Children's appearance is not likely to change, in ever more sophisticated ways, so as to still appear like children to humans but not to the algorithm. Thus, while not intrinsically harder than text analysis, it could very easily be the case that it is easier to flag videos as having children (and thus turn off comments) than it is to automatically police comments.

Note that they said it won't apply to videos of teenagers; my guess is that this is because it would be more difficult to automatically distinguish between teenagers and young (or just young-looking) adults.