In recent years I have tried, but often failed, to use “alt text” in my work when posting images online. I have failed dismally to go back and try to annotate all of the many images I have posted in the past, and I know I fail to be consistent in doing so in the present. Both of these failings are undoubtedly because of the time that this would take me, despite many platforms now encouraging its use, and my commitment to supporting people with disabilities (see my very old site at https://disabilityict4d.wordpress.com/).
For those who are unfamiliar with alt text it is the attribute in HTML that specifies alternative text for images, and is especially valuable for people with visual impairments, because it helps screen readers convey the meaning of images for them. The World Wide Web Consortium thus recommends that every image displayed through HTML should have alt text associated with it.
However, reflecting on this in the context of the increasing use of AI, has made me very aware of the ways in which alt text can be used by AI systems to describe images without the use of data labellers. At one level, this might be seen positively, because it can reduce the need for AI companies to use what is often termed “slave labour” to do the annotation (see Ganna Pogrebna, 2024). However, this would take away the very small income that such labour can generate, and is indeed valued in many parts of the world. Moreover, it is also a way that such images (or video) can be annotated for free for teh AI companies by the person doing the posting.
Much more worryingly, though, is the potential for alt text to be used maliciously. If, with their permission, I post an image of a friend/colleague online, and label this using alt text, their name will be forever attached to that image, and be a vehicle through which AI and search engines can identify them and link to further related images of them. This could, for example, readily be used to track and surveil them when travelling. However, it would be equally possible for someone else to write something unpleasant or abusive about a person as alt text on an image, and that too would be recorded so that AI could then be used to build very erroneous profiles of them.
I am inclined to think that the potential harms of this outweigh the benefits, although for innocent law-abiding people with visual impairments it would be an immense loss. Is this primarily a new way that the Digital Barons are deliberately exploiting us? Is that why platforms are incresingly encouraging us to include alt text when we post an image (as illustrated in the image above)?
Tim Unwin's Blog 