It turns out the “lesbian renaissance” only has 85 people. 

No, I am not talking about the Renaissance as defined by Chappell Roan, Billie Eilish, Bottoms, and Drive-Away Dolls. That Renaissance is well populated. 

It’s the Renaissance defined by Jenny Watson, a lesbian and self-described TERF (trans-exclusionary radical feminist) committed to the idea that lesbians can only be “biologically female.”

My number comes from Watson’s female-only lesbian and bisexual dating and community app, L Community, which took LGBTQ news and Twitter by storm last month when it claimed it could identify and exclude trans women to a rate of 99.89% accuracy using AI-powered “sex recognition software.”

As of Aug. 7, more than 60 days since launching, the website couldn’t even break 100 users. 

After reading L Community’s definition of biological sex – “biological sex is firmly linked to distinct reproductive anatomies dedicated to producing sperm or eggs for reproduction” – I realized that I – a transmasculine person– fit the bill for “adult human female.” 

So, I checked the box verifying that I was “biologically female,” snapped a picture of my face – and signed up for the dating app. I didn’t shave beforehand, so my testosterone-induced stubble remained in the picture. Chest photos were not required so my flat chest raised no alarms.

Not that any of that would have mattered, Watson is clear that her app can’t be trans-exclusive because “there are many biological women who identify as males and we would certainly welcome those women.” 

I paid and was refunded the $12.75 to verify my identity. And I was ushered into the community, which was notably silent. The only content was from Watson. Posts include telling members the proper dating app portion was on its way via an invite-only basis and asking if anyone wanted to join a Zoom meet up since “our recent event had only 6 attendees.” Another user posted sporadic lesbian-themed memes.

I used my legal name to register, as the platform requested. Conveniently, I haven’t changed my name to Henry yet. At the same time, I reached out to Watson multiple times for comment under the name I publish under and use. 

(To counter any claims of misrepresentation, my chosen and legal names are irrevocably tied together on the internet due to my brief time publishing with both. A cursory search of either name identifies both as associated with me). 

In response to an initial email request for an interview, she wrote “To ensure our message is accurately conveyed, I would prefer to answer your questions via email,” and provided the background “L’App is designed to create a safe and respectful space exclusively for lesbians, utilising facial recognition technology to ensure that only biological females can sign up.”

Watson noted, “This innovation addresses specific concerns raised by many in our community regarding their dating experiences.”

When I followed up with specific questions, as requested, such as the number of active users or their approach to people using the singular “they” pronoun or how they plan to approach intersex individuals, Watson failed to respond in a five-day comment period. I extended that to 7 days out of courtesy, and heard nothing.

Ten days after I reached out with my questions, Watson asked for another week to respond. I provided her with a work-week deadline and never heard back.

Watson’s stances on the non-binary, intersex, and trans community are of public record, however.

Watson had previously described a queer, non-binary musician – who happens to be in a relationship with a man – as “a straight woman LARPing.” She tweeted in dismissal of the inclusion of non-binary and intersex people in lesbian bars and lesbian history. 

In the same interview Watson said trans men were welcome on the app because they are actually women, Watson repeated that no trans woman could be a woman, to the surprise of the conservative interviewers who questioned if Watson’s conviction held “if they have gone through it, and they’re completely a woman now.” It, of course, being transition. 

By the logic presented in the interview, trans men who pass as men, who have testosterone levels equal to that of a cisgender man, and who have received top and bottom surgery are eligible for participation in the community, but trans women who pass as women, have received top and bottom surgery, and have testosterone levels of a cisgender woman cannot. 

Not that passing is something that every trans person wants, can do, or should be a necessity to gain respect or protection from discrimination.

Additionally, Watson’s app may not be open to cisgender women as well.

Watson was quick to tweet against Imane Khelif, the cisgender boxer whose gender was questioned by a coalition of far-right actors ranging from J.K. Rowling to J.D. Vance. (The only “proof” that Khelif has XY chromosomes comes from a highly discredited Russian sports organization).

The L Community website states that: “In humans, biological sex is firmly linked to distinct reproductive anatomies dedicated to producing sperm or eggs for reproduction. At birth, human reproductive anatomy is unmistakably male or female in over 99.98% of cases.” Meaning, that there are only .002% of people who are intersex. 

This statistic is categorically incorrect. The Cleveland Clinic estimates that 2% of people worldwide are intersex. Other medical and advocacy organizations consistently argue that the number likely is 1.7%, drawing from the research of sex and gender biologist Anne Fausto-Sterling.

Where did Watson get that number? It is likely from Leonard Sax, a medical doctor and psychologist, who has argued that 0.018% of people are intersex. Sax has also argued that gender is biologically hardwired between females and males on numerous occasions, including on conservative talk shows and for the far-right think tank the Institute for Family Studies.

Even if Sax’s and Watson’s proposed statistic was correct, Watson and L Community offer no guidelines about the inclusion of intersex people, regardless of their gender identity. Watson’s derision of Khelif suggests intersex people may not be welcome in the community.

This is not the only case where Watson’s assertions may be faulty. Watson initially claimed that her AI-powered software only messed up 0.10% of the time. She provided no proof to verify the claim.

Recent peer-reviewed research from CU Boulder studied gender recognition accuracy in multiple softwares and found that gender recognition software accurately categorized cisgender women 98.3% of the time, meaning that it miscategorized cisgender women 0.17% of the time, or a little less than double what Watson’s app does. 

Importantly, CU Boulder was examining some of the most advanced and well-supported models out there, looking at Amazon, IBM, Microsoft, and Clarifai programs. For those who don’t know Clarifai, it’s an AI-specific company that employs over 100 people. The rest need no introduction.

Not only is Watson working with a much smaller team — LinkedIn estimates 2-10 employees – Watson’s software also must account for the diversity of gendered appearances within the lesbian community, ranging from butch to femme, in addition to differentiating “biologically female” trans men from men and “biologically male” transwomen from women, meaning their software must be highly advanced. 

The Boulder research team found that transgender men were categorized as women approximately 38% of the time and men the remaining 62% of the time, meaning they are incredibly hard to accurately categorize in either direction.

Dr. Morgan Klaus Scheuerman, one of the authors of the CU Boulder study, said, “A lot of people have this view that tech is somehow abstracted from human bias or human values, but it’s not in any capacity.” While Scheuerman knew the topic of my interview, we only spoke about his research, not about the app specifically.

Biometric AI and computer vision – how computers can identify objects or people – consistently shows bias against transgender individuals

Watson’s team manually verifies sex from submitted selfies using a script on the website which uses publicly available datasets and APIs (Application Programming Interfaces). Per UCLA, APIs help dictate how software works and share information. There are several publicly available gender differentiation APIs. 

Scheuerman explains that, “at a broad level, most computer vision works by defining the categories which you want the system to recognize. In gender, this is often male or female.”

As Scheuerman’s research explains, large data sets of images, qualitatively labeled by people for specific characteristics like gender, can be trained to predict those qualities in future images.

Since the foundation of computer vision is human training, Scheuerman says, “these generative AI models, or these large foundation models, ideally can do anything you want them to do.” 

Fundamentally, Watson’s model wants to differentiate between ciswomen and transwomen. Since existing computer models successfully read transwomen as women most (87.3% per Scheuerman) of the time, Watson likely needed to train her model specifically for its task.

The specifics of Watson’s model remain under wraps. But ostensibly to get the level of accuracy, Watson’s model must have been trained on photos of both transgender women and cisgender women, in addition to transmen. This raises questions of consent. 

Where did Watson get the photos? Stock photo websites often include collections of transgender people available for republication, but some explicitly exclude their collections to be used in Machine learning or AI data, while others encourage it. Research has found that AI models often use copyrighted work as data to train models, regardless of if they have explicit permission. 

That is even if Watson used stock photos. “Scraping” data from publicly available sources like social media is very common for AI training and research and has previously been used to target trans people.

For example, an investigation by Vice found that the University of North Carolina Wilmington scraped more than 1 million images of trans people from YouTube without permission to create a dataset to learn more about terrorism. An interesting research question, seeing as a highly disproportionate number of terrorists are not transgender.

Although we don’t know how Watson went about sourcing the data used to train her model, the broader question remains: What would models think about their photos being used in this way? 

Shae Gardner, director of policy at LGBT Tech, who has worked in the field of tech policy and research for eight years, says, “While there has been zero transparency in how this app’s facial recognition system was trained, if it involved the non-consensual scraping and inclusion of images of transgender women, that constitutes a severe breach of privacy, trust, and consent.”

Gardner emphasizes that “developing a technology with the explicit goal of identifying members of a marginalized group raises significant ethical concerns. Openly stating an intention to use that technology to exclude said group confirms them.”

Scheuerman says that “a lot of people have this view that tech is somehow abstracted from human bias or human values, but it’s not in any capacity.”

He hopes that “the field of computer science would be more open to understanding these types of concepts [like equity and diversity] because they’re our responsibility and a moral responsibility. Plus, it’s actually valuable within the market.”

The politics of consent and AI are just beginning to be negotiated and already have led to multiple lawsuits.

The first trans-exclusionary lesbian app Giggle for Girls, started in 2019 by Sall Grover, is currently facing a lawsuit from a transwoman, Roxanne Tickle. The app shut down in August 2022 with 20,000 members. Grover’s Twitter bio says the app is under renovation and will be re-launched soon. 

Grover and Watson used to be collaborators of sorts, having joined each other’s podcasts to hype up the small world of female-only dating entrepreneurs. 

That collaboration seems to have soured as both are claiming to be the first trans-exclusionary dating app. Giggle started first, but Watson claims it did not begin to discuss dating – just finding community – until after L Community launched. Grover claims otherwise.

However, a dedication to in-person events is unique to Watson’s mission. She plans to open a bar in London for women — her definition — only. It will be a member’s only club, so the exclusion of transwomen is legal.

Watson recently hosted a counter event to London Pride, protesting trans and asexual inclusion at the event. Estimates Watson promotes put her event at 150 people. To put that into perspective, their event was under .005% of the size of London Pride. 

These numbers are not surprising. A 2023 YouGov survey found that 84% of cisgender lesbians think of transgender people “very positively” or “fairly positively.” Another 13% don’t care (“neither positively nor negatively” and “fairly”). Only 3% felt “very negatively” about trans people.

The Her App, a trans-inclusive lesbian dating app, that has critiqued Watson and L Community, has more than 15 million users. Grover’s app before it shut down was 0.0013% the size of that. Watson’s app is .000005% the size of that. 

Perhaps no comparison is more jarring to show that Watson and her followers are a stunning minority within the lesbian community.