Churchland on folk psychology and connectionism

His viewpoint on folk psychology seems fair enough. The basic argument put forth by Churchland on why we should not rely on folk psychology is that ideas in folk psychology are vague enough that they can be manipulated to justify any kind of description. Supporters of folk psychology can claim that their viewpoint are meant as an abstract description and not to be taken literally. Thus, if folk psychology can be manipulated to fit multiple different theories then having folk psychology doesn't serve any useful purpose.

Churchland in the same article talks about connectionism and how that is one of the more modern theories that can helop us understand how the brain functions. Churchland fully admits that just because we can create neural networks containing a few hundred nodes it does not mean that the brain or parts of brain function like that. Brain is innumerably more complex than the current neural networks that can be run on modern computers. Furthermore, some of the ways the neural networks function, such as back propagation to solve problems, is physically implausible biologically. Back propagation is implemented by our brains but not in the same way as neural networks function. These criticisms all seem fair.

But I personally feel that churchland fails to express why exactly are neural networks helpful in understanding how the brain functions. The solutions that are generated by neural networks can be different everytime we run it to come to a predefined solution, then what are we to gain from it?