AutoLike

Recommender systems make millions of tiny choices about what reaches our attention.

Recommendation systems shape much of what we see online, yet the algorithms behind platforms like TikTok, YouTube Shorts, and Instagram Reels remain largely opaque. In this episode, Kyle Polich speaks with Hieu Le, a researcher studying privacy, automated systems, and machine learning, about Auto-Like, a framework designed to audit social media recommendation systems through automated interactions. By framing the problem as a reinforcement learning task, Auto-Like systematically explores how user actions such as liking, sharing, bookmarking, and skipping content influence what platforms recommend next. The conversation explores how researchers can evaluate whether recommendation systems steer users toward specific topics, sentiments, or potentially harmful content without requiring access to proprietary algorithms. Hieu discusses the technical challenges of auditing black-box systems, the role of reinforcement learning in discovering effective interaction strategies, and how tools like Auto-Like could support researchers, policymakers, and platform designers seeking greater transparency and accountability in online recommendation systems.

Hieu Le

Federal Trade Commission’

I am currently serving as a Senior Technologist at the Federal Trade Commission’s Office of Technology, where I leverage my expertise on privacy, automated systems, and applied machine learning to further the agency’s mission to protect consumers and promote fair competition.