Risky Science Podcast · Risk Market News

Confidence as a Service With Eric Winsberg

·1 hr 1 min·10 clips

Westfall frames the interview around a hard question. Every model makes choices about what matters, what gets left out, and how much uncertainty the person using the output is willing to hear. Winsberg answers from the philosophy of science, but the conversation keeps pulling him back toward practical risk work. The result is a discussion that treats modeling as an institutional act, not just a technical one. COVID-SIM becomes one of the key examples. The show uses it to ask how a model can look authoritative while still resting on debatable assumptions. Climate models come next. Winsberg and Westfall discuss the areas where they are weakest for North Americans, especially wildfire, sea level rise, and tropical cyclones. That weakness is not presented as failure in the simple sense. Instead, it becomes a way to examine how uncertainty is handled when the stakes are high. The conversation then shifts to the trade-off between confidence and informativeness. A narrow band of output can look decisive. A wider band can be more honest but less satisfying to hear. Winsberg treats that as a value judgment as much as a technical one. He returns repeatedly to the difference between appearing precise and actually being useful. Westfall keeps pushing with careful follow-ups that ask what a model really communicates when it speaks with authority. AI enters the discussion as another layer of complexity. Once machine learning sits on top of physical models, the question of whose values are inside the output gets even harder. The episode keeps circling back to that problem without turning it into a slogan. It stays focused on the machinery of uncertainty, not just the headlines around it. By the end, the conversation has made a simple point feel sturdy: models are never just descriptions, because the people who build and use them are always deciding what counts as enough certainty. The closing exchange lands on that judgment. It is less a verdict than a reminder that confidence itself can be a design choice.

As heard by us

A careful look at how models package confidence, uncertainty, and judgment.

Risky Science Podcast frames modeling as a choice-filled act, not a neutral machine. Chris Westfall's conversation with Eric Winsberg moves from COVID-SIM to climate models to the layer of AI sitting on top of physical models, where the values built into the output are easy to…

Read the full review in PlayNext →

Why you'd press play

Pressure-test how a model handles uncertainty and values.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Risky Science Podcast