Simple "neutral" algorithms are generally easy to game. If you choose an algorithm that you already know will be gamed for harassment or whatever then you're essentially choosing that outcome.
Re:Gitlab: don't discuss politics at work
by ahnick on Wednesday October 16, 2019 @06:46PM
(#21276838)
The labeling algorithm is just a relatively naive example. More complex "neutral" algorithms can be designed. Mitigations for abuse that are "neutral" themselves may be added before/after initial implementation. (e.g To extend the labeling example, a moderation system could be added that allows users to flag inappropriate labels.) Also, I don't think it is always possible to foresee in what ways a system/algorithm might be abused; therefore, it might be necessary to address said abuse once it reveals itself.
Let's say for example you have a community contributed labeling system that allows users to label content. A recommendation algorithm might then be written that simply pulls the top 10 highest number of viewed pieces content that have the same labels as a piece of content the viewer just viewed. In this case, there is nothing in the algorithm that evaluates the content itself. The evaluation was done by users of the software, not the software itself.