SoylentNews

takyon [soylentnews.org] writes:

A new social network, Candid [becandid.com], will use machine learning to try and moderate posts [npr.org]:

We use a deep learning NLP algorithm, which basically looks at what you're saying and decides ... whether it's positive or negative. So it kind of classifies things as having a negative sentiment or a positive sentiment. It then gives it a score of how kind of strong your statement is — let's say you said something about someone or you threatened someone, it classifies that as saying, "Hey this is a very strong statement," because these kinds of categories are not good in terms of social discourse. And when we do that, we basically say if this thing has a score which is more than a particular level, a cut-off, then we basically take out the whole post. So whether it's self harm or like bullying or harassment, we look for certain phrases and the context of those phrases.

On the line between moderation and censorship

I mean, here is the thing between what is "loud free speech," quote-unquote, right? At some level you should be able to say what you want to say, but on another level, you also want to facilitate, you know, what I would say constructive social discussion. ... There is a kind of a trade-off or a fine line that you need to walk, because if you let everything in, you know the fear is that social discussion stops and it just becomes a name-calling game. And that's what happens if you just leave — like certain discussions, just let them be, don't pull things down — you will see they quickly devolve into people calling each other names and not having any kind of constructive conversations.

They've succeeded in getting some free press, if nothing else.

Original Submission