Digital Accessibility
Remediation: Can AI Tools
Replace the Human Touch?
By: Prudhvi Mandava
Artificial intelligence, or AI, is popping up all around us, bringing lofty promises of revolutionizing the world as we know it. We’ve seen the development of self-driving cars, the rise of generative AI apps such as ChatGPT, and websites and movie studios replacing the work of actual human writers with AI-generated content. Wherever we turn, we see a growing reliance on AI across industries ranging from journalism to finance to art.
But along with the increasing proliferation of AI comes a growing amount of backlash and controversy as we ask ourselves: Can artificial intelligence really replace human intelligence?
One developing field of AI tools involves digital accessibility compliance—adherence to Section 508-mandated federal guidelines and the remediation of public-facing web content. One such application that we’ll focus on here involves digital documents, particularly PDFs. Can we count on AI to make documents accessible to users with disabilities, or do we still need human-led remediation?
Let’s review some common remediation considerations to find out.
Alt Text
The purpose of alternative text, or alt text, is to accurately describe images in a digital document that someone with visual impairments would be unable to see. The descriptions are relayed to the user’s assistive technology, such as a screen-reading tool, which then reads the content aloud to the user. As such, the process of creating alt text does not end with inserting an image; consideration is required to describe said image in a meaningful way.
I recently reviewed a document in Microsoft Word, which features its own accessibility checker tool. As I scrolled through the tool’s automatically generated alt text, I came across the following description:
A picture containing text, map, screenshot, diagram
Not very helpful, is it? Is it a picture? Is it a diagram? What is the information being conveyed? You’d be surprised by how many automatic descriptions are generated this way. When a screen reader comes across vague, ineffective descriptions like this one, the overall comprehension suffers. However, something like this would be caught in a quality control check by a human content owner or remediator.
Auto-Tagging
When it comes to PDFs, the most fundamental aspect to consider is the tags, which are containers that categorize the content and place the content in a hierarchical order that can be easily read by assistive technologies. The tags are intended to identify the structure of a document—its headings, paragraphs, tables, charts, graphs, lists, and images—indicating the correct reading order.
For example, let’s say your PDF document features a table with several columns. A human would know to read the content by column, but a screen reader, unless told otherwise, usually reads content from left to right, top to bottom. So, if your table is mistagged or untagged altogether, a screen reader would not know to read the table the way it was meant to be read. Proper remediation is necessary.
However, when using software that claims to auto-tag the document, the software is as good as the samples recognized in the database. There’s a good chance that the content will be mistagged or inaccurately tagged. Sometimes, if the content was well formatted from the source document, the AI software will practically reset the near-perfect formatting as well as the associated alt text for figures. A simple recognition of formatted and designed documents by a human remediator would allow one to clear the minor issues as opposed to using AI to make the problem worse.
Security
One of the major concerns for any business or government agency is security risk. We can all agree that confidentiality breaches are bad for business. Third-party AI tools often require access to documents that may contain personal or proprietary information and can even introduce potential copyright issues. Human-led remediation ensures the privacy of both your organization and your end users. Humans can adhere to strict workflows and security checks for accountability when accessing said information.
Conclusion
Although AI-powered tools have come a long way in recent years, they simply cannot always generate documents that are fully accessible to users with disabilities. Such tools struggle to comprehend the full context of the document’s content, which can result in misinterpretation, poor user experience, and inaccuracies.
For websites and other digital documents to reliably be accessible to all users, they need the assistance of trained, human remediators. The humans here at The 508 Institute are experts in 508 standards who can teach you everything you need to know about remediating documents, creating accurate and effective alt text, testing websites for 508 compliance, and conducting rigorous quality control assessments.
Reach out to The 508 Institute today to learn how we can help; we’ve got Good Humans ready to assist you!
