50,000 scanners doing 1B scans a month means doing about 666 scans a day, it can't afford to be a relaxing affair: it's in-out for every patient as quickly as they can go. And a quick chat with Gemini [1] also highlighted another problem I hadn't thought of, which is that if you dunk 666 people into a pool every day, the water gets really dirty and unhygienic, so you probably need to re-circulate it and filter it, which seems like the kind of thing that would interfere with the ultra-delicate sonic wave reading. So let's just say that maybe we're not doing 1B scans a month with just 50,000 scanners in a beautiful golden spa. We'll have the scanners set in a white hospital setting (which can be kept cleaner more easily), water changed between every patient, one scan every 15 minutes for a total of say 40 to 50 a day, with a lot of additional costs and complexity, after a lengthy and expensive trial whose costs Midjourney will have to recoup somehow. Leading to something that is probably still
And in the entirely fantastical hypothesis that this ultrasound scanner was truly and genuinely perfect, this would be a pity, because the rules would probably be somewhat constraining its potential. But the rules weren't made for perfect machines based on the observation that we've never had one and we don't expect any in the future. If somehow this entirely novel technique developed in a very short period of time by a company with no medical technology experience with unrealistic targets turned out to be orders of magnitude better than anything like it ever made, then it would be a real pity that our world just isn't ready for that, and we'd probably clip its wings.
I kind of hate to be the party pooper but I'm an engineer and I deal in worst case scenarios [2]. That's my first instinct, to look at this kind of claim and go "hmmm". Ultimately, there's a reason why sci-fi doesn't become real easily, and that's because that "fi" part stands for fiction. But I also don't deny that, to some extent, we probably do sometimes lose opportunities because of this.
The idea of enhancing ultrasound imaging with machine learning and huge amounts of compute sounds interesting. The plan of having such a thing be used daily by everyone as a matter of course is a beautiful dream. Reality has a way to crush dreams of that sort. It does so partly out of brutal pragmatism, having learned that all too often dream peddlers are fools or con artists, and you can't fall for it. To an extent this ends up becoming hardened cynicism - we don't really admit a path for such things to exist, and to an extent we don't really dare dream them much at all, so I'm at least sympathetic to the vision (though let's be real, it's probably mostly just a ploy to drum up hype. We know how Silicon Valley works). I would like to believe that the sci-fi Midjourney Spa & Body Scan could be a real thing by the time I'm 45 [3]. I just know better than to rely on it.
Of all the LLMs around, Gemini. If Gemini can poke holes in your plan, your plan isn't great.
I also deal specifically in medical devices. It's not easy.
(it seems more enjoyable than colonoscopies or having a doctor stick a finger up your butt)