Gary Marcus

Everything should be a pure neural network processing, where search is done via some smart attention-based process. This is how--Yoshua Bengio argues--human think. Well, for one thing, we don't know how this happens. For another, a lot of automation/mechanization processes...

Do not try to mimic how animals/people think or move. One could insist that cars may need to walk and jump (and for rough terrain it's a must), but in most cases we can build roads and put cars on wheels. For many years to come this is likely gonna be an optimal solution.

I would tend to agree with that spirit. You want symbol manipulation but these symbols need to emerge as attractors in highly recurrent neural net. For one, because we have no idea what these symbols should represent at the lower level of cognition.

But that is almost entirely perpendicular to a question of whether hybrid GOAI and deep learning could bring useful applications. I think the answer to that is yes, but at the same time it is not a road towards solving AI.

Well these hybrids are heavily used already.