So he claims that there is always going to be more data than we can process. If we want to move towards building a true AI we need to focus on intelligent learning rules instead ?
The general machine that really learn's in the wild, anything and everything is the holy-grail, but its also the 'free lunch'.
For now its better to develop gadgets and hack code that solves explicit problems, I think in time there could well be a wiki-expert, where a generalized machine could call upon the wiki-expert to solve the specific problem at hand, or call 100's of such agents, and the only task is to pick the best.
For now we don't even have the best answer's for the specific problems, let alone tools to manage the general problems.
A robot is just that, specific tasks, but is it human? No, its not.
RP Feynman some 60 years ago said "We can't program a dog, because we don't even know how a dog works". Nothing has changed folks, we still don't understand how the dog works, then only then can we have a robot-dog, and then we can talk about a humanoid robot.