They could certainly do with a control group or three. The point they’re trying to make is that over 5 days of watching recommended videos the proportion that were misogynistic grew from 13% on day 1 to 52% on day 5. That suggests a disproportionate algorithmic boost but it’s hard to tell how much that was caused by the videos they chose to view.
A real world trial ought to be possible. You could recruit thousands of kids to just do their own thing and report back. It’s a very hard question to study in the lab because it’s nothing like the real world.
They could certainly do with a control group or three. The point they’re trying to make is that over 5 days of watching recommended videos the proportion that were misogynistic grew from 13% on day 1 to 52% on day 5. That suggests a disproportionate algorithmic boost but it’s hard to tell how much that was caused by the videos they chose to view.
A real world trial ought to be possible. You could recruit thousands of kids to just do their own thing and report back. It’s a very hard question to study in the lab because it’s nothing like the real world.