> the robot has a model of itself and the world around it, and when it "looks at the world" it's actually looking at that model. So do you! Perception is a bay

> the robot has a model of itself and the world around it, and when it "looks at the world" it's actually looking at that model. So do you! Perception is a bayesian inference problem! Maybe the actual state of the world is x, but you observe the latent z (eg photoreceptor output) which is related to x in some complicated way. What you are doing when you look at a scene and understand what's going on is inferring what underlying state of the world is most likely to have generated your observation! Theoretical neuro gets one surprisingly far in terms of gaining insight about the self. For instance, I'm a big sucker for the sampling hypothesis, which says basically that thinking is the process of drawing samples from the posterior p(x | z), states of the world which are likely to have generated your observation.