Statistics, causality, causal modelling, mediation analysis, counterfactuals, philosophy of research, causal representation learning, ethical AI, the power of abstractions, and systems thinking

Excellent review. One thread to pull on: active inference is in a sense an extension to Pearl causal inference that is more practically applicable to complex systems, as the agent's probing of (internal and external) environment and comparing with its assumed generative model is included explicitly as part of the theory. Some (myself included) are even working on higher-order model testing and learning in active inference, using exactly the concept of abstract frames a la Hawkins.

“Bayesian statistics is the only branch of statistics that also allows for subjectivity (prior beliefs are subjective)”.

‘Prior beliefs are subjective’. Surely that is not quite so – perhaps to refer to them as beliefs is also misleading. The predictive value of a medical test. like an x-ray for breast cancer. produces an estimate of the risk of that individual having cancer: however, that estimate is entirely dependent on the frequency of occurrence of the condition in the population in question. Is that not a prior probability? It may be one which is generally subject to scientific estimation, but nevertheless it is a prior probability, and it is not a ‘belief’.

Excellent review. One thread to pull on: active inference is in a sense an extension to Pearl causal inference that is more practically applicable to complex systems, as the agent's probing of (internal and external) environment and comparing with its assumed generative model is included explicitly as part of the theory. Some (myself included) are even working on higher-order model testing and learning in active inference, using exactly the concept of abstract frames a la Hawkins.

“Bayesian statistics is the only branch of statistics that also allows for subjectivity (prior beliefs are subjective)”.

‘Prior beliefs are subjective’. Surely that is not quite so – perhaps to refer to them as beliefs is also misleading. The predictive value of a medical test. like an x-ray for breast cancer. produces an estimate of the risk of that individual having cancer: however, that estimate is entirely dependent on the frequency of occurrence of the condition in the population in question. Is that not a prior probability? It may be one which is generally subject to scientific estimation, but nevertheless it is a prior probability, and it is not a ‘belief’.