I do think some tools and practices can be adapted. I think, for example, that expert systems -- whether classical or hybrid a la Watson -- could make a big dent in that $500 figure. That is definitely pursuable.
But humans inject errors, it's what they do best. And outside of a properly formal system (and per Godel, these aren't perfectly usable anyway) the best way to identify errors is ... with another human.
You might say that judges have been performing software inspections for much longer than we have.
Don't get me wrong, law is as slow as frozen honey. But the problem domain is heinously complex and I expect that any software system would reflect that complexity. Given our profession's iffy track record on creating items of massive size and complexity, I lack your transformative enthusiasm.
I am what you might call a cautiously optimistic conservative. Or a pessimistic radical. Not sure. One blogger I host talks about "radical centrism".
The only scope for human error in such a system is the process of coding itself, and massively distributing this and opening that process as much as possible makes any bugs in the system shallow by comparison, as well as easily demonstrable by test cases rather than actually requiring some clueless politician, lawyer or judge to make a terrible decision and someone else to suffer the consequences for it.