The advantage is that unlike the lateral inhibition, it does not require:

• a separate hard-wired receptive field for each neuron in the ring connecting to all other neurons in that ring with a specific pattern of excitatory and inhibitory connections

• and a separate set of those receptive fields for each kind of completion that the system performs, i.e. one for collinear, one for orthogonal, and one for diagonal completion, etc. Each of these sets of receptive fields would need their own set of neurons

• and those sets of neurons would have to be connected to each other with the proper excitatory / inhibitory connections to account for the complex interactions between different completion modes.

In other words, the harmonic resonance model offers a very simple mechanism with remarkably complex computational properties, which would require a combinatorially implausible set of neurons if implemented in an equivalent neural network model. This is the real power of the harmonic resonance principle, and I am amazed that this reviewer does not have the vision to recognize that fact.