TikTok
the University of California Berkeley
Faddoul
Faddoul’s
Twitter
Facebook
Monster Match
Gizmodo
Marc Faddoul
Recode
’d
No matching tags
No matching tags
Broadway
China
No matching tags
When artificial intelligence researcher Marc Faddoul joined TikTok a few days ago, he saw something concerning: When he followed a new account, the profiles recommended by TikTok seemed eerily, physically similar to the profile picture of the first account. And as he followed various accounts, Faddoul observed that the profile pictures of the recommended accounts seemed very similar to the profile image of the initial account. Those searches produced similar results: recommended follows tended to physically resemble the initial account followed, though they were not always the exact same accounts that appeared in Faddoul’s results. It’s worth noting that in our searches, Recode found plenty of recommendations that had no physical resemblance to the initial account followed.“What I suspect is happening is that TikTok is featurizing the profile picture,” he says, “and using these features in the recommendation engine.” But regardless of how results like these show up, a more general problem with recommendation algorithms is that they can ultimately confirm our preexisting biases.
As said here by Rebecca Heilweil