- Connect your experiences as a user or creator on social media to the ideas in the article. What, if any, resonated with your experiences? What, if anything, surprised you? In your own history of social media use, do you remember major changes to platforms that made the algorithmic mode of information propagation more prominent? How did you, and those in your community, react to those changes? How do you feel about them retrospectively?
- What do you see as some of the major benefits of social media? Which of these result from the baseline feature of simply connecting people online? Which ones result from recommendation algorithms? Which ones are specifically a result of engagement optimization?
- For most people on social media, the posts that they see in their feed will have, on average, more engagement than the average of their own posts. Explain the seeming paradox that most people are less popular than the average.
- Do you think it’s possible that recommendation algorithms can be far more accurate in the future (in the narrow sense of engagement prediction), or are there fundamental limits that will prevent this? For this thought experiment, imagine that platforms will have access to more invasive data about users than they do today, and that computational limits aren’t an issue.
- The article says that platform design consists of algorithms, policies, and UX (user experience). But it only talks about algorithms (in fact, only recommendation algorithms). How do the other components of design affect the phenomena described in the article, such as skewed information propagation or political polarization? What about other potential harms of social media, such as addiction, that the article didn’t talk about?
- The Haugen documents include many internal discussions at Facebook on the social impact of the platform, and the engagement-based ranking algorithms in particular. Pick a few documents to discuss (“Is Ranking Good?” and “Distributive Justice at Facebook” may be good places to start.) What do you think about the moral reasoning in these documents? How did Facebook employees navigate the tension between growth/revenue and their responsibilities to society? How well did their interventions work?
- Many scholars have proposed ways to go beyond engagement optimization. Two examples are the paper Building Human Values into Recommender Systems by Stray et al. and Bridging Systems by Ovadya & Thorburn. What are the challenges—whether technical, economic, or political—in making these ideas a reality on large social media platforms? What are some ways to overcome those challenges?
Arvind Narayanan is the Knight Institute visiting senior research scientist for 2022-2023.