In a previous post, I argued in favor of splitting robot learning into task-independent and task-dependent portions. The general version of this statement is certainly not a new idea. Indeed, my own thoughts on this topic are shaped by years of discussions with my advisor and there is a whole community of people, including some of my current colleagues, who’ve done work in this direction. However, I do believe that I have a new take on the specifics of this argument – more on that later, in my formal publications.
Meanwhile, in the previous post, I had waved my hands in the direction of the claim that kids seem to benefit from precisely this separation. Specifically, the randomness of childhood behaviours are aimed at acquiring powerful representations and foundational concepts in aid of future task-specific strategies and concepts. This op-ed piece in the New York Times does a better job of arguing this point. The author of that piece has also recently written a book entitled The Philosophical Baby, on the same topic. I look forward to reading it when I get some time. This article in the New Scientist magazine makes a similar conjecture about the role played by teenagers in human development.