1. Depriving the the public of decent encryption does a lot of harm. If you let some people in through the backdoor you're going to let bad people in too.
This is different from nuclear bombs. It doesn't harm the public to deprive them of nuclear weapons. I like having my financial information encrypted, but I don't want a personal nuclear bomb.
2. Non-proliferation of encryption wouldn't work. The algorithms and ideas are already widespread. Every university teaches it. Millions of computers have the code already. The code and ideas can be proliferated anonymously, instantly, undetectably and cheaply.
This is different from nuclear bombs. Nuclear bombs are much more complex and expensive to transport, manufacture and hide.
(I haven't mentioned AK47s, but I'd say they fall somewhere between between encryption and nuclear bombs in benefit vs harm to the public and in ease vs difficulty of non-proliferation, which is why there are more of them around than there are nuclear bombs.)
Back to HN's world, one of the basic principles of human factors is "make the right thing easy and the wrong thing hard." You might not be able to prevent people from doing the wrong thing, but there is still value in making it harder to do.
All you need to write an encryption program is an explanation of the concept behind a single algorithm, someone with math smarts, and almost any computer.
But honestly, I think it'll result in bad guys using the same backdoored crypto that everyone else will be using, relying on steganography and disappearing in the noise - so basically, they'll do things in the same way as they always did, since assuming all communications is broken by the government is simply good OPSEC.
Information about the uncrackable padlock is widely available. Bad guys will always be able to get these padlocks from the black market. The question is whether we want the good guys (i.e. us) to have the padlocks too.