Thread by @tobias_rees

arguably, the comprehension of what intelligence is, of what the world is and how it is composed that emerge from GOFAI and DL are radically different.

I don’t think Yoshua Bengio or @ylecun or @geoffreyhinton are saying every single aspect of GOFAI technology is useless.

Rather, the argue that the philosophical assumptions about intelligence and the world that underly GOFAI are untenable.
On the basis of this argument they reject projects that try to add DL elements to GOFAI systems.
I.e., they offer a philosophical critique and alternative.

While they may appreciate aspects of GOFAI, they vehemently reject the suggestion that they way forward is GOFAI-DL integration.
Integration may be most practical right now. But in terms of research or the search for a way forward, it is not.

Hence, they look for ways in which DL can arrive at generalizations.
Not, I repeat, because they consider GOFAI trash. But because the presuppositions about intelligence and the world GOFAI permits appear to them untenable. They unnecessarily limit research.

Please not I don’t judge: I am merely offering an interpretation of the three pioneers of DL.
It is how I understand them.

Call it an effort at philosophical sense making by way of attending to the underlying assumptions about what intelligence is, about what the world is.

🙏