Online content moderation
I saw a panel discussion on this at [[MozFest]] 2019, and also watched The Cleaners documentary.
It’s pretty grim stuff - both the distressing content that is created and uploaded around the world, and the way in which the people that are contracted to moderate this content are treated.
Overview
-
the job of content moderation of big online platforms
- it is usually outsourced to other companies
- low paid employees do the act of moderation
Who is setting the policies?
With their policies of what content is acceptable or not, the big tech firms are in some way determining what is acceptable to society.
- basically three/four private companies in the US
- e.g. terrorist groups as designated by US homeland security are used as guidelines for content moderation
- and they do it only based on on profit motives - e.g. reacting to bad publicity
Conditions for workers
Conditions
-
high level of things to see per day
-
Chris Gray, who worked in Ireland as a content moderator, I think suggested around 600 items a day
- 90% of it might be mundane, around 10% of it will be traumatic
- on The Cleaners documentary, I think they said in the Phillipines it’s a target of around 20-25,000 every day??
-
monitoring
- workers are monitored to see if they are making ‘correct’ judgements
- have to meet a quality target otherwise their employment is in jeopardy
-
post traumatic stress
- stress of seeing disturbing things, stress of precarious labour, stress of having to determine what is good, what is bad
Pay
- at MozFest panel Cleaners directors said payment in the Phillipines is $1-$3 a day
- Chris Gray said in Ireland around 12 euro a day I think?
- contrast both of these with the salary of a Facebook engineer…
- question: is it different types of content in different locations?
Support
-
content moderators are under NDAs
- they can’t talk about it with anyone, including friends/family
- talking about it would help with the trauma
Use AI instead?
Why not use ML/AI to moderate this content?
- AI can’t handle the level of complexity involved in some of the decisions
- dilemma: it would put people out of work - but it is unpleasant work
- even for the unpleasant stuff, human input would be required to train any machine learning process anyway
Legal action
- currently a legal action being taken against Facebook by Chris Gray and others
References
- [[The Cleaners]] (film)
- MozFest 2019 panel (probably will be online at some point)
Misc
The central problem is that Facebook has been charged with resolving philosophical conundrums despite being temperamentally ill-qualified and structurally unmotivated to do so.
If nudity can be artistic, exploitative, smutty, and empowering, then the depiction of violence can be about hate and accountability. A video of a shooting can be an expression of deadly bigotry, but it can also expose police wrongdoing. Distinguishing between them requires human decision-making, and resolving a range of contested ideas. At present, private institutions bear significant responsibility for defining the boundaries of acceptability, and they are not very good at it.