Post Archive

Artificial intelligence (AI) holds the promise of transforming every aspect of our lives. From self-driving cars to autonomous robots, AI-powered solutions are revolutionizing industry and transforming the way we live, work and play.

AI-powered accessibility solutions are an emerging field in accessible technology—and could one day make the Web universally accessible to people with disabilities. These bolt-on accessibility solutions promise a quick fix for accessibility and are alluring for businesses facing accessibility lawsuits or other compliance issues. But buyer beware— there are limitations to automated accessibility solutions, and these limitations could subject businesses to unexpected liability.

Aurora recently completed an audit of popular AI-powered accessibility solutions, and while we were impressed with some features, we did find accessibility errors in sites using these technologies. Common errors included:

• Images with low quality alternate text, and decorative images with non-empty alt text.
• Keyboard accessibility problems including missing keyboard focus, elements not receiving focus, keyboard trap, keyboard focus contrast issues, and other problems.
• Incorrect heading structure.
• Contrast controls that cause text in buttons and other controls to become unreadable.
• Video content missing a media alternative or text-transcript.

Despite the promise of automated accessibility solutions, there are many limitations to the technology that prevent these solutions from being totally effective in remediating accessibility barriers.

Common Accessibility Errors

Here are some common accessibility errors that we identified in our analysis of automated (AI-powered) accessibility solutions:

Low Quality Image Descriptions

AI solutions for image descriptions currently rely on OCR or other technologies to determine image content and meaning. Unfortunately, these technologies cannot determine if an image is purely decorative or has semantic meaning. Also, complex software may be able to identify an object in an image but determining the purpose of an image in the context of page content is much more difficult for machines. While Facebook and other large tech firms have had some success with automated image recognition, this is an emerging field, and there is much work to be done.

Keyboard Accessibility Problems

Keyboard accessibility problems can be difficult or impossible to identify and diagnose without human testing. AI-powered solutions can identify and attempt to fix common keyboard accessibility problems, but there are many problems that they may miss.

Ensuring that websites work seamlessly with a keyboard requires manual testing to verify that menus, form controls, and other components work well with a keyboard only.

Video Captioning Errors

Companies like Google offer auto captioning for video content to provide a stopgap or bridge to accessibility. Problems with auto captions are numerous and include missing speaker identification, grammatical and captioning errors, timing problems, and other errors. Videos with low-quality audio or background music can make accurate auto captioning difficult or impossible for speech recognition software. While this technology continues to evolve, human captioning is superior to automated captioning to ensure conformance with the Web Content Accessibility Guidelines.

Other Considerations

Legal Liability

Having an automated accessibility solution in place might discourage website owners from having their websites tested regularly to verify and document accessibility compliance. This means that websites could have undetected accessibility barriers and expose business owners to unexpected liability.

Promoting Best Practices

Automated accessibility remediation could cause developers to ignore accessibility, result in poor development practices, and cost business owners more money in the long run.

Ownership

AI-powered accessibility solutions add a layer of accessibility to your website and help to interpret and change markup that poses an accessibility barrier. Unfortunately, when you stop paying for services, you lose all accessibility features and the benefits that come with them.

Cost

The cost of automated accessibility solutions could easily surpass the investment required to build an accessible website for your business. For smaller websites (costing a few thousand dollars), the investment to build an accessible website would pay for itself in 2-3 years.

Conclusion

While AI holds tremendous potential for making the web universally accessible, there is much work to be done to improve the effectiveness and accuracy of automated solutions. Also, automated accessibility solutions can give businesses a false sense of security, and subject business owners to unexpected liability.

While these solutions are far from perfect, they may be a good temporary stopgap for companies working towards developing more accessible web content.

In conclusion, manual testing is the only 100% effective method to identify and address accessibility barriers. Manual testing with assistive technology will always be a best practice for ensuring that web content complies with Web Content Accessibility Guidelines.

At Aurora, we recommend both automated and manual testing to identify accessibility barriers, and we offer industry-leading support to remediate accessibility barriers.