Words. Ideas. Actions.

24 February 2023

I’ve been thinking more and more about content recommendations. Specifically how those recommendations are developed and delivered. With the Supreme Court weighing the role of recommendation engines in radicalization, along with my recent daily use of the new Artifact app and my never-ending desire for a great music recommendation engine, the idea of “content” in “content recommendations” has been on my mind a lot lately. 

There are plenty of more well-informed people writing about the Supreme Court. I’d recommend (I use that term knowingly) getting updates from NPR’s Nina Totenberg and analysis from Slate’s Slate Marc Joseph Stern and The New Yorker’s Kyle Chayka. What I want to talk about is how videos at the center of those cases ended up in front of the eyes of terrorists. To put it far too simply, our recommendation engines are broken. 

One of the reasons I think this is because our current systems and algorithms rarely reward curiosity. Here are a couple of examples: Let’s say you want to know why the bar across the street just erupted in cheers. You look on Twitter, for instance, to see that the Golden State Warriors are playing, and scroll through a few Tweets to learn that they pulled out a win in the last seconds of a regular-season game. The recommendation engine has no idea what motivated your search. All it thinks now is, “MUST SEND MOAR BASKET BALLS!” [Please read that in the voice of Frankenstein’s monster.] But you don’t really like basketball. You were just curious, and found an answer to your question. But now your Twitter timeline is suddenly injected with NBA-related Tweets you have no desire to see. It’s misunderstood you and the limited number of signals it can read about you. There’s no way it knows that you’re five-foot, six and couldn’t sink even a layup even if your life depended on it. But let’s look at another example, shall we?

For people who follow the news closely, there is a lot of value in having a wide array of sources to get your updates. But if you’re using an app like the new Artifact, the narrowcasting can happen pretty fast. If you read a New York Times article, for instance, a lot more Times articles end up in your feed. The same happens for topics. If you read about a local break-in in your neighborhood, you run the risk of having your feed overtaken by crime reporting. It almost makes you reluctant to stay informed for fear that you’ll only get those types of stories over and over. It’s like telling your grandmother when you’re 7 that you like penguins, only to get penguin-related gifts from her for the rest of your life. 

One last one, if you will. Say you have a gloriously broad set of music you like. And, occasionally, you explore bands you read about by looking them up on Spotify. So, you listen to an album you read about on Pitchfork, or wherever, and then, suddenly, your Daily Mix 4 is full of bands who think whistling and stomping and clapping are the-end-all, be-all of songwriting prowess. All because you wanted to try something new. 

I guess what I’m saying is, we need more control. We should be able to explicitly indicate what we like and what we don’t. If we were better at providing that nuance, I’d trust these algorithms more to recommend things to me. And as we move away from cookies and personalization, and toward privacy and control, we might be able to get there. As long as corporations can profit from it, I assume.

See you tomorrow?

10 January 2023

I come to you again tonight with a nugget from a podcast. This time, it’s a line from the recent episode of “Decoder with Nilay Patel.” It was a broadcast of a live discussion between Patel and Chokepoint Capitalism co-authors Cory Doctorow and Rebecca Giblin. If you’re interested in the cross-section of creative work and capitalism, the entire chat is worth your time. But Patel said something towards the end of their discussion that leads me to tonight’s post: 

“There’s a difference between having an audience and having an algorithmic audience.”

Last weekend, we watched “Marcel the Shell with Shoes On.” It’s based on characters created by Jenny Slate and Dean Fleischer Camp in the Marcel the Shell books that we read to our daughter when she was little. The movie is a sweet look at Marcel’s search for his family after a marriage breaks up. That search leads to him creating a YouTube channel to talk about the search. I won’t give away much more about the plot, but Patel’s quote reminded me of a line from the movie that I jotted down in my notebook when Marcel was talking about the people watching his video uploads: 

“It’s an audience, not a community.”

Both of these quotes are so important to the work I used to do. And both the nuance and the differences between an audience, whether algorithmic or not, and a community seem to be important all over again as we get word that Twitter is moving toward an all-algorithmic timeline. I won’t get into the ins and outs of what we prototyped and what we learned while I was there, but trust me when I say that I was very vocal about the design and product decisions when we were testing a purely algorithmic timeline for people. 

Which brings me to the nut of tonight’s reason for all this exposition: If we can trust people to build an algorithm, we should be able to trust them to curate a timeline. What do I mean by that? Well, let’s get into it. 

Years ago, while working with my friend Scott, the social web was just taking shape, and the cost of building an app was falling precipitously. We would often think about mashing up different existing app ideas to create a new one. One of my favorites was what we ended up calling Looksy. It was basically a way that you could share links with your friends to items you had seen around the web that were interesting to you, asking them to “give ‘em a Looksy.” This was back in 2008, when sites like StumbleUpon and Digg as well as RSS feeds were a lot more ubiquitous, and useful, than the walled gardens we have today. But core to our idea was that people, friends even, were trusted link sources for their other friends. That human curation, and the ability to target specific links to specific people, was an essential piece of the experience we wanted to create. 

Later, services like Nuzzle (whose parent company, Scroll, was bought by Twitter bought about a year before Jack Dorsey stepped down), started to fill the space we were looking to fill with Looksy, but it was more of a heat map of popularly shared links from the sources you had already trusted, whether from your Facebook friends or accounts you were following on Twitter. It was a great way for me to keep up with what the people I was interested in were interested in. Which was always one of the biggest attractions of Twitter to me. It was like I had self-selected a couple of dozen editors-in-chief for the most interesting publication on the web each and every day.

This curation concept was also one of the big reasons I always admired the Curation team at Twitter. They were constantly gathering and vetting and explaining the links and information and trending topics on the service. The Explore tab on Twitter became the fastest, and usually best-sourced, breaking news service on the planet. Without argument. I miss it every day. Especially when I need a quick update on something I’m not totally interested in. Need to know the score of a game you’re not watching, the Curation team had you covered with not just the scores but a few Tweet-length highlights so you could get the gist of any match. Natural disasters? The Curation team had the latest, accurate information from trusted sources served up at the top of your feed. And for the all-too-often breaking news of a terrorist attack or mass shooting? Unfortunately, the Curation team had a lot of practice, and brought their humanity and care into crafting updates with the right balance of information and empathy. 

I started my writing career as a news producer for Florida Public Radio. Drafting and editing scripts with our reporters was one of my favorite tasks. But it was tough going more often than not. So, when I see people who get it right, over and over, I have to stop and admire them. It may be why I miss that part of Twitter most of all. Because I know how much work it was. And what a talent it is to do it well. As we lose one more space where careful curation is being replaced by an algorithm which rewards engagement over curiosity, I am lamenting Twitter’s loss all over again. 

See you tomorrow?