The state space of a complex piece of software is vastly larger than that of a building, and the desirable states within that space are not continuous, while a building's desirable state space is pretty smooth. It's like the difference between walking along the knife edge of a fractal vs. standing on top of a mostly smooth hill.
Because of that, software simply can't be treated the same as other engineering disciplines, at least not yet.
> The state space of a complex piece of software is vastly larger than that of a building
Is it? Unless we can accurately simulate reality at the atom level and bruteforce building designs against every possible scenario to make sure it doesn't fail catastrophically, it's still all down to prior knowledge/experience, reasonable assumptions, approximations and measurements, just like in software. If anything, software is easier, as with enough efforts (formal methods, etc) you can get to prove the correctness of your software, but you can't really do so with a building.
A large chunk of software vulnerabilities is either due to outright incompetence, legitimate mistakes or cost-cutting. The example of an insecure, black-box backdoorable EC like in this article would have a "building" equivalent of using rebar or concrete of unknown specs and origin and then wondering why the construction collapsed.
Thankfully the liabilities associated with civil engineering means we mostly don't use unknown materials of shady origin and there are multiple layers of review to catch any oversights. The same can be applied to software, and while you're never going to get 100% in either domain, if software was as reliable as buildings (as in you can count the number of major collapses/bugs on one hand in the past few years) it would be a major improvement.
Disagree. First of all, the level of detail an atomic level simulation could provide is simply useless in the grand scheme of things. Classical physics is more than enough for the vast majority of use cases and they can be simulated by computers. Eg, you can programmatically measure the load bearing on any part of a structure. And the more important thing, engineers can usually fall back to safety margins that “linearly scale” the provided confidence in a given case. What I mean by that, is that if a given part has to be this stable, as per the calculations, you can trivially make it 3x wider.
While a while loop in a computer program will eat away at any sort of redundancy you may introduce, and the complexity of computability itself simply leaves behind even mathematics. There is no universal way to prove a non-trivial property of an arbitrary program.
Because of that, software simply can't be treated the same as other engineering disciplines, at least not yet.