With that said, I think there's two major reasons for bold "save the world" visions expressed by startups to lack viability. The first is that the symbiosis between startups and venture capital creates a cycle where the most successful players are inevitably end user-facing services optimized for rapid growth. There are a ton of problems whose solutions do not find under this model. In fact, being too big of a visionary is usually considered a negative: it likely means you have no product-market fit. The incentive henceforth becomes to achieve a maximum market valuation and preferably a low fair value (less effort in creation, ops and maintenance).
The second is what jerf terms the "BOAC fallacy" (http://www.jerf.org/iri/post/2916). Lots of programmers seem to have this odd magical thinking that they can trivially obsolete industries and professions by just applying statistical learning over some data sets, or by writing what amounts to a spreadsheet... except marketed and branded towards a given field. Doesn't change the fact that it's just a spreadsheet application, and that there are other intricacies that naturally impede automation of a certain area which you haven't accounted for.