The thing is we never know whether approving X is a type 1 error or not approving it is a type 2 error. So right now we cautiously make, say, 99.9% of type 2 errors and only 0.1% of type 1 errors. We need to relax the requirements a bit, so that the split is, say, 95%/5%, where the 5% type 1 errors are minor (as in, won't cause a huge catastrophe).