“Facebook’s AI is scanning your memes for hate speech” – (BusinessToday, 2018)
Recently, Facebook is under fire since it has become known that Cambridge Analytica used data from millions of users to try to influence elections. Consequently, Facebook was in the middle of a scandal and shares dropped rapidly.
Because of that and other reasons, Facebook has invested in a machine learning system called Rosetta that scans text in images and videos of all kinds for hate speech and other inappropriate language. Text recognition systems have proven to perform insufficiently due to the large amount of images shared on the social medium every day. Rosetta is able to understand context, because the system can distinguish between a joke and hate speech.
However, AI on itself is not the “magical solution” to hate speech (yet) and there are still quite some hurdles to overcome. A great example is Rosetta, which I already mentioned. It is not as good at finding hate speech as expected (and wished); only 38 percent of the discovered hate speech was flagged by the artificial intelligence system. This means that almost two-third is still done by humans. We can start wondering if AI will ever be as intelligent as we wish it to be and to what extent AI’s are able to operate on their own.
Human language and interaction is very complex and so broad, that whole studies are written about it. Is it therefore realistic to expect that AI will be able to find hate speech and other inappropriate language one-hundred percent of the time? As far as we know, we are not there yet, but who knows what future technological developments will bring.
Sources:
https://www.businesstoday.in/technology/news/facebook-artificial-intelligence-rosetta-is-scanning-your-memes-for-hate-speech/story/282296.html
Facebook data scandal: How could likes profile voters for manipulation?
https://www.cnbc.com/2018/05/15/facebook-artificial-intelligence-still-finds-it-hard-to-identify-hate-speech.html
https://www.researchgate.net/publication/300702987_The_Complexity_of_Human_Interaction
