On Mon, Aug 3, 2026 at 10:25 AM Matt Mahoney <[email protected]> wrote:
> ...How do we negotiate the data set? How do we prove causation? > Trees: Negotiate the data set. Forest: "Prove" causation. It's hard to overemphasize how important it is to see the forest. Let's not palaver about the trees. Let's focus on what qualifies a causal model. A causal model of the world starts with two things: 1. The state of the world at t_0 that contains within it some notion of change. This is what Newton did when he incorporated the first and second derivatives into his notion of state. 2. Invariant rules of state transformation from t_n -> t_n+1 If you are missing #1, then you are reduced to kinematics rather than dynamics. If you are missing #2, then all you have are statistics -- and that's true even if you have done a curve fit that has a time parameter. Another critical aspect of the "forest" is that it is naive as to which of the trees affect which other trees. Therefore, the architecture is a dense RNN (i.e., it can provide a pseudo-UTM since the number of trees is unbounded finite). By "dense," I mean simply that all "weights" are unconstrained. And I'm only using the dense RNN as a metaphor for the "no priors" of the "There's a forest," perspective. Once we have settled that, we can palaver a bit about the fact that the laws of motion are reversible, etc., but this is pedantry when we are dealing with neighbors arming themselves to kill each other over who gets to tell everyone else to abide by "narrative" of social causation. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T5b58bcc51c493d41-M506f0930ab2142a9e222d8b8 Delivery options: https://agi.topicbox.com/groups/agi/subscription
