I think I can describe the ‘slippery slope’ to an alien and I think this description offers an obvious (if difficult) solution.
The ‘slippery slope’ is what we call a situation in which good judgement and a certain algorithmic response to something correspond in one case, but not in others, such that following that algorithm in similar cases will cause us to diverge from our good judgement at some point. ‘Slippery slopes’ come up when our good judgement about something is partially but not entirely captured by a particular algorithm.
Our solution to slippery slopes in legal settings is the institution of the judiciary, who are supposed to make sure a law is applied in such a way that is consistant with good judgement rather than with the letter of the law, wherever these conflict. This is how we approach slippery slope problems generally, and it works so long as those in charge of applying algorithms have good judgement, which is common enough, if imperfect and difficult to teach.
Schelling fences are patches or qualifications on algorithms which attempt a better approximation of our good judgement, and while they’re useful, they’ll never be ultimately sufficient unless there actually is an algorithm which just is good judgement in general or in some specific sphere.
I think I can describe the ‘slippery slope’ to an alien and I think this description offers an obvious (if difficult) solution.
The ‘slippery slope’ is what we call a situation in which good judgement and a certain algorithmic response to something correspond in one case, but not in others, such that following that algorithm in similar cases will cause us to diverge from our good judgement at some point. ‘Slippery slopes’ come up when our good judgement about something is partially but not entirely captured by a particular algorithm.
Our solution to slippery slopes in legal settings is the institution of the judiciary, who are supposed to make sure a law is applied in such a way that is consistant with good judgement rather than with the letter of the law, wherever these conflict. This is how we approach slippery slope problems generally, and it works so long as those in charge of applying algorithms have good judgement, which is common enough, if imperfect and difficult to teach.
Schelling fences are patches or qualifications on algorithms which attempt a better approximation of our good judgement, and while they’re useful, they’ll never be ultimately sufficient unless there actually is an algorithm which just is good judgement in general or in some specific sphere.