Seeing Like an Algorithm — Remains of the Day

🌈 Abstract

The article discusses how the design of the TikTok app helps its algorithm work effectively, and how this "algorithm-friendly design" approach could be a model for other companies with machine learning-powered products.

🙋 Q&A

[01] How TikTok's Design Helps Its Algorithm "See"

1. Questions related to the content of the section?

• The article explains how TikTok's design, with features like full-screen video playback and paginated feed, allows the algorithm to gather clear signals of user sentiment and preferences. This contrasts with the infinite scrolling feeds of other social apps, which make it harder for the algorithm to accurately gauge user interest.
• TikTok's design creates a closed feedback loop where the app's tools and features enable the creation of the very content the algorithm needs to train on, solving a "chicken and egg" problem.
• The article argues that as machine learning becomes more important, companies will need to prioritize "algorithm-friendly design" that helps their algorithms "see" the data they need, even if it adds some friction for users.

[02] Comparing TikTok's Design to Other Social Apps

1. Questions related to the content of the section?

• The article contrasts TikTok's paginated, single-video design with the infinite scrolling feeds of apps like Facebook and Twitter. The latter make it harder for the algorithm to accurately gauge user sentiment, as there are fewer explicit positive/negative signals.
• Apps built around interest graphs, like Reddit, tend to incorporate downvoting mechanisms to help the algorithm weed out uninteresting content. But most major social apps rely more on positive engagement signals.
• The article suggests that minimizing friction for users is not always the best approach if it comes at the expense of providing clear signals to the algorithm.

[03] The Importance of Algorithm-Friendly Design

1. Questions related to the content of the section?

• The article argues that as machine learning becomes more central to products, companies will need to prioritize "algorithm-friendly design" that helps their algorithms "see" the data they need to perform well.
• This may involve adding some friction for users, like TikTok's paginated feed, in service of providing clearer signals to the algorithm.
• The article suggests that companies with dominant market positions may be resistant to this approach if it means disrupting user experiences that have worked well for them so far.
• However, the author believes that creating a strong feedback loop between the app's design, user interactions, and algorithm training can build a powerful competitive moat that is difficult for rivals to replicate.