AI and Human Collaboration in Accessibility Work – The Binclusive Example

Anton Sten’s article provides us with a good framework for understanding AI’s capacity. AI output always tends to stay at a “good enough” level. Therefore, human effort is necessary for a truly good output. The mistake made in the marketing of accessibility products is exactly here. Solutions that claim you will become accessible with a single integration and that you don’t need to put in any effort simply don’t work. What actually works are solutions that start the work with AI and then easily incorporate human effort with AI support. The products launched by the Binclusive are great examples of this approach. Here are explanation of 2 products:

• Binclusive Audio Description Tool: By uploading your video to the audio description tool, you can get an AI-generated description output in a short time. Uploading the video, skipping the settings, and pressing the button is the basic flow you can proceed with. Although the tool works well, you get a “good enough” output with this method. You can specify the points that need to be described in detail by writing prompts through the tool. You can also adjust the level of description detail you want from the settings. After receiving the output, you can edit it through prompts as well. The important thing here is that it serves as an assistant that accompanies you and simplifies your process, rather than the “give us the video and we’ll describe it” approach.
• Binclusive PDF Remediation Tool: By uploading your PDF to the remediation tool, you can generate an accessible version of it. The “Reader-focused” option does not guarantee protecting the visual structure of the PDF, but “Preserve Source” adds semantic tags while maintaining the visual design. The feature I like most here is the ability to edit the output after receiving it. The tool mostly identifies elements that should be headings, accurately tags lists, and adds alt text to images. On the other hand, it may not always correctly determine whether a heading should be an H2 or H3. Or if an image should be marked as decorative, you have to change this from the tool. As a note, the tool is not a PDF editor; it is actually a simple tool provided to you for editing HTML tags. Once again, this is a great example of the editability of AI-generated output.