Re:Deep Learning’s Impact on Image Processing, Mathematics, ...
by xamuel on Sunday May 07, 2017 @08:37AM
(#14285140)
>An algorithm that explains ... isn't any more moral
Not sure what "moral" means, but two things:
1. If black boxes make unexplainable decisions, that gives too much power to the person who reads the box's answer,
2. If the child services algorithm explained its prediction involved e.g. alcoholism, then the parents could try to quit.
One example is in the child services. We've got enough digital historical data from child services, that we're able to build models, that let's us predict whether or not a child will have to be removed from abusive parents 5-10 years from now.
This information is relatively useless though, because there is no way to explain how it works in a sense that the general public will accept.
The public is probably going to adjust eventually, but until that happens, you'll see an increasing push for documentation on how a result was reached.
Ironically an algorithm that explains how it ended up with the results that it did, isn't any more moral than an algorithm that didn't. I mean, you're losing you autonomy and freedom when the government knows what you'll do before you do either way.