I liked this post. It correctly imho emphasizes theoretical & mathematically substantive approaches, a ‘big tent’ approach and ties it in with the kind of conceptual questions we’d like to attack in AI alignment.
You might like to look at Crutchfield’s epsilon machines, see e.g. https://warwick.ac.uk/fac/cross_fac/complexity/study/msc_and_phd/co907/resources/dpf.warwick.slides.pdf
Especially with regard to a good definition of abstraction I suggest taking a look at this paper: https://arxiv.org/pdf/cond-mat/0303625.pdf
I liked this post. It correctly imho emphasizes theoretical & mathematically substantive approaches, a ‘big tent’ approach and ties it in with the kind of conceptual questions we’d like to attack in AI alignment.
You might like to look at Crutchfield’s epsilon machines, see e.g. https://warwick.ac.uk/fac/cross_fac/complexity/study/msc_and_phd/co907/resources/dpf.warwick.slides.pdf
Especially with regard to a good definition of abstraction I suggest taking a look at this paper: https://arxiv.org/pdf/cond-mat/0303625.pdf