Facebook – The Dunce pt. III

AI so dumb, it accepts ads from ISIS.

• Despite its rhetoric regarding its progress in the fight against fake news, even the most obviously fake placements still seem to be getting through, highlighting once again that at the root of almost all its problems lies its weakness in AI.
• Tests carried out by Vice News (see here) and Business Insider (see here) were successful in that VICE News was able to post political advertisements paid for Mike Pence (VP USA), Tom Perry (DNC Chairman) and ISIS and Business Insider posted advertisements paid for the now defunct Cambridge Analytica.
• Ironically, advertisements claiming to be paid for by Hilary Clinton were blocked.
• Facebook has said that no foreign advertisers are permitted to place political advertisements, but VICE News was still able to re-post exact copies of advertisements allegedly used by Russian agents in 2016.
• This is yet another indication of just how poor Facebook’s AI is and that anything that is not explicitly programmed for is likely be able to slip through the net.
• A decent AI would have known that Mike Pence and Tom Perry are active politicians without having to be explicitly told and I am pretty sure that it blocked Hilary Clinton only because it was explicitly told to do so.
• I am surprised that no one at Facebook seems to have told the algorithm that ISIS is a world known terrorist organisation.
• Furthermore, in AI terms, this is not rocket science.
• Image recognition, voice recognition and video processing are complex AI problems, but this could have been spotted with a simple text search.
• For the last 2 years, RFM has ranked Facebook dead last in the AI arms race that is currently being run among the digital ecosystems and these findings indicate that this state of affairs has not changed.
• The understanding of one’s data is only stage one of three when it comes to creating and deploying AI to improve one’s digital life services (see here).
• It is because Facebook cannot get past stage one, that it is having to hire armies of humans to analyse its data because its machines are too stupid.
• To make matters worse, humans are ill-suited to this task as to prevent objectionable content from trending, it needs to be seen and identified very quickly after it has been posted.
• Much of the human fact-checking that Facebook does takes days to flag content as fake or objectionable and often it has trended and is in decline before it is removed.
• This is the worst of both worlds as the humans are really expensive (as declining EBIT margins show) and they can never be as good at this particular job as the machines can be.
• This is why AI remains by far the biggest problem that Facebook has today, and while progress is happening, it is painfully slow.
• I think that the net result will be financial underperformance for the next couple of years meaning that the time to get back into Facebook is not even close.
• I am sitting on the sidelines until I can see a meaningful improvement in its AI.
• There is worse to come.