Because ad-hoc tests, tests, tests, ... with the verbosity they add (tons of uneeded tests that could be type checked instead), and the arbitrariness (you have to remember to add a test -- instead type checks are always run) and checking only some code paths (type checks propagate everywhere) are in any way better?
Not to mention that you can have type checks AND (less, complimentary) tests at the same time.
At least us software guys have an advantage: the hardware guys will always have to live with failure rates.
But the standard happily declares software to be bug-free, as long as you do the required testing and stuff. ;-)
But aren't those tools (that can build trust) something that could be built-in & available to all users of the language, therefore avoiding "reinventing the wheel"? Why would you want to do by hand something that can be automated?
Fact is, if you can't trust your language because of its type system, you develop tools to build that trust around your particular application of the language. If that is 'too much work for you' .. then by all means, avoid the work by reinventing the wheel. In Academia, you can get away with that because you don't have customers, and you don't have an industry demanding results. At most you have peers whom you have convince that steering their groupthink in your direction is worth the effort.