This week’s sciku is inspired by a new hypothesis about the purpose of dreaming. No one really knows why we do it, but dreaming must have some beneficial function. Though it’s only 2% of our body mass, our brains consume 20% of our energy. If we just turned them off at night, we’d save 50-100 calories, and that’s a large price for any animal scrambling for food as our ancestors have throughout evolutionary history. If it wasn’t worth the resources, we wouldn’t do it. And yet we do.
The new explanation, proposed by a neuroscientist at Tufts University, was inspired by a problem that arises in machine learning. If you give AI a task to learn, like navigating a car from one point to another, it does a much better job of solving the problem specifically, rather than generalizing what it learns to other environments. This “overfitting” has always been a challenge with AI, and is often solved by adding extra noise into the system to prevent the AI from settling onto a precise solution.
It’s only a speculative idea for now, but maybe that’s the purpose of dreaming. We learn about our environment during the day, but then play with what we learned mentally at night, so that our concepts of what we’ve experienced don’t become too rigid to apply to future situations.
I was re-reading Iain Mcgilchrist’s excellent book “The Master and His Emissary” the other day, and I can’t help but think the bicameral nature of the brain might be playing role here, too, as the right hemisphere is the “generalizer,” harvesting information from the left’s narrow and concrete view of the world to make broader associations it can assimilate into a bigger picture that the left hemisphere is never able to see.
This is the stuff of poetry and all art, of course—the associative, inarticulate understandings of the right brain. The stuff of dreams. So this new idea is interesting to me, though whether it can ever be confirmed remains to be seen.
satellites skimming the surface