Some Analog Recommendation Algorithms
On finding your next book (show, album, etc.)
I worry more and more these days about the Recommendation Machine—the feed of endless “for you”s and suggestions and “trending now”s.
I don’t worry because I think computer algorithms do a bad job suggesting quality content; in fact, on average, they almost certainly do a better job than I do on my own. But the key phrase here is on average. A feel-good family comedy is always going to get 4.5 stars—while a bizarre cult classic (for some a 1, for some a perfect 5) will appear in aggregate ratings like a 3. Most great works of art are polarizing, loved by some and loathed by others.
When you’re stopping for food on a road trip, it’s perfectly reasonable to to simply minimize downside—to look for a restaurant you can safely assume is good anywhere. But when it comes to books or films, this strategy is garbage. We shouldn’t be optimizing for merely decent. We should be optimizing for the peaks, not the average—the stories that echo in our minds for years, the films that leave us speechless in the dark.
I’ve argued in the past (borrowing liberally from Adam Mastroianni’s “Pop Culture Has Become an Oligopoly”) that we should ignore more recommendations. In that piece, I advocated for a different set of social norms around recommendations. No one in this story is a villain—we’re all just responding to incentives—but the net result is stale: ever more sequels and reboots and repeat bestselling authors.
At the end of the day, though—literally and figuratively—individuals do need some heuristic for actually deciding what to watch or read. What’s the alternative? The strategies below aren’t particularly groundbreaking, but they are effective, time-tested, and easy for virtually anyone to implement.
Make a List — It sounds so simple, but the key is understanding why lists help. When we fall back on digital recommendation algorithms, it’s often because we’re in a suggestible position: We want to watch TV, say, but we just finished our last series—so what do we do? Open Netflix and scroll through the “Top 10 this week” section. There’s nothing wrong with this, morally that is, but it means that our choice set is vastly restricted and is being dictated by an external party with God knows what incentives. Keeping your own running list of recommendations, by contrast, means that at least in the moment, you are the one driving selection. (Keeping a list also helps buffer against short-lived hype; you can see what titles keep coming up year after year in conversation, and which ones fade after the initial media blitz dies.)
Find Trusted Recommenders — Of course, some recommendations carry stronger weight than others. If you’re lucky enough to have someone in your life whose taste is well-aligned with your own, you can simply outsource much of your search to them. Big data may be able to get your taste mostly right—but it’s unlikely to surprise you with a sleeper hit or a film so bad it’s good, the way your childhood best friend might be able to.
Find Trusted Curators — In absence of an individual you know personally, an editor or independent media organization is a decent substitute. This means, for example, subscribing to a publication that resonates with you, or finding an independent cinema you like, or relying on a favorite podcast host to offer you suggestions. (I think it’s probably fair to say that this is the analog recommendation algorithm that has been most thoroughly displaced by digital ones?)
Swim Upstream — Yet another kind of trusted “recommender” is someone who produces work you like. Read your favorite authors’ favorite authors! Check out your favorite artists’ favorite artists!
Follow a Thread — If you find yourself drawn to a certain topic, it can be very rewarding to read or watch multiple works about it. For some, this might literally mean doing research, but I personally find the process more enjoyable and interesting when the scope is broad or the thematic connection is loose. In recent months, for example, I’ve been reading some books that deal, generally, with “personality”: an overview of the Enneagram; an old psychology textbook; and the character-driven classic, Middlemarch. Far from being a review of scientific literature, this approach lends itself to more random, combinatorial insights. The goal is not systematic understanding but to put the texts in conversation with each other and see what emerges from the gaps.
Necessarily, these analog algorithms are less efficient than their automated counterparts: If you want to find more hidden gems, you probably have to wade through a lot more crap. (Personally that’s a price I’m willing to pay, but it does imply an important corollary: All else equal, analog discovery should probably increase your willingness quickly abandon whatever you pick up.)
It’s possible that in the future, our digital overlords will improve so much and become so creepy individually attentive that they are able to deliver personally tailored, high-quality content without you needing to wade through it all yourself. But right now, for those of us who crave originality, who yearn to be shocked, moved, and forever changed, I think the hunt is worth it.