I don't think the assessment that they were part of a future algorithm is too far off. Given enough expert input picking a fitting image/video or headline should be doable. Worst case you reduce the number of humans needed to one quality assurance person. Algorithm says "this headline, this image"...yes/no;fix.
