Taking effective results in machine learning, and somehow assuming that they apply to cognition - simply because neural nets were inspired by our limited knowledge of neural signaling and structure - is like trying to apply aircraft engineering to studying ornithology. For a better articulation of this point (from the reverse direction) check out the paper 'Could a Neuroscientist Understand a Microprocessor?' from 2017 - https://journals.plos.org/ploscompbiol/article?id=10.1371/jo...
