I read the article and I find this fascinating. I am happy they are giving it some thought. Pronoun political correctness (a.k.a. thoughtfulness or respectfulness) aside, if our society's training data leads AIs to make assumptions, we should be aware of it and be careful to detect it.
An incorrect pronoun isn't going to kill someone -- but we are gradually handing over more and more decisions to technologies that rely on AI/ML. Perhaps investigations into incorrect pronoun assumptions can lead to improvements in assumption errors in other areas (e.g. you think your self driving vehicle doesn't need radars and that only using cameras is sufficient? Just because it looks like a big fluffy cloud doesn't necessarily mean you can safely fly/drive through it).
Back to just pronouns though: if someone says "My teacher assigned me to read Act I of Macbeth tonight", most people would avoid a reply that uses he/she until an indication was given or they'd just ask "Who's that, Ms. McFadden? Yeah, I know, she assigns way too much homework!". If a human can be smart enough to get it right, then I'm glad folks are working on AIs getting it right too (or, for now, not making assumptions until they can get it right).
The fact that AIs are advanced enough for us to be thinking about these kinds of details is wonderful! :)