Breaking out of the algorithmic comfort zone

Read it first on my Substack

One classic type of algorithm problem is the tradeoff between explore and exploit behaviors. Say we want to go out to eat. We only have limited time to choose a restaurant, and limited money to spend. Are we best served by exploring for new spots, in the hopes of finding a great new restaurant? Or is it better to exploit the places we know about and already like, accepting that they might not be the best possible option?

This is, I think, an interesting lens through which to view our online behavior, especially when it comes to consuming content. I would argue that, for good reason, most of us spend most of our time online in full exploit mode.

Toward what we know

I tried to find data on the typical number of channels an active Twitch or YouTube user subscribes to. These aren’t easy numbers to come by, but we can make an educated guess that the median number is quite low. We can reasonably assume this because the count of subscriptions per channel is extraordinarily skewed — on YouTube, only seven channels have over 100 million subscribers, while there are… thousands? millions? of channels that have 1–10 subs. This probably indicates that most people are subscribed to relatively few channels, and that those channels tend to be among the most popular…