Kagan MacTane (he/him)Quick #accessibility tip: Rather than writing an alt text that says nothing but "image", you might as well just write one that says "fuck you".
Daniel Howard 🌻@kagan I know a guy who has AI fill in the alt text. And I'm like, couldn't the screen reader just use an AI to fill in the alt text? Maybe then the blind person could use an AI that suits their needs. You're just dropping slop into the equation. But maybe I'm wrong.
Aaron Sawdey, Ph.D.
Jamie Teh@acsawdey @dannyman @kagan (Blind screen reader user here) To be fair, I can understand why writing alt text might be hard for some people, especially if they're new to it. Writing in general doesn't come naturally to some folks, especially if it's out of their comfort zone. So, I think it might be reasonable to use AI as a starting point, as inspiration for an image description. Sometimes, it's easier to know what you want if you have something imperfect to start you off. But the catch is that this only works if the author closely scrutinises what the AI produces and edits it for context and accuracy, and I suspect many wouldn't bother, which makes it hard to recommend this.
Tiota Sram@jcsteh @acsawdey @dannyman @kagan yeah, I've seen a fair amount of clearly AI alt text on here, and lots of it misses important context, while >1% of it incorrectly describes what's in the image.
1% might not seem like much, but imagine if your social media client replaced 1% of pictures with unrelated images. It would mess with your head quite a bit if you had no way to tell other than sometimes the image contradicts the text of the post.
Another point in these conversions: writing alt text for my images often helps me improve the text of my posts, and this is even more true in other contexts like when I'm making presentations or other teaching materials. I've put together many diagrams that were vastly improved by having to stop and think about how to describe them, leading me to reorganize the diagram to make it easier to describe. Even when I can't edit the image, I can often sharpen the surrounding prose after being forced to express in words what the image shows (most common case is realizing the in made doesn't clearly show the thing I'm trying to illustrate so I add both specific alt text and extra supporting text to point out what I talking about).
@tiotasram @acsawdey @dannyman @kagan Further to this, if you're going to use AI and not bother correcting or at least checking it in any way, that's probably worse than not providing any alt text at all. That way, even if the reader chooses to use AI, they at least know it's potentially inaccurate and are aware that they need to make their own judgement, use additional/multiple tools, etc. I still contend that AI could help some people by providing a starting point for their descriptions, but *only* if they're going to use that responsibly. (Which is the big problem with a lot of AI things: it just isn't used responsibly.)
@jcsteh @tiotasram @acsawdey @dannyman I've seen a fair number of AI alt texts that I thought were better than nothing. They managed to describe the image well enough for understanding, although generally they were overly-verbose and included irrelevant detail.
I've also seen them miss the point so badly that they were useless or worse. While those were a large percentage, I don't think they were a majority. (IME, of course.)
@jcsteh @tiotasram @dannyman Also, I don't think I've ever seen AI give a decent alt text for anything humorous. Cartoons, jokes, anything like that — as @acsawdey pointed out, AI lacks anything resembling context, and context is critical for delivering jokes properly.