It's much easier to write software that does something, than it is to write software that does the right thing every time. If an NSA system fails to see 10% of traffic, that's acceptable. If an NHS system incorrectly rejects 10% of claims, that's a disaster.