Preventing an AI catastrophe: a brief intro to analysis by synthesis

An essay argues that AI safety debates conflate alignment with basic epistemics, using the 1983 Petrov nuclear false-alarm incident to show that the Soviet satellite was not misaligned but simply lacked a world model. It revives Ulf Grenander's Pattern Theory, which treats pattern analysis and pattern synthesis as sharing one probabilistic model, and contrasts it with the backpropagation approach that displaced it.