Sites such as LinkedIn, Facebook, Instagram etc are driven by algorithms. These algorithms are generally opaque, often intrusive (in terms of the automated decisions they are making based on what is often extremely personal data), and rarely revealed1 publicly in any real detail.
What all of these obscure algorithms appear to have in common, however, is that their behaviours range from bizarre to badly biased when they are starved of input to work upon.
My facebook accounts have zero personal information about me – they are exclusively for accessing information which can only be found on facebook and which can only be easily accessed when logged in. And naturally I only use them in private/incognito/InPrivate™ browser sessions. The algorithms desperately try to push content to keep my attention, but what they dredge up is fascinating – mainly in how utterly & completely they fail in this task. The “people you might know” suggestions are hilarious and weird in equal measure, and comforting (to me) since they have yet to suggest anyone that I actually know.
Whilst my boy-mode LinkedIn page does contain quite a lot of information, my girl-mode LinkedIn pages does not. Thus it is fascinating to see the types of jobs their algorithm considers suitable.
It makes me wonder how the algorithm perceives, assumes, or deduces gender – because there is no “gender” setting in LinkedIn to explicitly specify the user’s gender. There is this in the help section (although note the “Last updated: 2 years ago” which is inherently suspicious for something described as “being rolled out”!)
Note: If you choose the pronouns she/her or he/him from the pronouns dropdown (options available in select regions only), we may use this selection to infer your gender.
LinkedIn help
I didn’t select “she/her”, I used a custom “She/Her when in girl mode” description, but two years on it may be more flexible in its assumptions.
There is also this which is much wider:
LinkedIn may infer your age and your gender based on information in your profile. For example, we may infer your gender based on your first name or on the pronouns used when others recommend you for skills. We may infer your age based on when you graduated from school. We do not infer age or gender based on profile photos.
LinkedIn help
So I’m going to assume that the algorithms are classifying me as female from the (limited) profile content, most obviously from the first name.
Now take look at the jobs which are emailed to me and described as “matching my skills” – noting that I have (somewhat tongue-in-cheek) described my skills list to LinkedIn as “Nightlife, Nail Art, Shoes, Makeovers, Partying, Wigs, Nail Care, Makeup Artistry” which may directly explain one of these, and possibly indirectly hint towards two others. But where do the rest come from? Piercer? Model2?(!) Spa Consultant? Mystery Shopper? Cocktail Server⁈ Freelance Model⁈ Part Time Administrator (work from home data entry)? Independent Optometrist⁈⁈
Interestingly, the algorithm has completely ignored my membership of the LinkedIn group “IT/Engineering Freelancers in London” which would seem (to me) to offer a substantive hint about what might be relevant.
My suspicion is that these suggestions are the result of biases based upon the gender determination, exposed by the paucity of information upon which the algorithms can operate. These biases probably come from unconscious biases held by the creators of the algorithms.
Maybe – just maybe – if I write a little more technical content, the algorithms will pick this up and start sending me irrelevant and unsuitable IT jobs too… Although these days I host it off-site leaving thin pickings actually within the LinkedIn site.
Overall, my approach will remain that of starving these algorithms. Partly because I don’t want them manipulating what I get to see. But also because when starved their behaviours are interesting and revealing.
Sophie Baskerville 