What I love about this post is that, at heart, it challenges the very ability to be rational, it hints at the possibility that what is most effective (and how we ultimately function) is as statistical learning machines whose understanding and predictive ability lie in structures very different from the ones we use to communicate. Our language and formal argument structures define a space of understanding that is more a reflection of our communication technology than the true nature of things. In this way, Occams razor and Kolmogorov complexity reflect a motivation to communicate efficiently (both with ourself and with others). It can have practical utility in the sense that a simpler conscious model is easier to analyse and in that sense we have more chance of finding its flaws (predictive failures). In this sense a simpler theory (in the sense of being more comprehensible) is more likely to be correct if it explains the same evidence, as it literally has more analysis performed on it, because it is easier to think about. Or in an AI sense, has a larger (mentally synthesised) evaluation set.
What I love about this post is that, at heart, it challenges the very ability to be rational, it hints at the possibility that what is most effective (and how we ultimately function) is as statistical learning machines whose understanding and predictive ability lie in structures very different from the ones we use to communicate. Our language and formal argument structures define a space of understanding that is more a reflection of our communication technology than the true nature of things. In this way, Occams razor and Kolmogorov complexity reflect a motivation to communicate efficiently (both with ourself and with others). It can have practical utility in the sense that a simpler conscious model is easier to analyse and in that sense we have more chance of finding its flaws (predictive failures). In this sense a simpler theory (in the sense of being more comprehensible) is more likely to be correct if it explains the same evidence, as it literally has more analysis performed on it, because it is easier to think about. Or in an AI sense, has a larger (mentally synthesised) evaluation set.