I was browsing Wired as a fine news consumer is want to do, and I saw an article by the title “Can We Still Rely On Science Done By Sexual Harassers?” I thought this was a bad idea, then I read the article and was more confused. While I think you should read the article yourself, I wanted to pull out some quotes, and provide a summary so you can understand my annoyance (if you’ve read this article, skip to the bottom of the below section).
Sexual harassment and abuse are present in every field, and big examples in Silicon Valley and the media have colored the content we consume (paragraphs 1-3).
“There’s a word for that kind of work: ‘problematic.’ It’s stuff you love tainted by people you hate.” “Is Ender’s Game less of a masterpiece for Orson Scott Card’s homophobia? Maybe. Looking hard at the flaws of the artist is an important way to engage with the art.” (p 4)
Sexual harassment and abuse occurs in science as well, but unlike art, the results and conclusions of science are intended to be objective and separate from their human creators. Yet humans do bearing (p 5-6).
Godwin’s law (p 7). An astronomer at UC Berkeley prevented students from doing their best work but discovered bunches of exoplanets (p 8-9). There were social and professional consequences but the science he developed is still used and he gets acclaim for having discovered things (p 11).
John Searle, also a professor at UC Berkeley, had a lawsuit filed against him. Searle developed a thought experiment called the Chinese Room. (p 13-14)
“In brief: A guy stands in a sealed room with a slot in the wall. Every so often a slip of paper comes through the slot with some Chinese characters on it. The guy looks up at a display in the room and sees another set of Chinese characters expressing the right response. He writes those on another piece of paper and pushes it back out the slot. So: Does the guy in the room understand Chinese?” (p 14-15)
(Apparently this has to do with AI and consciousness, but it’s not important here (p 16)) The important take away is that the Chinese Room thought experiment is separate from Searle’s sexual assault/ harassment transgressions (p 17).
Bias in machine learning and computer science is a compounding force and has influenced the databases we teach machine learning on. Bias in programmers also prevents them spotting problems as computers learn (p 19-20). While this might not be bad for some programs, on the margins it could hurt those who are overlooked by the algorithm (p 21).
“If the lesson of #metoo is that monsters are everywhere, they’re in Silicon Valley, too.” (p 22)
Machines learn from what we teach them, and we will embed them with our bias (p23).
So, why can’t we trust science done by sexual harassers again?
This article confuses two serious problems (almost certainly to capitalize on clicks in the wake of the #metoo campaign) the first being sexual harassment in science (but more generally all industries), and the second the bias we build into AI systems we create. The connection between these ideas is weak in reality, and more so in this writing.
The professors used as examples, both created work which endures without the stain of association. Those exoplanets are going to sit up in the sky for a small fraction of eternity. Conversations about strong AI will persist until we have an answer, weather understood through the Chinese Room or another thought experiment. The best thing we can do to remove the “problematic” nature of these findings is to do what we do with all knowledge, abstract it away from its roots. We don’t worry ourselves with the pros and cons of nineteenth century Christendom when we use Mendel’s theories of selection. We have taken the useful knowledge, removed the “problematic” history and allowed the facts to exist on their own. The same thing we have done to most knowledge throughout history. Remembering the “problematic” origins of our space program, country (USA), and language is the realm of historians not scientists or engineers*. The best thing we can do as average scientists is forget about the assholes.
*Not to say knowing the history of your field is unimportant, only that it is accessory to the fields focus, discovery and invention respectively.
But back to the problems. Sexual harassment exists in every industry. Apparently because bad behavior is too? And while power allows these transgressions to go unpunished, as in the case of Weinstein, these people exist at all levels within groups. Pretending like the problem only exists in the directors, producers, and tenured faculty, throws the baby out with the bath water.
My problem with this article is that it’s not the sexual harassers who are injecting bias into our machines, its literally almost everybody. I’m not sure if fixing the white guy problem will instantly solve machine learning bias, but diversity probably will, cause the same problems occur with other homogenous groups. If that isn’t a screwed up parallel to the algorithms are racist posts, I don’t know what is.
What parallels there are between bias and sexual harassment, is in their distribution. Bias exists in (probably most but at a minimum) tons of people and is distributed throughout groups. There are people with strong bias and those with weak bias, but like sexual harassment we only hear about things when they go wrong (and if anything, our bar for evidence should be lower than it is). The mistake is thinking these are the same groups. You can have unbiased people who sexually harass, and the most bias person imaginable who can keep it in their pants. But this has nothing to do with the facts these people create. No one’s science is beyond reproach, but writing it off because of who discovered or created something is bad science and worse for our society. Asking if we can still rely on science done by sexual harassers is the same as asking if we can believe the science done by the victims. While bias is its own problem, it doesn’t track with harassment that neat and clean, and maliciously tangling them together for nothing but a click-bait headline is disingenuous.
P.S. Chinese has been used as a synonym for incomprehensibility, just, like, Greek. I bet if we peered into Searle’s head when he first thought up the Chinese Room problem, the man in the room would be balding with has brown hair, a little pudgy and in need of glasses. Because he defaults to his own body image, just like you did when you first read the scenario.