How Community Notes Reduce Viral Misinformation | Keith Coleman, Jay Baxter | TED

Overview

This discussion explores the development and impact of Community Notes, a crowdsourced fact-checking system on X (formerly Twitter). Keith Coleman and Jay Baxter detail how the platform combats misinformation, fosters trust across political divides, and leverages AI-human collaboration to scale its effectiveness. They also envision a future where social media promotes common ground rather than division.

Notable Quotes

- A random person on the internet wrote a note, and suddenly, leaders of the free world changed their public statement. That’s like a superpower for people. – Keith Coleman, on the power of Community Notes.

- Our algorithm takes advantage of partisanship and polarization to ensure notes are fact-checked from every angle, resulting in highly accurate and neutral content. – Jay Baxter, on the surprising agreement mechanism.

- If all we did was pursue the areas of agreement, I think people would be pretty happy with the direction the world was going. – Keith Coleman, on the potential of identifying common ground at scale.

🛠️ The Mechanics of Community Notes

- Community Notes allow users to add context to misleading or incomplete posts, including those from heads of state or official accounts.

- Notes are only displayed if rated helpful by users with differing perspectives, ensuring neutrality and trust.

- The system is open-source, allowing anyone to verify the algorithm and data, which fosters transparency.

- Unlike traditional fact-checking, which can take days, Community Notes often appear within hours, sometimes as quickly as 20 minutes.

📊 Leveraging Polarization for Accuracy

- The surprising agreement algorithm identifies notes that are rated helpful by users with opposing political views.

- Polarization is used as a feature, not a bug: people with strong opposing views are more likely to rigorously fact-check notes, leading to higher accuracy.

- Notes that fail to meet quality standards or are overly biased result in contributors losing their ability to write notes.

🤖 AI-Human Collaboration

- AI is integrated into the system to draft initial notes, which are then refined by human contributors.

- Human feedback on AI-generated notes serves as training data, improving the AI’s ability to produce accurate, neutral, and well-sourced content.

- This collaboration accelerates the speed of note creation while maintaining quality.

📉 Impact on Misinformation and Behavior

- Posts flagged with Community Notes see a 50% drop in reposts and engagement, driven by organic user behavior rather than algorithmic downranking.

- Authors of flagged posts are more likely to delete their content, reducing the spread of misinformation.

- The system has proven effective across diverse topics, from AI-generated war imagery to political claims, demonstrating its scalability.

🌍 Building a Pro-Social Media Future

- A pilot program highlights posts liked by people with differing perspectives, encouraging constructive dialogue and reducing polarization.

- The team envisions applying this approach beyond social media, such as in policymaking, to focus on areas of agreement and foster societal progress.

- By identifying and amplifying common ground, the system aims to transform social media into a tool for unity rather than division.