Deferring decisions to an algorithm doesn't absolve the owner from responsibility for its actions. If the consequences are unknown, why should it be allowed to use it?

For things that are crimes only if the intent is there, it certainly does absolve the owner from responsibility. There's no duty to preserve everything that might be evidence in some manner (because, really, everything might be) - if evidence gets destroyed as a byproduct of normal operations, that's not prohibited.

Because if the consequences are unknown, there is no intent, and intent is necessary for complicity.

If Google/people at Google made an algorithm to intentionally delete criminal evidence, that would qualify. Having an algorithm that deletes lots of things, and happens to delete evidence does not.

Couldn't that also imply that someone else would need to view the algorithm to verify it's intent?

The parent post mentions they've been reviewed by a human (appeals process) and been rejected anyway.

That should get past any "it woz the algorithm that did it!" arguments about intent.

If I'd write an algorithm for a self-driving car that just drives it in a straight line with high speed.. I'd be still responsible if the car kills somebody and couldn't just say that "whoops it was just the silly algorithm" even though I didn't specifically intend to kill somebody.