Perhaps it seems even more desirable to eliminate all macroscopic entities except those needed to fix inputs, and read outputs. That is, perhaps it would be better if the control variables and the mechanisms which manipulate them were not macroscopic slits, but themselves quantum phenomena.
Perhaps not. Penrose makes some interesting comments [Penrose, 1989, p 403, 171-2,] concerning the physics of computation that are relevant here. He argues that we can only have computers built out of macroscopic objects because of the discreteness of the quantum level; if there were no underlying discreteness, then there would be an unacceptable degradation of accuracy within any computational system. Furthermore, at least part of this discreteness is provided by the collapse of the superposed wave packet, so a purely quantum computer which does not have its superpositonal states collapsed now and again by macro objects, may be less powerful than a hybrid classical/quantum one.
On the other hand, this limitation of purely quantum systems may only be an impediment to traditional, von Neumann style computation. Neural networks, in that they are more robust and noise-tolerant, do not require as high a degree of accuracy, and thus quantum implementations of them may be able to function adequately without frequent ``observations'', which collapse the superpositional states.