Isn't he simply questioning an assumption? (assumption being that every bit of image always needs an alternative that is not in normal flow of page)

Ian's often repeated naive-sounding "Why would you do that? What's the use case?" has shot down many feature requests that seemed good in theory, but weren't that useful in practice.

Perhaps description of the video makes description of poster redundant? Maybe one can use ARIA? Maybe media framework can extract description from the video file?

Even if there is a way to describe the poster, will authors actually care to do it? Will they ignore it (naively assuming blind users don't watch videos) or fill it with boilerplate?