Focus

Even just focusing on prediction errors is still too noisy. If every time our models were wrong we had to focus on figuring out what was wrong and how to update our model, we’d be completely stuck in analysis paralysis!

Because our brains are laser focused on meeting our needs, we only really need to focus on the prediction errors that are relevant to our needs. Neuroscientists call this “attention” – effectively a prioritization of errors so we can get our models right where they matter most, and tune the rest out. We better get that “breathing” thing right before we start worrying about whether we understand Kant correctly.

Sometimes, this is incredibly narrow — if we are dying of thirst, the only real need that matters is water, and the only surprises that matter are ones that help us build a better model for how to find water.

As we feel more “safe”, we have the luxury of attending to surprises things that may be helpful at some point, but aren’t urgently needed right now.

Solution: Attention allows us to zoom in and out on what’s directly relevant to us, and tune out signals that would otherwise be distracting.