Over the past few years, innovations in artificial intelligence (A.I.) have had a profound impact on everyday life.
A.I. played a critical role in the development of COVID-19 treatments and vaccines, and virtual assistants have become more useful thanks to new natural language processing models. Within the next decade, automotive manufacturers are expected to roll out self-driving vehicles and “robotaxis,” potentially reducing traffic fatalities while freeing up space for parks, housing, and business development.
These developments are only possible if computer systems can perform tasks that usually require human-like intelligence. A.I. systems solve problems by perceiving objects visually, recognizing speech, and making decisions based on situations — tasks that normally require input from a human being.
Those capabilities may change the landscape of digital accessibility. Machine learning could reduce the cost of testing and remediation, allowing more websites to remove the barriers that affect real users with disabilities.
In order to create accessible content, creators need to follow established standards.
The Web Content Accessibility Guidelines (WCAG) are the basis of many accessibility laws, and websites that conform with the Level AA guidelines of the latest version of WCAG are considered reasonably accessible — and compliant with the Americans with Disabilities Act (ADA), the European Union Accessibility Act, and other accessibility laws that apply to online content.
But while WCAG contains straightforward guidance for improving the internet, many businesses have limited resources for compliance testing. Automated tests are an inexpensive way to find certain WCAG issues and find ways to remediate those barriers.
The problem: Most automated tools aren’t intelligent — they can’t perceive certain types of content, and they rely on strict rulesets to operate. When accessibility issues fall outside of those rulesets, the tools may report a false negative (they fail to identify the issue) or a false positive (they identify issues that don’t impact accessibility).
For example, automated accessibility tests can determine whether websites use alternative text (also called alt text) for images, but they can’t judge whether the alternative text is useful. If an image of an orange contains the alt text “apple,” current tools can’t identify the mistake. The tool won’t report an accessibility issue, but the poor alt text could affect real-life users.
Related: Is Automated Testing Enough for Accessibility Compliance?
Automated testing tools have improved substantially over the past few years, but they still have limitations. That could change in the future. By leveraging innovations in machine learning, accessibility tests can mimic the natural behaviors of users and provide more accurate reports.
For example, an advanced A.I. tool could compare each image with a large dataset of similar images to identify inaccuracies in alt text, then flag potential issues for review. As machine learning improves, A.I. might be capable of making necessary changes automatically — without human input.
Similarly, A.I. tools will be able to consider the context of web content to make better suggestions. Fully automated audits may be able to detect misapplications of ARIA roles, keyboard accessibility issues, and barriers that only affect certain technology platforms (such as specific versions of screen reader applications).
Related: 5 Quick Ways to Self-check the Accessibility of a Website
Human judgment will remain a critical component of digital accessibility for the near future. Manual audits are essential for ensuring conformance with WCAG, and while machine learning will make auditing easier over time, we strongly recommend using a combination of manual and automated tests for full compliance with the ADA and other accessibility laws.
With that said, artificial intelligence will gradually reduce the cost of accessibility testing and remediation, allowing more creators to improve their content. Online accessibility is always worth the investment, but as A.I. reduces the cost of that investment, more websites (and users) will enjoy the benefits.