Recommendations
How to find cool stuff.
published on 10.25.23
note: I focus on reading recs in this, but it applies to nearly all content you'd want a recommendation for. I also have a ton of writing on why you'd want a recommendation in the first place, but removed it as it felt more like background context.
How did people get recommendations before the internet?
With the internet, these methods haven't drastically changed, they've just expanded. The main difference is that you can get "in person recs" from strangers anywhere on Earth. You don't have to be physically near them or actually be in person. These kinds of recs are happening on modern forums, like Reddit and Twitter. The "exploring worlds" method is functionally the same in the internet-era, but it is just easier and more powerful now. This is because things are so well-documented and (crucially) the documentation is so accessible. All the crucial information for the "exploring worlds" method is indexed online. To figure out what authors George Saunders enjoys is very easy, to find a list of every James Baldwin book ever written is trivial, and curating an information circle (through follower lists, RSS, newsletters, podcasting) is the status quo. And finally, in many ways, the "trusted curators" method has also been strengthened by the internet. Authorities are more decentralized and you can therefore find a more personalized authority (whether a "booktuber", Goodreads reviewer, or specific person sharing their thoughts online) you actually trust. You don't need to defer to The New York Times Best Sellers list or Oprah's Bookclub. "Authorities" are more niche now and their decision making processes are more transparent. Furthermore, you don't need to make it past the gatekeepers to gain access to authoritative sources any longer. You don't need to be admitted to Harvard to gain access to their syllabi and you don't need to travel to the local book store to chat with expert readers.
All in all, that isn't bad progress for the internet-era rec. But I still find it surprising we haven't innovated on our methods for recommendations given how connected we are now. It feels like the improvements to the recommendation methods are all incidental to the internet. In other words, there have not been many deliberate attempts to improve the recommendation by software companies. Social Media (like Twitter and Reddit) have helped with getting better recommendations, but not by explicit design. They've done this simply because they are modern (virtual) gathering places. And where people gather, recommendations get made. Yes there are individual YouTube channels, Tik Tok users, and sub-reddits that are based around recommending good stuff - but that is just because they are platforms for expression and many people are interested in recommendations (giving and getting).
An objection to this framing might be that these companies knew the best way to share specific recommendations is to have a decentralized culture, where groups of users interested in different things can build little niches in the broader ecosystem. But I don't think the goal was ever to improve recommendations. Rather, the goals of these companies seems to be more general. Something like: let people connect with the groups and interests specific to them. Recommendations are (again) just a subset of those interests.
Another objection to my framing might be that the algorithms so central to the most popular online spaces are an innovation on recommendations. They are sort of. The underlying idea of this use of the word algorithm (as in "Tik Tok's Algorithm") is to observe a person's behavior and then recommend what they would enjoy next. That sounds like a recommendation method distinct from the previous three discussed. Indeed, I think the algorithms designed by modern companies like Google, YouTube, Amazon, Facebook, Twitter, Tik Tok, Netflix and others are a kind of recommendation innovation. They just aren't explicitly aimed at long form content (with the exception of Netflix and the other streamers who have in many ways shown how effective this new method is at serving up long form content). Tik Tok is recommending you individual videos they think will occupy you for 30 seconds and keep you scrolling. Twitter is recommending you individual sentences. Facebook is serving you ads and... friend recommendations? Honestly the best recommendation algorithm for longer form reading might be Amazon recommending you books based on your purchase history. What might we discover if we aimed this algorithmic method of recommendation at books and other long form writing like essays and articles? How can software explicitly aimed at improving recommendations change the way we find good writing?
I am going to stop here for the moment and come back to explore the below questions.
- What are the exceptions to this? Which companies have explicitly aimed to improve the recommendation with software? Goodreads, Medium, Substack, etc.
- Why has there been so little innovation around this?
- Other ideas for improving recommendations with software.
Collected reading on this topic