[snrji]

People tend to do a hard distinction between symbolic AI and machine learning, but actually some machine learning algorithms are based on symbols (eg. decision trees and association rules build logical rules). I recommend Pedro Domingos book, The master algorithm, in which he describes the "5 machine learning tribes" (one of them is referred as "the symbolists") and advocates for a unification of different machine learning algorithms. He even proposes a particular instance of algorithm that would fulfill these criteria: Markov logic networks. He has developed an implementation, called Alchemy (https://alchemy.cs.washington.edu/).

If by symbolic AI we mean GOFAI, expert systems etc, I don't think that there will be ever a resurgence. But if by symbolic AI we mean machine learning algorithms that are somehow based on symbolic reasoning, I do think that there will be a resurgence. In particular, this resurgence will start when: a) Deep learning arrives to its limit (ie. research gets stuck) and/or b) Someone finds a scalable and SOTA-ish way to integrate symbols into gradient based algorithms.