This thread is for comments on the Human decision principle that will form part of Prostasia Foundation’s draft Best Practice Principles for Sexual Content Moderation and Child Protection. Please consult that master thread for background information.
Here is the current text, to be used as a starting point:
Automated content restriction should be limited to the case of confirmed illegal images as identified by a content hash. Content should not be added to a hash database or blocklist without human review.
Please post your comments and suggestions below.
Here’s how it looks with this suggestion. I’ll push this into draft version 0.2 of the principles:
The only issue I see with this(and I Am Not A Computer Science Expert), can’t you change a file’s hash entirely by modifying a single byte of it, and wouldn’t that make hash verification easy to bypass and therefore relatively pointless?
The innovation behind the PhotoDNA system that is used for scanning CP/CSEM is that it is robust against changes like that. It basically breaks the image (or video, now) down into units that can be identified even after it is rescaled, color-shifted, or subjected to other deformations. So it has proved to be pretty effective.