Nice talk, here are few comments/observations;<p>Kant: He is correct that AGI is a philosophy problem (more specifically epistemology). Unfortunately, his hero is why he will never solve it unless he rejects Kant's false theory of consciousness (i.e. consciousness creates reality) and concepts.<p>Model: At 8:50 he has a slide showing the model of the AGI receiving inputs from the environment and output of actions on the environment. At the most fundamental level, the flow diagram should have arrows on both ends to incorporate the most general form of subject/object, observer/entity relations. Without those arrows he is implicitly trying to solve the "narrow" AI problem he claims to want to transcend.<p>Data Overload: At 14:30 he says one of his goals is to solve the data overload and how to decide what to do in the face of that and, more generally, system complexity. The epistemological solution is to identify the <i>principles</i> that are valid in a given context but principles are just a type of concept, hence the need for a (non-Kantian) theory of concepts.<p>Ethics: To me this is the best implication of AI research; forcing scientist to confront the false premise that there can be no scientific basis for ethics. And by ethics I don't mean ethics for robots or AI; I mean what are the principles of ethics for human beings. This problem must be solved before you can apply it to AI. In my judgement, the first principle is that all humans should go by reason but proving that and defining what you mean by reason and how it works is mired in non-sense from Kant and Hume. Step one is a clear theory of concepts.<p>FTFY: He joined with Google and put their motto on a slide which I have revised for clarity "To organize the world's information and make it universally accessible and useful (for Google marketing)" ;-) A word of caution that their goals may be different than his.