We can't change the past. When someone does something wrong, the act of saying "Sorry" doesn't help. Actually feeling sorry doesn't help, either. Saying or feeling sorry can only help as part of a process that _decreases the measure_ of the wrong across the multiverse. We can't change _our_ past, but we can update on its evidence—use the memories and records of it as input to a function that _changes who we are_ in a way that makes us perform better in the future (which is somebody else's past). And we can create timeless incentives: if people _know_ that history (and the court system) has its eyes on them, they might do things differently than they would if they knew no one would ever hold them to account.
The _update_ part is more important than the timeless-incentives part. The first duty is to investigate exactly _what happened_ and _why_. If you can learn the causal graph, you can compute counterfactuals: _if_ this-and-such detail had been different but everything else had been the same, what would have happened instead? If you can compute that if this-and-such detail had been different, then something better would have happened, then you can make advance plans and take advance precautions to make sure the analogous detail takes a more favorable value in analogous future situations.
We can't change the past. When someone does something wrong, the act of saying "Sorry" doesn't help. Actually feeling sorry doesn't help, either. Saying or feeling sorry can only help as part of a process that _decreases the measure_ of the wrong across the multiverse. We can't change _our_ past, but we can update on its evidence—use the memories and records of it as input to a function that _changes who we are_ in a way that makes us perform better in the future (which is somebody else's past). And we can create timeless incentives: if people _know_ that history (and the court system) has its eyes on them, they might do things differently than they would if they knew no one would ever hold them to account.
The _update_ part is more important than the timeless-incentives part. The first duty is to investigate exactly _what happened_ and _why_. If you can learn the causal graph, you can compute counterfactuals: _if_ this-and-such detail had been different but everything else had been the same, what would have happened instead? If you can compute that if this-and-such detail had been different, then something better would have happened, then you can make advance plans and take advance precautions to make sure the analogous detail takes a more favorable value in analogous future situations.
projects / Ultimately_Untrue_Thought.git / commitdiff