# Surfing Uncertainty: Prediction, Action, and the Embodied Mind  ## Metadata - Author: [[Andy Clark]] - Full Title: Surfing Uncertainty: Prediction, Action, and the Embodied Mind - Category: #books - Summary: insert summary - My notes: - Summary: The brain acts as a prediction machine, using past experiences to make sense of new sensory information. It constantly adjusts its understanding to minimize errors in these predictions. This process helps us perceive the world more accurately, but it also means our perceptions can depend heavily on prior knowledge. - Source File: Surfing Uncertainty_ Prediction, Action, and the Embodied Mind by Andy Clark - Oxford University Press (2019).pdf ## LLM Chats ## NotebookLM ## LLM Audio ## Highlights > The links to imagination and dreaming are then close at hand, for such systems command a generative model capable of reconstructing the sensory signal using knowledge about interacting causes in the world. That process of reconstruction, tuned and deployed in the presence of the sensory signal, paves the way for processes of outright construction, able to form and evolve in the absence of the usual sensory flow. ([View Highlight](https://read.readwise.io/read/01jkzjmxmqwwnre7hxy7ga20dt)) > The illusion occurs because (as we saw in [chapters 1](#acprof-9780190217013-chapter-1) and [2](#acprof-9780190217013-chapter-2)) our visual experiences do not simply reflect the current inputs, but are greatly informed by ‘priors’ (prior beliefs, usually taking the form of nonconscious predictions or expectations) concerning the world. In this case, the prior is that surfaces tend to be equally reflectant rather than becoming gradually brighter or darker towards their own edges. The brain’s best guess is thus that the central pairing involves two differently reflective surfaces (two different shades of grey) illuminated by differing amounts of light. The illusion occurs because the image displays a highly atypical combination of illuminance and reflectance properties and the brain uses what it has learnt about typical patterns of illumination and reflectance to infer (falsely in this case) that the two tiles must be different shades of grey. ([View Highlight](https://read.readwise.io/read/01jkzj5ax3vgvn5gaq9gydg666)) > An important feature of the internal models that power such approaches is that they are *generative* in nature. That is to say, the knowledge (model) encoded at an upper layer[4](#acprof-9780190217013-miscMatter-9-note-61) must be such as to render activity in that layer capable of predicting the response profile at the layer below. That means that the model at layer N + 1 becomes capable, when operating within the context of the larger system, of generating the sensory data (i.e., the input as it would there be represented) at layer N (the layer below) for itself. Since this story applies all the way down to layers that are attempting to predict activity in early processing areas, that means that such systems are fully capable of generating ‘virtual’ versions of the sensory data for themselves. ([View Highlight](https://read.readwise.io/read/01jkzj1fws5gtdmjdebsyzn625))