Web Accessibility and AI

There has been a lot of talk about Artificial Intelligence (AI) lately and how it is being used. As with almost any technology, it can be used beneficially or maliciously. Unfortunately what we see in the media mostly these days is about how AI is being used maliciously. Deep-fakes for audio and video, for sextortion, cheating, plagiarism, misinformation, and disinformation.

Let's take a look at how AI can be beneficial as it relates to the development of web content.

As web developers we look at new technology with an eye towards whether or not it can make our tasks more efficient. One of the tasks that we deal with now on a daily basis is whether or not the content we help our clients with is accessible. AI could be very beneficial to our work toward this goal. If our programming applications can provide algorithms to be sure that the code we are writing is following accessibility guidelines, we could allow AI to embed the appropriate attributes into our code from the start.

Having an AI assistant be able to evaluate the structure of the content and code that has been built and pointing out any errors or best practice hints would be very valuable.

This kind of tool would not work like AI content writing applications like ChatGPT where you're asking the application to write content based on an amassed knowledge base. The applications we use for coding would be specifically checking against known and accepted guidelines and rules to make recommendations based on the recognized coding structure.

Outside of these standalone programming applications, we see the potential for AI to help with content development itself. Specifically related to the non-textual elements of the content. We look at elements like images, charts, graphs, and infographics. How could AI help with making these elements more accessible?

In reading an article called "Opportunities for AI in Accessibility" at A List Apart and a couple of other articles linked there, here are some of our own thoughts on the subject.

There are AI tools out there that are now able to analyze and provide descriptions or Alternative Text (alt text) for images. At the moment these are still at a stage that the interpretation is specific to the image or graphic itself, the AI has limited capability of interpreting the image within the context of surrounding content. For now a human review of the content created by AI is still necessary to be sure that what is created is relevant to the context in which the image exists.

These tools are also not fully available in the content management environments that our websites exist in. The computing power for these AI tools is more than a web host would be able to provide for the average website and therefore are a tertiary tool that can be applied through an API.

Google's Vision AI and OpenAI's ChatGPT and InstructGPT are well known examples of these tools. They are now in regular use for content creation, but they are still under development. Both organizations have strong teams supporting them and are applying a lot of resources to their AI goals. These tools will definitely be part of the web developer's toolkit going forward, but they will still need that human input for applying the content to the live web page or application.

At some point these tools will have APIs to regularly integrate into most CMS solutions and AI will be able to supply the copy that content administrators want. The only caution we would have is that these AI tools are only as smart as their knowledge base allows them to be, and we as developers need to be conscious of content biases in order not to propagate misinformation or misinterpretations.

AI is going to be part of our future. It's best to get on board now so that we can help it down a more beneficial path and that we have an understanding of how it can be used to make the lives of people with disabilities more manageable.

to blog