Shenbo Xu: Your AI Is Confident. That Doesn't Mean It's Right | Ep 1315

In this episode of The Digital Executive Podcast, host Brian Thomas sits down with Shenbo Xu, MIT PhD and co-founder of Kapnova, to make a distinction that most business leaders have heard but rarely internalized — and that costs companies millions when they get it wrong. Shenbo opens by tracing his path from MIT's AI lab through quantitative research at Point72 and foundational model training at Scale AI, explaining that each stop sharpened the same core insight: prediction tells you what will happen, but causal inference tells you what happens if you actually do something about it. Those are fundamentally different questions, and almost every AI system in deployment today is only built to answer the first one. He illustrates the danger with two vivid examples. First, a retail CEO who copied Apple's no-discount strategy without understanding that Apple's customers and his own were nothing alike — the sales crashed, and he lost his job. Second, eBay's discovery that years of spending on branded search ads — the ones triggered when someone literally types "eBay" into Google — produced almost no incremental value, because those customers were already going to the site anyway. The model said the ads were working. The experiment proved otherwise. This, Shenbo argues, is the gap that lives inside almost every ROAS number ever reported: prediction models get credit for outcomes that would have happened regardless. Causal AI builds an actual model of what drives what, so executives can ask what will happen if we do this before committing the budget. He closes with a forward-looking take on where causal reasoning is heading — pointing to recent DeepMind research suggesting that any AI capable of generalizing across new environments must have learned something causal — and argues that the next generation of decision-making tools will be built on mechanism, not just pattern, making them more honest, more durable, and genuinely trustworthy when the world changes.

If you liked what you heard today, please leave us a review - Apple or Spotify. 

Learn more about your ad choices. Visit megaphone.fm/adchoices