Post by Janne Ahrens

AI is great at politics. Until it isn't. A researcher at the University at Buffalo just exposed a critical blind spot — and it changes how we should think about AI in public discourse. Professor Yini Zhang feeds millions of social media posts into machine learning models to study how political conversations evolve online. Gun violence. Misinformation. Democracy. The big stuff. Here's what she found: AI nails the easy calls. → Does this post mention a specific gun law? ✅ AI handles it. AI stumbles on the hard ones. → Does this post frame gun violence as a systemic problem? ❌ Still needs a human. That gap matters more than most people realize. Because the difference between "mentioning a topic" and "framing a narrative" is exactly where political influence lives. It's where misinformation hides. It's where public opinion gets shaped. Zhang's solution isn't to ditch AI — it's to keep humans firmly in control of it. Her team manually codes thousands of posts to build a "ground truth" before AI ever touches the data. The result? A methodology where AI scales the work, but human judgment defines the standards. In a world where AI is increasingly shaping what we read, believe, and vote for — this kind of human-centered rigor isn't optional. It's essential. The takeaway for anyone working with AI: Scale with machines. But never outsource your judgment to them. What's one area in your work where you think human judgment should always stay in the loop — no matter how good AI gets?