> Our conceptualization of the problem domain is fundamentally insufficient. There have been people in the AI community saying this since at least the 80s/90s

> Our conceptualization of the problem domain is fundamentally insufficient. There have been people in the AI community saying this since at least the 80s/90s (e.g. Hofstadter). It's an old idea that has been difficult to get much traction on, partially because it's a long way from applications. NN, SVN, etc. for all of their limitations can draw that line pretty easily.