Commments on ms. #99-082: "Harmonic resonance theory: An alternative to the "neuron doctrine" paradigm of neurocomputation to address gestalt properties of perception," by Steven Lehar
This is a very ambitious but ultimately unsatisfactory paper. It addresses one of the most important unsolved problems in perceptual theory: the mechanism by which certain holistic (gestalt) properties of perception arise from neural activity. The author proposes the (somewhat) novel metaphor of some sort of neural harmonic resonance to replace the "neuron doctrine" that dominates current theories of neural computation. (I say that this metaphor is "somewhat novel" because Gibson used a mechanical resonance metaphor for his theory of information pick-up many years ago.)
The big problem with the present paper is that the author never gets beyond the metaphor stage. Yes, harmonic resonance and standing waves on vibrating plates appear to bear some interesting relations to Gestalt phenomena of perception, but this observation leads to no useful theoretical work. There are some very sketchy ideas about how neural activity might support resonance phenomena, but even these are mostly handwaving. No percptual phenomena are explained and no predictions are derived or tested.
This failure to go beyond the simple harmonic metaphor is particularly unsatisfying because there are dynamic neural network theories that exhibit several of the holistic properties that the author claims cannot be explained within the "neuron doctrine" approach, including multistability, emergence, and filling in (e.g., see the PDP volumes, 1986). I presume that Lehar would include such neural network theories within the general umbrella of "the neuron doctrine," although there are some odd sentences in the very beginning of the paper that might be construed as indicating that the author intends to constrain the Neuron doctrine to feedforward architectures. If so, then he is setting up a straw man, because nobody really believes in strict feedforward models of vision -- or anything else the brain does.
For a paper on this topic to be acceptable for publication in Psychological Review, it would have to fill in a lot of the details that are so obviously missing here. Important phenomena would have to be explained in a concrete, detailed way, predictions of new effects would have to be derived and tested, and the results would have to be compared to alternative theoretical approaches. Unfortunately, none of this crucial work has been reported in the manuscript.