This week contained two mini-lectures, one a recap of how to design with accessibility in mind and the other looking at how to design cohesive multi-channel experiences.
Accessible design is something I believe to be important, not only in digital design but everywhere from buildings to experiences. Too often accessibility is the last consideration and this causes solutions added at the end of the process, rather than incorporating accessibility into the design from the start.
Often as designers, we can also fall into this trap, thinking accessibility is all about passing the WCAG guidelines for colour contrast, but there are many more things to consider. This week I want to look at the role AI could play in accessible design as well as any issues it could throw up.
While AI is still a major topic in technology, with many stories about how the latest iPhone is going to incorporate its power, I have not come across any articles considering how AI could help (or hinder) accessible design. During this week’s lecture, we discussed some possible uses and I wanted to have a more thorough look at what is being used and how well it is working.
I found these two articles, the second a response/follow-up piece to the first looking at how AI is used to create Alt-text and the author’s opinions on its success.
https://www.joedolson.com/2023/06/accessibility-and-artificial-intelligence/
https://alistapart.com/article/opportunities-for-ai-in-accessibility/
Last Accessed 01/05/24
The author of the first article is less than impressed with the current use of AI to create alt-text. He points out that while AI can generate a description of the image, alt-text is designed to give the meaning of an image to those who cannot see it. This requires a description of the image in context with the rest of the content alongside it. An image description doesn’t offer this, it will simply describe a man standing beside a building, rather than considering the importance of the building or the man himself. The author of this article suggests that while AI is great for creating image descriptions and these descriptions can form the basis of alt-text, currently we need a human in the loop to create useful alt-text. A use of AI that the author mentions, and I think would be useful would be an AI assistant in a code development tool say for HTML, this AI would suggest accessible code, such as pointing out the need for alt-text on images but would leave authoring the alt-text to a human. Having created a few websites now I know how easy it is to neglect to put alt-text on images, and I may be making other mistakes that affect the accessibility of these websites, so I would welcome an AI which guided me rather than doing the work for me.
Another area that AI is often cited as having the potential to revolutionise is data analysis, one of the hardest things I have found is creating useful alt-text for graphs, charts and infographics. As these rely so much on their visual appearance to impart information, trying to create a text alternative that has the same impact and conveys the same meaning feels nigh on impossible, combined with the urge to keep alt-text short so as not to impact the flow of the piece these elements are placed within, puts this task into a list of tasks I try to avoid. Could AI use data analytics and text generation to provide short, useful, contextual text alternatives for charts, graphs and infographics? I believe it could although it would take a determined development team and human oversight to build an AI capable of doing it reliably and to a high enough standard. This for me is one of the key aspects of AI, it can do amazing things but needs specific tasks and oversight to ensure the quality and relevance of output. Major LLMs such as Google’s Gemini and OpenAI’s ChatGPT are incredibly impressive but being trained on open Internet-sourced datasets has built-in issues of the Internet such as exclusion, bias and inaccessibility. For AI to take the next step it needs better training data related to specific tasks, a one-size-fits-all AI is currently not available.
AI has the potential to revolutionise online accessibility, but only if humans decide to build and train the correct types of AI and oversee their outputs to ensure they do what is required, otherwise AI for accessibility could easily be a missed opportunity and potentially exclude people from its uses, increasing exclusion online and in digital services.
AI is an area I feel I will be reading about and assessing for the rest of my career as it inevitably improves and takes on different roles, currently for accessibility it is not quite where it needs to be to be something I would use day-to-day but the potential is there. Hopefully, soon we will see AI tools assisting designers and developers to create more accessible technology products and allow full access to the power of technology to as many people as possible.