I appreciate this discussion bc it helps shine a light on the complexity of these problems. Two things to note as we all work to tackle inauthentic behavior & deception. 🧵
https://twitter.com/etbrooking/status/1461735800702488579
1. There’s a big behavioral difference between spammy amplification and complex IO;
2. Platforms traditionally approach each differently for a reason — each represents different behaviours and has different incentive structure.
2. Platforms traditionally approach each differently for a reason — each represents different behaviours and has different incentive structure.
Boosting shares and likes is a numbers game to make content look more popular than it is. It can be used on political content or fake sunglasses (or both).
Either way, it’s on the simpler end of the spectrum.
@markhansontoo discussed it last year about.fb.com/wp-content/upl…
Either way, it’s on the simpler end of the spectrum.
@markhansontoo discussed it last year about.fb.com/wp-content/upl…
The best way to tackle a numbers game like this is to counter it at scale via auto detection & removal — no matter the content.
Some accounts will always slip through the net, but billions don’t (FB reports that quarterly). Fakers adapt; the challenge is to keep adapting too.
Some accounts will always slip through the net, but billions don’t (FB reports that quarterly). Fakers adapt; the challenge is to keep adapting too.
Now back to complex IO: AFAIK, our research community has not seen examples of successful influence operations and foreign interference being built solely on fake likes and shares for years.
Efforts like that are basically one-trick ponies.
Efforts like that are basically one-trick ponies.
The IO we’ve seen recently have typically combined many different tactics.
Fake likes/shares can be one of the (less effective) ones, but there are so many others, e.g. backstopped cross-platform personas, fake media brands, or co-opting real people: about.fb.com/news/2021/05/i…
Fake likes/shares can be one of the (less effective) ones, but there are so many others, e.g. backstopped cross-platform personas, fake media brands, or co-opting real people: about.fb.com/news/2021/05/i…
As @davidagranovich has already pointed out, it’s really important not to conflate the different kinds of inauthentic activity.
Even within influence ops, it’s important to assess their likely impact based on the available evidence.
Even within influence ops, it’s important to assess their likely impact based on the available evidence.
https://twitter.com/davidagranovich/status/1439020401946816516
This handy scale is designed to estimate the impact of IO based on observed evidence, categories 1 → 6.
If an op stays on one platform/community, it’s a category 1. It goes up if it reaches more communities, platforms, mainstream media or celebrities.
brookings.edu/research/the-b…
If an op stays on one platform/community, it’s a category 1. It goes up if it reaches more communities, platforms, mainstream media or celebrities.
brookings.edu/research/the-b…
Always remember, some influence ops seek to overstate their own impact - remember the IRA in 2018?
The more we can scrutinise whether they *actually* got anywhere, based on the evidence, the stronger our defender community will be.
The more we can scrutinise whether they *actually* got anywhere, based on the evidence, the stronger our defender community will be.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
