Thread by @ylecun

@mcCronjaeger @BloombergME The list is much too long for a Twitter thread.
I'll leave that for FB's comm people to do.

@mcCronjaeger @BloombergME More importantly, the whole premise of the article is wrong.
The SAIL / Responsible AI group's role *never* was to deal with hate speech and misinformation.
That's in the hands of other groups with *hundreds* of people in them.
In fact, "integrity" involves over 30,000 people...

@mcCronjaeger @BloombergME So the central theme of the article, that RespAI wasn't given the necessary resources to do its job is patently false.

Second, AI is heavily used for content moderation: filtering hate speech, polarizing content, violence, bullying, etc...

@mcCronjaeger @BloombergME The idea that these things are not taken down is false.
For example, 94.7% of hate speech was taken down preemptively by AI systems in Q4 2020.
The rest was taken down after user flagging and manual review....

@mcCronjaeger @BloombergME Then there are independent academic studies that show that the growth of polarization in the US (and Canada) has been happening for the last 4 decades or so and is *not* correlated with social network usage.
pnas.org/content/114/40…

@mcCronjaeger @BloombergME What's more, other OECD countries have seen a *decrease* of polarization over the last several decades (e.g. Sweden, Norway, Germany...). The US/Canada are a special case.
This can't be explained by the use of social media.
nber.org/papers/w26669

@mcCronjaeger @BloombergME Then, the idea that FB shows polarizing content because it's good for "engagement" is ridiculous.
The criteria on which Newsfeed is based were completely overhauled in late 2017/early 2018, *precisely* to minimize polarizing content, clickbait, information bubbles, etc.

@mcCronjaeger @BloombergME The one thing that's true is that FB assumed good intent from user for too long, and exposed itself to being exploited by people with nefarious intent: stoking ethnic conflicts, influencing foreing elections, etc....

@mcCronjaeger @BloombergME But once the company realized what was happening it quickly took corrective measures: protecting the integrity of elections, taking down Myanmar government pages spewing hateful propaganda, etc.

@mcCronjaeger @BloombergME The thing is, it's not easy to do all this. You need content moderator that speak all the world's languages, or else you need AI for translation and hate speech detection in all languages.
FB actually has the world's best Burmese-to-English translation system.
Guess why!