Artificial intelligence has not found its newton, we cannot rely on it

“Artificial intelligence has not found its newton, we cannot rely on it” — Gary Marcus

2019 is almost over. Looking back on this year, AI and deep learning are undoubtedly hot topics, but there are also many questions about AI. Due to different views on the development of deep learning, Gary Marcus, a professor at New York University and a former Uber AI Lab director, questioned the 2018 Turing Award winner Yoshua Bengio’s position on deep learning battle.

Five years after the rise of Deep learning, it’s time to critically reflect on the capabilities and deficiencies of deep learning. Is deep learning overhyped, or is it just the throbbing of new things replacing old ones? To what extent has artificial intelligence developed?

Gary Marcus has published a long article critically discussing the current state and limitations of deep learning. The article points out 10 challenges facing deep learning:

1、Deep learning currently requires a lot of data.

2、Deep learning is currently too superficial and does not have enough capacity for migration.

3、Deep learning has no natural way to deal with hierarchical architectures so far.