Maybe because the terminology is incredibly misleading? The whole point of representation is that behavioral invariance is thanks to symbols in the brain, ie, reliable proxies. What the experiments show is precisely that system invariance is not due to component invariance (=symbols).