Bench Talk

(Source: Andrey_Popov/Shutterstock.com)

Artificial intelligence (AI) has influenced or even revolutionized many areas of our society, and with recent developments such as deep learning or reinforcement learning, this trend will continue to accelerate. These developments are particularly interesting for people with physical disabilities, such as impaired vision, speech disorders or missing or damaged limbs. They open up opportunities for inclusion and participation in society that until recently seemed unattainable for disabled people. Technology can do so much more than just making life easier or more convenient. Instead, the aim is to enable them to do activities that were previously difficult or impossible.

Although assistive technologies for the blind, such as screen readers or braille displays, have been around for some time, AI has recently led to significant improvements. With more powerful speech synthesis algorithms and smartphones and dedicated assistive devices, the use of text-to-speech (TTS) technology has become much more flexible, enabling mobile ad-hoc use, such as listening to an e-book hands-free. This is also closely related to advances in image processing, especially in optical character recognition (OCR), to make text accessible for speech synthesis. A good example of this is Google Lens, which can, for example, recognize and process text in a camera image of a traffic sign. The same logic applies to speech-to-text (STT) apps, which enable mobile and hassle-free text communication without the need for a keyboard or special software, as was the case in the past.

However, a crucial aspect that requires even more attention is the user interface design and, in particular, error tolerance. For ordinary users, the wrong recognition of a command or a failure to load are simple annoyances, but this can be a serious obstacle or even life-threatening for a disabled person if they cannot make an emergency call. An example of this gap in interface design is when a voice interface requires a touch input, which a blind person might not be able to perform at all, in the event of an error.

TTS and STT can also be used to support people with speech and hearing impairments. For people with the motor skills for typing, TTS can give them a voice. One of the best-known examples is the late Stephen Hawking, who used specially designed speech synthesis software to make himself heard. Similar software is now widely used.

Live transcription is an example of STT. This is a popular field of research primarily for commercial reasons, but reliable and flexible transcription also allows deaf people to consume previously inaccessible media. For example, YouTube now enables live subtitles for many languages, making many entertainment and educational resources available to a wider audience.

Although many advances have been made that indirectly help people with impairments, areas of AI research are directly aimed at alleviating the impairment in close cooperation with medical research. One of the most interesting fields of research in this regard is the development of prostheses. Unlike early prostheses, which were completely immobile, mechanically flexible limbs have become commonplace, and recent advances in algorithm design combined with more powerful and compact chips lead to significant improvements in this area. A modern prosthesis reacts much faster to input from the nervous system because of improved signal processing algorithms and adapts appropriately to the environment, for example, to different floor conditions. This allows for much more natural, intuitive movement and reduces the disparity between artificial and natural limbs.

Another traditional approach to improving the lives of disabled people is, of course, physical therapy. In the past, assessing a person’s physical condition was the responsibility of the doctor or therapist. However, with time constraints and incomplete information, it is often difficult to create a tailored, optimized treatment plan. This often results in the use of textbook approaches that do not do justice to the specific situation.

This can be better managed with AI. For example, image recognition software can analyze minor inconsistencies in gait and posture to suggest custom exercises, and progress can be tracked much more closely, allowing for faster improvements. This also leads to greater motivation, as the person being treated achieves success faster and plays a more active role in setting goals and providing feedback on what is working and what is not.

Many AI technologies already exist that improve the lives of disabled people, either as by-products of regular technologies or that are designed specifically for them. However, many technological challenges still prevent AI from having a greater impact on this large and particularly vulnerable segment of the population. To make significant progress toward inclusion, policymakers and companies must focus more on taking disabilities into account during product development and, ideally, also include disabled people in the development process.