A TikTok novelty: FACE-BASED FITLER BUBBLES
The AI-bias techlash seems to have had no impact on newer platforms.
Follow a random profile, and TikTok will only recommend people who look almost the same.
Let’s do the experiment from a fresh account:
1/6
Conversation
Clearly, recommendations are very physiognomic.
But it’s not just gender and ethnicity, you can get much more niche facial profiling.
TikTok adapts 'recommendability' on hair style, body profile, age, how (un)dressed the person is, and even whether they have visible disabilities.
5
80
415
FEATURE or BUG?
Similarity is expected from a recommendation engine, but not homogeneity, nor does it have to be based on physical appearance.
TikTok’s assumption is that, if I follow a blond teen, it is because I like blond teens, not for her humour or the music she features.
6
41
359
Social Media have been known to create filter bubbles for political opinions. TikTok seems like the first major platform to create such clear physiognomic bubbles.
3
60
417
Can you post a link to any research on this? I’m curious from an ML perspective and an ethics view. Would serving more disabled posters to a disabled user be bias or helping build community and connections?
2
31
This was just an observation, I did not do research on it. But an interesting track to explore!
About disabled users, I agree that there could be value in theory. But they'd need to ask this community for their opinion, and make this an intentional feature (which i doubt this is)
1
2
48
Show replies
1
1
4
thanks for the pointer! I had seen that, concerning for sure...
Quote Tweet
TIKTOK INTERNAL MODERATION GUIDELINES LEAKED by @theIntercept
TikTok explicitly prescribes moderators to demote ‘abnormal body shapes’, ‘ugly facial looks’, ‘dwarfism’, ‘beer belly’ or ’too many wrinkles’. twitter.com/theintercept/s…
Show this thread
3
Hi Marc! This is most interesting. How exactly did you produce these results? I've just tried to reproduce them without success.
1
17
Start with a new account. Go to the default thread 'For You'. Swipe right to see an account. Click 'following' or the down arrow. 3 suggestions show up. These are the one displayed on my thread. ( I have the full screenshot but I thought it'd be better to remove the usernames)
1
16
Show replies
Couldn't that somehow be the result of appearance-based human biases (the good ol' "people who follow A1 also follow A2 and A3") feeding into some sort of clustering algorithm, rather than the algorithm itself being appearance-aware/-biased?
1
4
yes, it could just be the result of collaborative filtering. But if appearance is the main driver for people to follow other profiles, then a collaborative filtering model will basically be capturing appearance.
1