With enterprises in the race to embrace digital change, other tech companies are doing what they can to make this happen. AudioEye, a digital accessibility firm, is no exception to this. They have embraced growing changes, like machine learning, to accelerate accessibility and help organisations achieve digital inclusion standards. Sean Bradley, CoFounder at AudioEye talks of why machine learning is so vital to today’s evolving tech sector.
At what point did you realise that a change needed to happen in digital accessibility?
Beginning in 2015/2016, businesses started reaching out to us having found themselves on the receiving end of website accessibility claims, despite having neither a) any control over their source code or b) any understanding of where to begin in order to address the claims. The traditional approach toward accessibility was too slow, too expensive, and too impractical for the vast majority of businesses seeking to rapidly address the issue. Speed-to-compliance and an outsourced solution quickly became necessary, prudent, and cost-effective for these businesses. Addressing these business needs was central to [our] approach and technology-driven solutions.
How are you using machine learning to help firms achieve digital inclusion standards?[The technology we use] can identify accessibility errors, associate those errors with specific elements or contexts, and fix those errors, programmatically. When similar fixable errors are identified, we can make a determination of whether or not an error can be fixed, programmatically, based on the results of heuristic analysis. As an example, a website may provide links to 3rd party destinations promoted from various iconography. If not labelled properly, users of assistive technology may not know what link the iconography takes the user to. Once remediated and validated in one instance, future instances sharing similar iconography/link failures, may be remediated, automatically, and as a result of the fixable error having already been remediated in prior and unrelated instances.
Where do you think machine learning can have the biggest impact in general?
From suggesting potential fixes for issues of accessibility to identifying issues that today require a human AT tester to identify. We force a future wherein manual QA testing is far less time consuming, more targeted, and vastly more effective than it is today. In a general sense, ML is intended to make computers smarter and more effective while reducing the burden placed on humans performing highly repetitive and monotonous and/or complicated tasks.
How did you involve software testing and other types of tech in your development?
Software testing is an important component of any technology company. We are actively building technology that will take traditional software testing methodologies and multiply their benefits by utilising the efforts to train models which ultimately will reduce the level of effort required for testing going forward.
What do you see happening with the future of ML?
In the future, we will be able to more accurately test for issues that currently have tests in numerous vendors’ test suites but extend beyond the possibilities of today by learning to detect issues only detectable by human testers. By evaluating the pages we have performed work on and analysing the page content and structure both before and after AudioEye’s remediations have been run, the future of ML for us will enable computers to begin comprehending a whole new set of issues currently undetectable by computers. We may see a time in the near future where 90%+ of the issues of accessibility on any given web page can be completely eradicated in near real-time, while also providing manual AT testers with highly targeted test plans to increase their efficiency and effectiveness. And of course, all of the efforts put forth by those manual testers will feed back into the models, further training the systems and making them smarter, increasing the percentage of issues identified and fixed, for a future goal of complete accessibility automation for HTML-based content.