Algorithm Design

The Problem with Engagement Algorithms

Today’s social media algorithms optimize for engagement: likes, shares, comments, time-on-platform. The problem is that outrage, fear, and conflict are highly engaging. Divisive content gets more reactions, so it gets shown to more people. This isn’t a bug—it’s how the system is designed to work.

Decentralized alternatives like Mastodon remove the algorithmic amplification, but they don’t fix the underlying dynamic. Content that gets shared still appears in more feeds—that’s just how sharing works. Without centralized data, you can’t build algorithms that actively counteract this tendency.

Bridging Algorithms: A Different Approach

Researchers have developed an alternative: bridging algorithms that elevate content appreciated across different groups, rather than content that provokes the strongest reactions within one group.

The concept comes from Aviv Ovadya (Harvard Berkman Klein Center) and Luke Thorburn (King’s College London), who define bridging systems as those that “increase mutual understanding and trust across divides, creating space for productive conflict, deliberation, or cooperation.”

Instead of asking “will this content get engagement?”, a bridging algorithm asks “does this content resonate with people who usually disagree?”

This Already Works

Community Notes on X/Twitter uses exactly this principle. A fact-check note only appears if it’s rated helpful by users who historically disagree with each other. The algorithm uses matrix factorization to identify different perspectives automatically. Research shows users who see Community Notes are 20-40% less likely to agree with misleading content.

Taiwan’s Polis/vTaiwan platform took this further for national policy debates. By rewarding proposals that appeal across partisan divides, it resolved previously deadlocked issues including ride-sharing regulation and online alcohol sales. As Taiwan’s former Digital Minister Audrey Tang put it: “We flipped the incentive for going viral from outrage to overlap.”

Why This Requires a Cooperative

Bridging algorithms need two things that don’t normally go together:

1. Centralized data — You need to know what different groups look like to find content that bridges them. Decentralized systems can’t do this.

2. Aligned incentives — A platform optimizing for ad revenue will always drift back toward engagement maximization. The business model undermines the algorithm.

A cooperative solves both problems. Centralized technology makes bridging algorithms possible. User ownership means the platform’s legal duty runs to members, not advertisers—so there’s no structural pressure to abandon bridging for engagement.

User Control Remains

Bridging algorithms are a default, not a mandate. Users will still be able to:

• View posts chronologically
• Prioritize content from specific people or topics
• Adjust how much algorithmic curation they want

The feed algorithm will be open source and regularly audited. When you see a post, you’ll understand why it was shown to you.

Learn More

• Bridging Systems — The foundational paper by Ovadya and Thorburn
• Bridging Systems Project — Ongoing research and resources
• Center for Humane Technology — Broader work on humane technology design