Combining causality with algorithmic information theory — LessWrong