The title might be a bit of an overstatement, but the basic argument is that on major platforms, nonabusive adult content was systematically banned when far-right extremist content flourished without censorship. This is especially bad for LGBTQIA+ people, who both lose important connections and suffer from right-wing attacks. Although we often say that all tech starts with porn, Monea argues that that’s really true only of standard heteroporn. Only when “media technologies are accessible and affordable but not yet overly regulated [can] more niche pornography … flourish. Wars on porn often crush niche pornography first due to its lack of access to capital. These first victories often exhaust the political capital of anti-porn crusaders and appease at least conservatives by achieving a heteronormalization of pornography. overbroadly and ends up censoring large amounts of nonpornographic content, particularly sex education materials, LGBTQIA+activism, and LGBTQIA+ community-building discourse. This overbroad censorship is especially prevalent once the rhetoric of children’s unwanted exposure is used to drum up support for anti-porn regulation. Once this rhetorical trope is leveraged, it easily becomes possible for people to perceive the unwanted censorship of some nonpornographic material as immensely preferable to even a single piece of pornography slipping through and being seen by children.”

You probably know about algorithmic bias; there’s also algorithmic bias in AI porn, given that the images available overrepresent mainstream heterosexual porn. Even if you don’t care about porn, Monea notes that this overrepresentation means that algorithms are better at distinguishing porn from non-porn for LGBTQIA+ content, which affects what gets screened out. Meanwhile, workers are labeling training datasets with terms like “closet queen.” “The architecture of the dataset demands that stereotypes about what constitutes the successful performance of a particular sex, gender, and sexuality become hardwired into the visual dataset. Regardless of which images end up populating the category, the category’s very existence determines the way a computer will see—it will see stereotypically. For example, two men hugging, especially from behind, is a key indicator of closeted homosexuality.”

Consider the fraught question of the oft-censored “female nipple.” As a transgender woman asked, “At what point in my breast development do I need to start covering my nipples?” Monea argues that it would be presently possible to identify all human nipples and just have a setting: I’m willing to see nipples/I’m not willing. But heteronormativity makes biased filtering seem like a better default. (I think Monea understates the complexities here: breastfeeding parents tend to object strongly to being classified with other nipple-barers, and whether or not they should, that does matter to the conditions platforms face.) Likewise, Facebook distinguishes between “real world and digital art in the context of adult nudity and sexual activity because we have historically found that digital images are hypersexualized.” That sounds ok, but consider who historically and today is more likely to have access to the means and training to produce oil paintings and sculptures v. digital art: there’s a bias against LGBTQIA+ communities. Monea argues that this kind of monoculture impairs sexual imaginations; “while SafeSearch does not really seem to have ever succeeded in preventing adolescents from accessing pornography online according to most studies, it is successful in heteronormatively channeling adolescent porn use.”

Monea doesn’t spend enough time on the “compared to what?” question and could have benefited from some Evgeny Morozov. He wants us to have the ability to “take random walks through and engage in serendipitous discovery of new materials in our digital pornographic milieu.” But it’s not obvious to me who actually wants that. ( )