Astral Codex Ten

active inference

4 min read

active inference

Article

active inference is a recurring concept in the Astral Codex Ten archive, appearing 3 times across 3 issues between June 15, 2021 and June 14, 2023. The archive places it in contexts such as “The theory of “active inference” adds another layer of complexity here”; “The “active inference” hypothesis - a leading theories of how brain motor centers work”; “in active inference, a step is just a fixed false belief that your leg has been placed in front of you”. It most often appears alongside 5-HT1A, 5-HT2A, ADHD.

Metadata

  • Category: Concepts
  • Mention count: 3
  • Issue count: 3
  • First seen: June 15, 2021
  • Last seen: June 14, 2023

Appears In

Source Context

Recovered passages from the original issue text. When the raw archive preserved outbound links inside the source passage, they are listed directly under the quote.

Carhart-Harris On Serotonin
June 15, 2021 · Original source
The theory of "active inference" adds another layer of complexity here; it posits that sometimes your brain automatically resolves prediction error through action. If you were expecting to be well-balanced, but actually you're off-balance, you'll reflexively right yourself until you're where you expected to be. At its limit, this theory says that all action takes place through the creation and resolution of prediction errors - I stand up by "predicting" on a neurological level that I will stand up, and then my motor cortex tries to resolve the "error" by making me actually stand.
All of this was in the paper and my review, but I like the way George ties it together with problems of active inference and the adjusting-predictions vs. changing-the-world tradeoff. If true, this should be testable on the very small scale, with predictions around perception and movement.
Information Markets, Decision Markets, Attention Markets, Action Markets
March 28, 2022 · Original source
But my favorite example is from neuroscience. The “active inference” hypothesis - a leading theories of how brain motor centers work - says that the brain is entirely a predictive organ, but that the easiest way to predict your body position is to cause it. The brain registers a prediction of 100% that your arm will move, and then - in order to “win” its “bet” - moves your arm. “Every prediction market is also an action market”, indeed!
Inline links: active inference
The Canal Papers
June 14, 2023 · Original source
We can convert from a model of stimulus-action valleys to a model of Bayesian beliefs. Given that you’ve started moving your leg to walk, you have a high prior (or an “extremely precise belief”) that you should bend your knee a certain way. Since the authors are all good Fristonians, they don’t really distinguish between beliefs and actions; in active inference, a step is just a fixed false belief that your leg has been placed in front of you; its mismatch with reality can only be corrected by actually moving the leg.