Paperonium

Children’s media · JAMA Netw Open · 2024

The paper’s summary: following video recommendations brought more gender stereotypes over time. Its own data: fewer.

The paper’s summarygender stereotypes “increased”no changeIts own results and chart“decreased”: about 25% → 17%
The paper’s summarygender stereotypes “increased”no changeIts own results and chart“decreased”: about 25% → 17%

Did gender-stereotyped pictures grow more or less common over 20 recommendation steps?

The paper says

On a video site children use, researchers searched 12 terms, opened the top result, then followed randomly picked recommendations for 20 steps, rating the 12 recommended videos’ preview pictures at each step. The summary says gender-stereotyped pictures grew more common the further they went.

Its own data show

Its results and discussion say these pictures “decreased”; a minus sign in its table says the same. Its chart shows about 25% of pictures gender-stereotyped in steps 1–5, 17% in steps 16–20 (our reading; exact counts aren’t printed). Summary and results both call the change very unlikely to be chance, but disagree on its direction.

A paper’s summary is the short overview at the top, which many people read instead of the whole paper. This study tested a “rabbit hole” idea: that problematic content grows the further you follow recommendations. The summary rightly says most such content did not grow, but wrongly says gender stereotypes did. In fact they shrank: no problematic feature grew.

“The summary” means the paper’s abstract: the short summary printed at its top, which is the part most readers see.

Read the paper →Full technical write-up →

Algorithmic Content Recommendations on a Video-Sharing Platform Used by Children · JAMA Netw Open (2024) · doi:10.1001/jamanetworkopen.2024.13855