Decision Management Community

A right-to-the-ground discussion about Machine Learning (ML) is happening on LinkedIn. After François Piednoel de Normandie explained why “ML can neither be safe nor secure“, Philippe Kahn wrote: “First off, you’re right: ML isn’t perfect (yet). But neither is my coffee maker, and yet I still trust it not to flood my kitchen—most days. What keeps both from going rogue? Guardrails! In ML, these aren’t just buzzwords; they’re built with everything from rule-based checks to AI agents that monitor, validate, and correct outputs before they reach the wild. You can think of them as the seatbelts and airbags of the AI world.

And about those “very large datasets”—there’s real wisdom there. Big data helps models generalize, spot edge cases, and avoid overfitting. It’s like training a chef with every recipe in the world, not just the ones from their mom’s cookbook. Sure, sometimes the soufflé still collapses, but with robust safeguards, you’re much less likely to serve raw eggs. There are frameworks, and protocols designed to catch anomalies—like the ML-On-Rails protocol, which flags weird inputs before they cause trouble (so your robot doesn’t recommend yoga poses during a sensor meltdown). And let’s not forget, even human brains make mistakes—sometimes with coffee makers and sometimes with math.” Link