RECENTLY, A FRIEND sent me this meme.

I asked around and found @Rivelino to be the initial poster. (A whole piece should be written on how shitty that account is, but that’s for someone else.) See below for the first usage I could find (at right) and the reasoning behind it (at left).

Given that there are TEDtalks on the right power postures for excelling at service jobs that won't ever pay enough to let you own a house or raise children, the "Don't Lean In" meme is hardly more ridiculous. In my view, this meme and other dominance hierarchy memes that reach the mainstream after originating in this heavily online community (e.g., the Virgin v. Chad meme and its variants) are often: 1. hilarious, 2. semi-accurate if you squint your eyes at society enough, and 3. largely benign. Here's the Virgin vs. Chad meme getting posture-checked.

But why should anyone care about people on Twitter drawing green lines on photos of other people? Generally, they shouldn’t. Essentially, this is just another iteration of the alpha/beta dominance hierarchy format that is ubiquitous within heavily online communities. As artist Joshua Citarella pointed out, however, the aesthetics of the Don’t Lean In format—green lines generalizing the relative positioning of figures in an image—has a lot in common with how computer vision algorithms are often displayed. I find this as good an opportunity as any to discuss the systematizing design reasoning that increasingly appears linked to the incel-adjacent online world and the way in which its generally regressive ideology may be gaining visibility and acceptance in the mainstream. As someone who works in machine learning, I can’t help but think about how the green lines in the Don’t Lean In memes are similar to an old school computer vision preprocessing technique called morphological skeletalization.

At a high level, this works by iteratively scanning an image and eroding the borders of each object pictured, pixel by pixel, until only a single line of pixels remains. This same process can also be useful in recovering line data from corrupted images such as fingerprints (the canonical example). Also relevant here: the applied field of computer vision that is skeletal body tracking.

But the meme also contains strong aesthetic links to the field of machine learning. Image classifiers often utilize a pure green bounding box to denote the region of interest containing the image's target object.

As this same green features prominently in the corporate branding for the company I work with, I asked the founder what significance or use-value he felt it held.

Looking deeper, I found that the human eye is actually more sensitive to certain pigments, and pure green the most. This is because the photoreceptors in human eyes are unevenly distributed across the visible light spectrum. By contrast, birds and bees have more evenly spaced receptors—and therefore more equal sensitivity to different tones.

The evolutionary reason we developed this kind of vision (called asymmetric trichromatic vision and shared by all primates) is thought to be for how it helps us identify small differences within a wide range of plants: for example, to better determine when fruit is ripe or what is or isn't edible. There is also a compelling explanation regarding primates’ sensitivity to red pigments, which is useful for social coordination (blushing, being red with anger, pulling a whitey).

In 2019, Kate Crawford and Trevor Paglen launched a viral, facial recognition app called ImageNet Roulette. Matching the aesthetic conventions of the machine learning field, the app used pure green (RGB schema 0,255,0) to draw the bounding box around the user’s face on screen. (Likely, green is the default selection color for computer vision systems not only because it is so easy to specify, but also for how easy it is for the human eye to see when overlaid on any other given image). From there, the app would assign you to a class,  with labels from the innocuous and inane (anchorperson, skier, healthy) to the blunt and problematic (kept-woman, money-grubber, slave), which it had inferred from your image. Users were left to debate what about their visual appearance caused the algorithm to assign the specific class it did. Of course the point of Crawford and Paglen’s work, at least in part, was to demonstrate what biases get hardwired into the process of mechanical “understanding” of images from human generated labels. For me, seeing it pop up as a face filter on Instagram solidified this connection between the seeming authority of a solid green bounding box and mechanical judgement.

I think it’s important to contrast the rising ubiquity of mechanical judgement with the original Mechanical Turk, a truly impressive 18th century parlor trick involving an “automatic” chess opponent (after which Amazon named its b2c slave-labor-as-a-service). Created in the latter half of the 1700s by Wolfgang von Kempelen, a Hungarian civil servant, the contraption worked via a series of levers, magnets, and strings, allowing a human operator to hide inside its candlelit wooden box and control the attached "robot" to play a game of chess. The operator’s seat could even be moved around so that prior to a demonstration, nearly all of the doors of the machine could be opened revealing only gears, pulleys, and empty space, ensuring that the illusion was convincing.

After debuting the machine at the Austrian royal court, Kempelen toured it around Europe before passing it along to other owners. With a skilled operator inside, the Turk could win most games, including with such notable losers as Ben Franklin and Napoleon (who tested the machine by cheating multiple times during his games). Even Gary Kasperov's knee-jerk reaction was to claim there had to be chess grand masters secretly remote-operating Deep Blue when he started to lose after game 2 of the 1997 rematch.

While the roots lay in these earlier events, it feels like there has been an inflection point in people’s confidence in technology’s “judgement” within my lifetime: one that budded with the machine recognition of cats in Youtube videos and blossomed with 18-time world champion Go master Lee Sedol losing to Google DeepMind’s AlphaGo. The shift we’ve witnessed is humans now placing trust in a system not for producing results that we can confirm to be true (cats) but for producing results that are explicitly beyond the human capacity to independently verify. We have come to grant a system value because, not in spite, its use of artificial intelligence.

In short, we can think of this shift as an appeal to algorithmic supremacy.

In Silicon Valley, this is not a new trend. Since ~2015, founders have had better chances of raising rounds of funding by claiming their system to be "AI." A 2019 report from MMC Ventures found that companies labeled “AI” on Crunchbase-like sites for European start-ups, attracted between 15 to 50% more funding. The same report found that roughly 40% of these companies in fact had absolutely no true AI at all at the core of their IP! (Fwiw: in a market that will support an AI powered toothbrush, there is little incentive for founders to correct the tagging if mislabeled as an AI company. They’ll get there eventually, right? Growth mindset!)[1]

As this appeal has spilled beyond the zone of Modafinil-fueled pitches to VCs and into the Overton window, I see this 'appeal to algorithmic supremacy' providing two rhetorical functions: 1. no matter which conclusion is reached, the result is given more authority due to the fact that it’s come from a computational process rather than the irrational depths of the human mind; and 2. any fault in the conclusion is shifted away from the presenter, allowing them to shrug and reply, “It’s what the machine says!”

It is extremely unlikely that @Rivalino thought through this string of connections before making the first Don’t Lean In meme. Probably, the aesthetic choices were even subconscious. And yet they nevertheless channel a much larger trend of hyper-systematic thinking that I have been seeing grow in this online space. It’s important to stress that, for a lot of incels, “not having sex” may be what initially brings them to the community, but what many stay for is a collective acceptance of (or even obsession with) the seemingly hard realities of dominance hiearchies in society. And it’s understandable how, when faced with a process as complex as socialization, there’s this desire to just formulate your thesis via two green lines. In fact, quantitative thinking-driven reasoning isn’t even what we need to worry about with this crowd. The whole logic is so self-limiting and rigid that it hardly has broad appeal. (If you don’t know what I mean, go on incel.wiki and see the obsessive love of systematizing for yourself.)

Of course this, too, is nothing new. But what does feel novel to me, or at least like something that I’ve only recently noticed, is how different parts of these highly online, right-adjacent circles have become markedly more aware of their aesthetic strategy. I'm thinking, here, of such Gen-Z heirs to Mencius-Moldbug-style neo-reactionary discourse as Kantbot, whatever that perfume guy's name is, BAP, and that Twitter with stock-photo-esque images of beaches and attractive women interspersed with the occasional image-plus-antisemitic-caption of a buff white guy.[2] (There are also some old NRx Tumblrs of architecture that precede all of this but they’re too rinsed by now to include).

Before saying more, I want to be clear that I have no interest in giving these figures any kind of special importance and I sincerely do not wish to platform them. (To at least minimize the signal boost, screenshots in this piece omit usernames.) There is, however, something going on here worth observing—an advanced version of the systematically-minded design thinking of incel communities, which is being applied to what remains a generally regressive set of trad values (i.e., the racial prejudice and gender norms of the “simpler” pre-modern time they prize).

Complicating this is that the memes that circulate in this world tend to rely on many layers of irony and meta-irony. So my guess is that their core values end up remaining largely opaque to everyone who shares them, even to those on the inside. In a good number of the accounts I come across, the prevalent way of making a claim is what you could call argument-by-moodboard. (AOC’s skincare-routine-as-political-key can’t compete.) I fully get it though. These layers of irony, this symbolic deflection—like owning a copy of the 1748 novel Clarissa and doing kettle-bell reps—not only allow for plausible deniability (‘it’s just a novel,’ ‘just an exercise routine’) but are honestly much more attractive than spending all day online and talking to the internet about hating women.

Another point of clarification: I do agree that modernity has its problems. I just firmly do not see the way out as backwards! And it is troubling to me that some of the public figures within these “return to tradition” communities are: 1. learning how to aesthetically encode their messages in a way that can communicate complex systems of hierarchies in a single image (as compared to the left’s tendency to relay complexity with long-winded, detailed language), and are 2. all but excited that, by default, their desired “beautiful” futures favor white men (and their subservient white female partners) above all. I find this repulsive and feel the need to warn everyone that this is the impression I'm getting here.

I struggle with my relationship with this content a lot. I generally don’t think one should attempt to stop art or, for that matter, any kind of creative expression. It’s also usually futile to try. But aesthetic literacy is a reasonable goal.

When, in 1909, F. T. Marinetti published the first Futurist manifesto on the front page of Le Figaro, his embrace of mass cultural production and mechanical reproduction was seen as a thrilling pivot of the avant-garde (previously associated with bohemian enclaves). Ten years later, the Futurists—who, not unlike the aforementioned realm of the highly online, increasingly propagated chauvinistic “traditional” values—would become a mainstream political project that profoundly contributed to the formation of fascist ideology. But thanks to the Futurists’ status as avant-gardists on the one hand (popular conception of artists as good-hearted and socially progressive) and embrace of utopic-seeming technology on the other (progress free from the problems of irrational human bias!), their ideas enjoyed easy passage to the mainstream.

Hopefully I’m completely paranoid and this only looks like a growing issue to me because I'm online as much as I am. It's easy to forget that just 22% of Americans are on Twitter. But if this aesthetic turn is real, then an algorithmically reductive, systematized obsession with a regressive male ideal (and gendered counterpart) is being exported to the mainstream, encrypted in memes that critique modernity while claiming masculinity is a perfectible machine.

1.  It's important to remember, that Facebook did this in 2015. They launched a personal assistant named M, after acquiring Alex Lebrun's startup wit.ai, claiming it was a "human powered digital assistant." The idea was simple: have a chatbot handle the things for which there are API calls (Uber scheduling, checking the weather, etc.) and have a human do the hard stuff, with each exchange carried out entirely within the Messenger app. The "trick" was that the humans would record their method of solving more complex tasks and future iterations of the digital assistant would be trained on the solutions the human contractors found.

Unsurprisingly, the project shut down in 2018 with few details ever revealed, but one can imagine what a nightmare it must have been. I really would love to know how Facebook asked its (almost certainly minimum wage) contractors, whose job title was literally "trainer," to logged their "solutions" to complex tasks or how much money the company ultimately burnt on the project. We do have an estimate for the latter: fin.com, a company started by the co-founder of Venmo and (after M launched) an ex-VP of product at Facebook, was doing a chatbot assistant very similar to M, charging users $1/min with a minimum monthly spend of $120. They have since pivoted into a b2b analytics firm.

2. The majority of the content of this essay was written two weeks ago. Between then and now (while the piece was with NM being edited), there was a large fracture along the virgin/chad axis of the heavily online right, from what I can gather in part caused by Kantbot facedoxxing himself in Alex Lee Moyer's shitty documentary and the BAP crew finding that as good a time as any to separate themselves, calling KB fat and ugly.

It was always essential that at some point the kantbot/logo/etc ppl would have to define themselves against BAP style bc they are so opposed. The kantbot of “tfw no gf” is incompatible with the faceless semi-porno assemblage of beach model pics that comprises “BAP imaginary.” @mcrumps

I have a softer spot for KB and Logo than I do the BAP crew. Even if he wraps it in a larger, coded vision of the future, KB speaks about the problems of capitalism in a way that doesn’t diverge too far from leftists I appreciate. Meanwhile, the BAP crew’s philosophy just straight-up seems to be about gaining strength and power so as to not be a subservient Bugman; it’s as if the only selection pressure on their ideaspace is whether something makes you bigger and more bronze. (I can't wait for the first sun emoji Twitter person—a reference to the Black Sun—to get skin cancer due to "bronzing"...)

One last note on this topic before someone drags me: I am aware of how terrible Logo's poetry is (tho with this whole crowd it’s hard to tell when it’s trolling or serious or serious trolling, etc.) and that's not what I find concerning. Instead I'm thinking about how the idea of getting tan and muscular, and going outside with your hot girlfriend has mass appeal. And how this particular variant of the online right could be capable of seamlessly carrying its encrypted message to a much wider audience.

Duncan Wilson manages the hand tracking team at Ultraleap. He is based in the Bay Area. Comments and thoughts can be sent to dncnwlsn(at)protonmail.com and desk(at)newmodels.io.

Published: May 2020