Measuring the Wrong Things: Machine Learning, Human, Learning, and AI
What do you think? This morning, my husband sent me this article that posits that “male characters are four more times more prevalent in literature than female characters.” (As an aside, Matt’s commentary with the link was “Ironic since men don’t read books.” He’s being funny. But let’s for sure save space for a post about how reading is seen as a “women’s activity” for another day!)
Apparently, researchers at USC have used machine learning algorithms to scour 3,000 pieces of fiction ranging from sci fi, to romance, to adventure and mystery. And they have discovered a 4:1 male-female literary imbalance.
Honestly, I’m not really intrigued by this study, as — let’s face it — it’s originating from an engineering department and one of the co-authors is a Machine Learning Engineer at Meta. I repeat, a Machine Learning Engineer. Ok, cool technology. (And I’m not denying their quantifiable findings, although they do acknowledge some limitations and challenges.)
Here’s what I’m way more interested in: When dealing with real people who are using fiction to learn — let’s call them human students, usually under age 18 — my fervent hope is that curricula provide not just a balance of works written by both male and female authors, but works of literature that include all sorts of characters. I have a small sample set (my kids’ two high schools plus my own experience), but I think they do. Feel free to correct me if your school experience was reading only books by John Donne and John Steinbeck and John Updike and John Grisham (hopefully not that last one) that involved characters named John. I genuinely believe that in 2022, the majority of teachers from all sorts of schools provide a thoughtful curriculum *on this particular topic.* Am I naïve?
So, that is cool what this machine learning algorithm can do. Books *usually* don’t have the same cultural resonance as television or film — and there are waaaay more of them. Which is why I don’t think AI is the best tool for measuring societal impacts of this particular medium. Maybe the utility is in investigating the books that non-robot beings have to read for school…perhaps there would be story there? 🤖📚
originally published on instagram