I think that's what you're asking? It's a good concept that would be difficult in practice. Technically, we could have the software check when someone posts a message to see if the post contains an image. If it does, we could pop up a page for the user asking them to describe the image. But, this would get complicated. What if there are multiple images in the post? Do we handle them all on one page, or do we do a single page for each image?
Unfortunately this would also be onerous for people, especially for those who post OPs with many images, like the toon posts or the Sunday LOLcats. You also have the problem that people would not necessarily describe the images properly. Even if we force them to enter text, they might just put a bunch of spaces, or they might enter an incorrect description because it's funny, or they might write something obscene. That means we'd then have to separately monitor every image description that was entered, essentially creating a separate moderating system for image descriptions.
Those are the initial concerns off the top of my head. I agree that it would be a great thing if all the images were correctly captioned, but as you can see there are many practical hurdles to overcome.
I think we all know the obvious downsides of AI, but that doesn't mean there aren't specific situations where it will actually be useful rather than harmful -- for example, apps for people with vision impairment that can describe images which don't have alt tags are already being created. This kind of tech will quickly overtake any efforts we can make to accurately describe images at our end; it may just be better to leave the problem to third party apps.