Well, the question is, do the specific effects of damage look more like the effects that the "radio receiver" hypothesis would predict, or the ones that the "electronic brain" hypothesis would predict?

There is a big difference between an audio distortion and a semantic distortion. The radio-receiver hypothesis predicts that we can introduce audio distortion, but not that we can make the voice stop talking about vegetables. If we can only get the former sort of effect, then we are messing with a device that didn't understand vegetables in the first place; it did not contain any circuitry whose patterns correlated with facts about vegetables, only with radio and audio signal processing; the knowledge of vegetables was elsewhere. If we can get the latter sort of effect, then the device did have some patterns that had to do with vegetables.