Metacognition of Concepts

Deep neural networks – how do they think?

Deep neural networks – how do they think?

Deep neural networks perform astoundingly well on a whole range of difficult tasks. But what is going on inside? Are their computations anything like human reasoning? This paper shows that there are in fact two quite different ways that their internal processing works. This new distinction at the level of computational processes allows us to see more clearly how deep networks have differed from human deliberative reasoning, and also how the newer models have moved closer to the structures underpinning human general intelligence.

Shea, N. (2021), ‘Moving Beyond Content-Specific Computation in Artificial Neural Networks’, Mind & Language[open access pdf] [link]