Brian Gallagher in Nautilus:
Lucid dreaming is quite peculiar. We become aware that we are dreaming. In normal dreaming, we lack this reflective capacity. Lucid dreamers report that these experiences are extremely vivid, fantastic, and perceptually immersive, like virtual reality. In our new paper, we wanted to explain these differences in a model using the predictive coding framework. The main idea is that the brain is a prediction-generating machine.
Say I see someone in a dream. She’s probably my sister. No, she’s my girlfriend. No, she’s my mother. My brain is trying to make the best guesses of these images. And there is no constraint, no bottom-up input coming from the external world to fit or to shape these predictions. So the brain is just jumping from one prediction to the other. What we argue is that, in lucid dreaming, this is different. I see someone that speaks, let’s say, in a language that is different from the language that I know she usually speaks. This creates a prediction error. And I’m not changing the identity of the person. Instead, I realize, “Okay, something is not going on correctly here.” This is a momentum for lucid dreaming, this prediction error, that will trigger the insight that I’m in a dream. We call this a superordinate self model: “I am dreaming. I’m lying in bed. But I’m having a dream and I’m having these ideas.” This will create a top-down model to which everything that is strange and surprising will be easy to accommodate.
More here.