Gary Marcus

Strawman alert: To extent @tdietterich directed this line at me "I object to the argument that says “Today’s deep learning-based systems don’t exhibit genuine understanding, and therefore deep learning should be abandoned" it is an absolute misrepresentation of my position. 1/2https://twitter.com/tdietterich/status/1188551504388452352 …

My *actual* position is NOT that we should abandon deep learning, nor that deep learning can't get us to shallow understanding in sense @tdietterich has described, but that deep learning *alone* can't get us to deep understanding. Chp 4 of http://rebooting.ai  explains 2/2

Lack of shared definition makes this discussion a never-ending debate. If you define "deep understanding" as "human-level understanding", I guess that all AI algos (connectionist/symbolic/hybrid) will fall short of reaching this goal except if they model human brain mechanisms

Lack of reading hurts a lot too; we gave many examples in Chapter 4 of http://Rebooting.AI ; not one of my critics has actually gone through those examples and given a counteranalysis.

Please, don't take me wrong. I found your examples very informative in your inspirational book. My point was just that we need to separate the different meanings of the word "understanding" to foster progress on this topic. Not an easy job, but a needed step in my opinion

i am fine (as said elsewhere) distinguishing shallow b deep understanding; chapter 4 was about why text processing isn’t deep enough understanding

my larger point (not directed to you) is that this whole discussion would improve if framed around an analysis of whether chapter 4 is accurate, rather than soundbites.