Matthew Miner

Mind the Algorithm

September 28, 2023

As recommendation algorithms creep into ever more services, I find myself working to keep these algorithms on the right track.

I first noticed this with Netflix. My roommate shared his password with a friend who used it when babysitting. His recommendations devolved into an incongruous mix of macho action movies and Care Bears.1 Now that I pony up for my own Netflix account, I hesitate to share my own password lest the algorithm start recommending Paw Patrol (I’m a Top Wing guy).

More recently I stumbled across a cover of The Doors’ People Are Strange by The Cog Is Dead. Unbeknownst to me, they’re a steampunk-themed band (think top hats and accordions). Adding the song to my library wreaked havoc on Apple Music’s weekly “New Music” playlist. For a month it served a heavy dose of steampunk-adjacent music. I have no qualms with the steampunk aesthetic — brass goggles and dirigibles, why not — but the musical genre, I now know, is not to my taste.

Worse was my brief foray into remixes of the Super Smash Bros. theme. I gather that fans of this niche also enjoy rap songs about Animal Crossing, because that’s what Apple Music recommended next. Lesson learned: nip that Nintendo nostalgia in the bud.

One more example. Every day Reddit emails me trending posts from subreddits I visit. Unfortunately it includes not only subreddits I join, but any that I visit while logged in. After I replaced my office chair it thought I wanted the latest scoop from r/hermanmiller. When I bought a new printer (a Brother HL-L2370DW, if you care to know), the daily digest yielded printer-related links until I manually told it that printers are boring.

It’s not all bad. Sometimes I’m tempted by a YouTube video that I know is trash. But I stop myself — as morbidly curious as I am about Tucker Carlson’s latest diatribe, how will it affect my recommendations? I don’t want my YouTube homepage dominated by alt-right bile. I’m better off without the brain rot.

I’m not against recommendation algorithms. It takes effort to keep them in tune, but they introduce me to great artists that I would never find otherwise. And they have opportunity to improve. I’m intrigued by emerging platforms like Blue Sky that support custom recommendation algorithms. As for RSS, my go-to source for interesting links and sharp writing, the dumb-but-ideal algorithm of chronological ordering is here to stay.