Our team recently published a paper introducing a tool to reduce echo chamber effects on social media. Essentially, we’ve developed an “allostatic regulator”, grounded in psychological theory, that can be applied at the inference layer of pretty much any existing recommender.
Think of it as a code wrapper that helps machine learning algorithms offer a more diverse content diet, by subtly nudging users away from extreme algorithmic reinforcement, based on a user’s recent content viewing history. We've shown it works in simulated environments, but the big next step is real-world deployment and testing at scale.
We believe this tool has significant potential for platforms like Google and Meta – not just for user mental health, but also as a long-term strategic benefit for their ecosystems. So we're really keen to get this into the hands of researchers and engineers who are working on these problems. If you're at Google, Meta, or elsewhere and this resonates, we'd love to explore how we might collaborate on testing it in a live setting.
You can read the full paper here: https://www.tandfonline.com/doi/full/10.1080/29974100.2025.2...
Any advice or feedback is very welcome!