Can we Create Healthier Social Media Algos? w/ Researchers from UNIMIB

Описание: In this episode, I sit down with the research team, from the University of Milan-Bicocca, who wrote the paper “Balancing Benefits and Risks: RL Approaches for Addiction-Aware Social Media Recommenders.” Their work represents a rigorous and nuanced attempt to mathematically model the tension at the heart of the social media business model: how to balance user engagement with long-term well-being.

The first half of the conversation focuses on the paper itself—its core methodology, the design of the simulated user environment, and the way reinforcement learning algorithms can optimize for more than just click-through rates. We explore how the team modeled addiction, agency, and healthy decision-making, along with the findings from their simulations.

From there, the discussion zooms out. What are the current frontiers in recommender system research? How can we shift the incentives of tech companies to care about long-term user flourishing? And what kinds of policy or institutional frameworks might actually support healthier design choices in real-world platforms?

It was fun to engage with an entire research team for this podcast. Each person brought their own perspectives and ideas to the discussion–we even got an anime reference at one point.

The team was pretty pessimistic about the prospects of businesses transitioning to healthier recommender algorithms on their own, which I found interesting. Instead, they felt policy intervention and, especially, educational reform would be the drivers that move us in that direction.

Our hope with this podcast was to have a discussion with people at the cutting edge of a research field. We hope you enjoy this discussion and let us know if you’d like more of this kind of perspective.

You can learn more about the team's current research efforts by checking out their Social Media Data Collection Project: https://biconnect.psicologia.unimib.i...

Connect with Tusk & Quill on socials: ‪@TuskandQuill‬

And check out our website for more articles and podcasts: www.tuskandquill.org

CHAPTERS

Introduction 0:00 - 5:15
Understanding the research approach and findings 5:15 - 25:30
Advice for building healthier recommenders 25:30 - 33:00
Cutting edge of designing healthier recommenders 33:00 - 46:20
Where should we intervene? 46:20 - 1:01:35
The future of AI and social media dynamics 1:01:35 - end