Listen to this story
AI has the potential to revolutionise the way humans interact with their devices, especially for those with health conditions and physical impairments. Whether it’s text-to-speech models in screen readers for people with visual impairments, or intelligent accessibility changes for those with mobility issues, accessibility for all is an important part of software development and AI innovation.
To raise awareness on this issue, entrepreneurs Joe Devon and Dennison Asuncion launched Global Accessibility Awareness Day (GAAD) in 2012. Yesterday marked the 11th anniversary of GAAD, and keeping in line with this theme, both Google and Apple released new AI-powered accessibility features for their devices. Out of all the battlefields tech giants can pick, this is likely one for the best cause.
AI for all
On the occasion of GAAD, Google, in a blog post detailed a host of improvements it has made across its product lineup. Firstly, it spoke about Google Lookout, an app launched in 2019 to help visually impaired individuals navigate the world with image recognition algorithms. Targeted at visually-impaired individuals, this application is now getting an upgrade that uses AI to generate alt text for images on the internet that don’t have it.
Android’s AI-powered Live Caption feature also got an update with the feature being rolled out to more phones over the summer. In addition to this, speech-impaired users can use Live Caption to type back responses over calls and have the response read aloud to the other caller.
Apple, on the other hand, released one of the most revolutionary accessibility improvements to their devices. Made for users diagnosed with ALS (amyotrophic lateral sclerosis) or other conditions that impact speech ability, Apple released a feature called Personal Voice. By reading aloud 15 minutes of randomised text prompts, users can train an AI model to sound exactly like them.
To keep the process secure and private, all machine learning processing is done on-device. Personal Voice is also integrated with Live Speech, another accessibility feature that allows users to convert text to audio on phone and FaceTime calls.
Additionally, Apple also introduced a feature for the visually impaired. Adding to the detection mode in iPhone’s Magnifier, Apple introduced another feature, known as Point and Speak. It uses image recognition algorithms to process data from the camera on the device to announce images and text to users.
While a majority of these features reduce the barrier to entry for individuals with disabilities, integrating AI in this field must be done with a measured approach, lest it raises security concerns later down the line.
Balancing harm and benefit
The primary exhibit here is Apple’s Personal Voice feature, which, on a surface level, is a life-changer for those who wish to interact with the world in their own voice. However, on the flip side, it can also be used for malicious purposes. Suppose a mal-intentioned party got a user to go through the 15-minute prompt required to train the personal voice, the output can be used as a method to launch a social engineering attack on the person.
While the current crop of AI-powered accessibility features come with on-device ML processing, the issue of data privacy also tends to arise in such matters. If the accessibility data is not processed on the device and it is uploaded to the cloud, it means that a majority of the users’ data is exposed, leading them to sacrifice accessibility for privacy.
AI bias is another issue when it comes to accessibility services. Due to the lack of data on individuals with disabilities, AI algorithms can be biased or under-fitted for this demographic. Even though these accessibility models are a good step forward in using AI to enable technology usage for all, there still needs to be many steps taken for safe and responsible AI in accessibility.