Greggor Mattson

(Posted 9/13; updated 9/14. My original critique is here; and my resource of relevant blogs, media reports, and Kosinski statements is here).

“Under ethical review,” announced The Outline, of the Wang and Kosinski pre-print paper that controversially claimed to use face-recognition technology to detect intrinsic differences between homosexual and heterosexual faces. The statement came from the editor of the Journal of Personality and Social Psychology (JPSP), which peer-reviewed and accepted the paper. This smacks of the journal throwing Wang and Kosinski under the bus [UPDATE: Hal Hodson reports that the publication will go ahead, as it should].

This post explains why the journal bears the brunt of the blame here, and how this controversy is symptomatic of two larger problems: the crisis in academic peer review, and the general disregard of academic expertise on sexuality and LGBTQ lives. The LGBTQ community has long had concerns about research, privacy, and consent: we don’t treat gay dating profiles or gay bars as public places, and neither should researchers.

Critics of Wang and Kosinski’s study, including myself, raised many ethical questions that were unaddressed in the paper. Kosinski has subsequently addressed most of them on Twitter and in his Authors’ notes, and I accept his statements. It would be a shame if this controversy prevented other authors from sharing their work. The journal should own this paper and let it appear in print. The journal’s peer review process should have asked questions about ethics before accepting it, and its statement should have announced it was reviewing its procedures, rather than seeming to displace blame onto the authors [AGAIN: THE JOURNAL SEEMS TO BE DOING THE RIGHT THING].

We only know about the paper because the authors generously shared a pre-print version of their paper, a point first raised by Philip N. Cohen on Twitter, when he wrote, “I also object to trying to stop publication of an accepted paper. They did the right thing by posting an accepted paper preprint.” The furor around the paper’s many flaws (in which I have actively participated), was only possible because the authors are complying with valuable open scholarship values of sharing things as soon as they are available (full disclosure: I have never done so myself). If the journal wanted a discussion of ethics, it should have insisted on them before accepting it. Kosinski has reflected at length on ethics before, and could have done so briefly in the paper if they’d asked.

This controversy is symptomatic of two larger problems: the crisis in academic publishing peer review, and a general disregard of sexuality expertise in the academy. Academic publishing is so “staggeringly profitable” that for-profit publishers create ever-more titles that rely on the unpaid labor of scholars to review, universities and governments demand more publications as metrics of institutional quality, and scholars have ever-more demands on our time. (If you’re new to this massively lucrative scandal, google academic publishing and any of the words: oligopoly, profits, broke libraries, unpaid reviewers).

One troubling outcome of this proliferation of scholarship is that journals often solicit reviewers from authors. This risks groupthink and compromises peer review. It is an inevitable part of trying to produce scholarship beyond subdisciplinary silos that we venture into research areas that have their own vast research literatures. It is impossible to know everything, and that’s why we have a community of scholars who review papers for free. Peer review or an editor should suggested the authors discuss their ethics, reel in their conclusions, and shed some unwarranted theoretical connections. When authors suggest reviewers who inadvertently share our own blind spots, big problems get missed. I don’t know whether author-suggested reviewers were used in this case at JPSP. And if they were, this is not Kosinsi and Wang’s fault, but the journal’s, and the broken system’s.

I am assuming that JPSP reviewed this paper under its normal processes. My discipline of sociology had a scandal where expedited review pushed out Mark Regnerus’ unethical research so it could influence legal cases against same-sex marriage. Again, if this happened, it’s on the journal, and more reason they should be reviewing their procedures.

My last, broad point is that even gold-standard peer review regularly fails on the topics of sexuality and LGBTQ lives. In Wang and Kosinski’s case, the paper traffics in outdated and stereotypical conceptions of sexual orientation and gender expression, describes homosexuality as an issue of “character,” and links grooming styles to prenatal hormone exposure. Any scholar of LGBTQ studies or sexuality would have caught these errors and insisted upon changes or citations to justify these claims. Sadly, this is part of a general problem much larger than JPSP or Wang and Kosinski.

There is a longstanding paradox in the peer review of studies on sexuality and LGBTQ studies. These are relatively low-status research topics in all disciplines, devalued both by our colleagues and our institutions. One result of this devaluation is that researchers who know little about the field nevertheless feel qualified to review or comment, as if there’s no “there there.” And it means that those of us doing the research are either a) asked to review everything across the transom, or b) to not have the status that brings us to the attention of editors in the first place.

There are particular ethical concerns for scraping big databases for information about LGBTQ people. Anyone who has ever spent time on dating site profiles for gay men has come across a cut-and-pasted statement like this:

WARNING: Any institution using this or any of its associated sites for study or projects – you DO NOT have my permission to use any of my profile or pictures in any form or forum both current or future. If you have or do, it will be considered a serious violation of my privacy and will be subject to legal ramifications.

These warnings are a reminder to the viewer that although dating profiles may seem public to an outsider, we treat them as private inside the community. There are strong informal norms, for example, against talking to a coworker about the content of their dating profile–we will both pretend we haven’t seen each other online. We treat them as private because they contain information beyond sexual orientation that are often stigmatized within and beyond the LGBTQ community: the degree to which we are out, our preferences for specific sexual behaviors, HIV status or medical conditions, or information about our relationships. Most dating sites require a user to subscribe to them to view this more detailed information and so there is a barrier to entry.

This is quite similar to longstanding norms about gay bars. Premises in the U.S. that are licensed to serve alcohol are legally public (“public houses” or pubs), but historically we didn’t talk about seeing someone in front of outsiders (I think this norm is observed in many parts of the country and among many age cohorts). To outsiders, gay bars seem public. To those inside the community, they are not, and we’re very protective of our own.

The cut-and-paste warnings on dating profiles first emerged, so far as I can tell, in response to police agencies using hookup websites in the early 2000s, and later because of the constant ethical scandals from researchers treating gay dating profiles as public. The most egregious of these happened only last year, when Arhus University researchers released a database of 70,000 OKCupid users without their consent.

These warnings are a reminder that LGBTQ people already care about privacy and are aware that pictures may be used to identify us, and thus we did not need Kosinski and Wang to lead us. The social movements for disability rights and sex workers rights have an expression: “nothing about us without us,” an insistence that they should be partners in research rather than its objects.

These are the ethical questions I would ask of any big data project that is scraping user profiles to glean information about sexuality or LGBTQ lives, especially on gay dating websites:

• Did researchers have the permission of the company involved? Facebook and OKCupid have permitted such research. This should be disclosed. It alerts readers to the fact that this is research validated by the company, and thus less likely to be critical, but also that the research did not violate the site’s terms and conditions. These are to protect the company, of course, but they also shape the expectations of users, including the question of whether they have a reasonable expectation of privacy. As the Arhus University OKCupid scandal underscores, these expectations are not set by researchers, but by those being researched. Whatever decision was made, tell us in the final published product.
• Did researchers open a profile in order to gain access to dating sites? There may be reasons to justify this, but such requests should be stringently considered by institutional review boards for the protection of human subjects (IRBs), and the practice must be disclosed in studies. This is the equivalent of deceptive research, and conducting it without participants’ consent. This is a high bar to clear, as it should be. In a tweet, Kosinski agrees that such a practice violates research ethics; such a statement needed to be in the accepted paper regarding the dating site their project used to test their algorithms.
• How did researchers deal with individual profiles that included demands to be excluded from research? They must be excluded from any dataset. I have no idea whether their use opens researchers to civil legal actions. But ethically speaking, these are subjects who are explicitly refusing to participate in research and I cannot imagine any reason to override their wishes.
• Wang and Kosinski claim their primary goal was to expose the vulnerability of LGBTQ people to face recognition technology. Hacking norms are a useful metaphor for understanding why these good intentions were insufficient ethical justifications for the way they conducted their research. In a private communication, Jeff Lockhart notes that “white hat” hacking happens when organizations request help identifying and fixing their vulnerabilities. “Black hat” hacking is done surreptitiously for personal gain. He deemed this “grey hat at best,” done with good intentions but still violating key ethical principles like being invited to address a problem and attempting the fix the vulnerability before publication. There are limitations to this metaphor; I am not suggesting Wang and Kosinski hacked any database. As Dan Simpson blogs, both the footnotes of the original paper and the HRC/GLAAD press release describe such a conversation among the parties… and then the research came out and HRC/GLAAD aired their disagreements in a scathing joint press release. While this contains many factual errors it was not, as Kosinski suggested at one point, an warranted smear campaign. It seems there was an attempt at gold-standard ethics here by Wang and Kosinski, and then a breakdown in communication.

Again, these concerns should not bar Wang and Kosninski’s paper from being published. The journal should have addressed them before acceptance, and now it should solicit rebuttals to publish alongside it.

There is simultaneously too little research on sexuality and LGBTQ people, and too much bad research. With some tweaked claims, better ethical procedures, and more transparent disclosures in the paper, the Wang and Kosinski preprint might have been a modest contribution to AI research showing that a trained algorithm could outperform untrained/unmotivated humans at identifying “out” gay people based on their photographs. Researchers do not have to be LGBTQ to do good research on LGBTQ people, but such research needs to attends to the norms of privacy and consent inside marginalized communities. Individual researchers are flawed, peer review even more so, and there are lessons here for all of us.

I benefited from helpful comments from  Jeff Lockhart, Evangeline Heiliger, Patrick O’Connor, Urayoán Noel, and Amaha Kassa.