Why Neurons Cannot be Detectors: Shifting Paradigms from Sherlock Holmes to Elvis Presley?


The practice of treating neurons as detectors is ubiquitous in the neuro-science community and in AI as well, in the context of neural networks. But there are a growing number of cognitive scientists who think that the representational paradigm is ill-suited to this level of explanation. In this paper, I rehearse William Ramsey‘s powerful critique of neural-detector attribution, focusing on his argument that Dretske-style information theoretic accounts of representation fail to justify the practice. I then take this conclusion a step further by arguing that not only does this particular justification fail, none at all are possible. The conclusion that we need to let go of the representational paradigm is not a negative one though, I shall claim, because it liberates us from the kind of misguided thinking that leads to theoretical dead-ends. Once we see this, we are free to investigate new, more fruitful, paradigms.

Back to Table of Contents