Why current Machine Learning fails ...

I’ve been thinking a lot about how little we genuinely understand about the forces that shape our lives, especially when it comes to health and human connection. The U.S. Surgeon General recently published an eye-opening advisory on what he calls the “epidemic of loneliness and isolation,” and the findings honestly shook me. According to that report (https://www.hhs.gov/surgeongeneral/priorities/connection/index.html), chronic loneliness is more damaging to long-term health than inactivity, obesity, and even excessive drinking. We’ve spent decades obsessing over physical metrics—steps, calories, lab values—while missing something that quietly harms people more than many of the conditions our entire healthcare system is built around. That realization made me stop and question how many other impactful factors we simply don’t measure, don’t notice, or don’t have the language to properly describe.

If we’ve overlooked something so fundamental for so long, it raises a deeper question: what else is happening beneath the surface of human life that our current models—whether scientific or machine learning—just aren’t capturing? This is exactly why I’ve become so focused on world models. Traditional machine learning depends entirely on the data we choose to collect, but that approach only covers the narrow slice of reality we already understand well enough to label. World models are different. They try to learn the underlying structure of the world: the dynamics, the relationships, and the hidden causes we don’t directly observe. Instead of just predicting based on correlations, they attempt to represent how the world actually works.

Non–world models are pattern recognizers.
If something wasn’t in the dataset, they often:

• break
• hallucinate
• or output nonsense

They cannot reason about unseen scenarios because they never learned the underlying dynamics — only examples.

For me, that’s the promise behind Active Models. It’s a commitment to building systems that can reason about the deeper forces shaping health and behavior—even when those forces are subtle, invisible, or overlooked. Loneliness is just one example of something we didn’t see clearly until it was almost too late. If we want AI to help us build a healthier, more connected world, we need models that understand more than the data we already know to look for. We need models that can uncover the hidden structure of life itself.