The Impact of AI on People with Disabilities

Artificial intelligence (AI) is rapidly advancing, poised to significantly influence how we interact with both digital and physical environments. As we aim to build a more equitable world, it's crucial to consider how this development impacts accessibility and disability.

AI involves programming machines to emulate human cognitive functions such as learning, problem-solving, perception, decision-making, and language understanding. These technologies present both significant opportunities and risks for people with disabilities.

In part one of a three-part series on AI and accessibility, I’ll explore how AI technologies affect the daily lives of people with disabilities, including the accessibility benefits and the potential barriers introduced by AI implementations.

How AI is already advancing accessibility

Currently, AI is employed in various applications that enhance accessibility, such as generating captions and transcripts, supporting chat functions, creating alternative text, and providing object, sound, and face recognition. Two particularly notable implementations are the Seeing AI app by Microsoft, which provides AI narration of a scene, person, or surrounding objects (among other features), and the Be My Eyes volunteer service, which offers an AI-powered feature to assist blind individuals in recognizing objects and text, including screenshots and interfaces.

On platforms like Apple and Facebook, AI enhances image searching, text extraction, and content translation, and provides alternative descriptions for images, enriching access for users.? For example, on iOS, AI is used to help blind users know what the photos they have taken contain and to help blind people take pictures by providing guidance on the visual scene.

AI combined with glasses or a mobile device can overlay information about the environment to users with low vision, supporting navigation, providing recognition of people, and identifying and reading text.

AI-powered captions and transcripts have allowed people who are deaf or hard of hearing to access spoken information in real time in situations where an interpreter or captionist is not available. While this technology is not perfect and does not always provide the same level of accuracy as a human, its widespread availability offers solutions at scale and creates access where there was none. AI captions and transcriptions can also form the foundation for more accurate human-corrected captions and transcriptions.

The risks AI poses for people with disabilities

Despite its benefits, it’s important to note that AI can also introduce potential biases and discriminate against individuals with disabilities. For instance, the U.S. Department of Labor has issued warnings about AI's use in employment screening processes, which might inadvertently exclude people with disabilities based on certain traits or characteristics. Additionally, in early AI research, autonomous vehicles failed to recognize wheelchair users in crosswalks in some situations.

Beyond biases, AI can also misinterpret and confuse the unique accessibility needs of people with different disabilities. AI models sometimes generate responses that appear accurate and unbiased but may, in fact, be erroneous. For example, I’ve had a few conversations with AI-powered chatbots in which these bots indicated that closed captions would benefit users with low vision.? While captions can certainly benefit a broad range of users, these experiences suggest that AI may not fully understand what features are needed for different people. ?Transparent labeling of AI-generated content and clear notification of AI's involvement are becoming essential practices to mitigate confusion and ensure reliability.

Frameworks like the Blueprint for an AI Bill of Rights by the U.S. White House illustrate how risk models can be applied to AI, assessing where the technology can be deployed safely, particularly in contexts involving people with disabilities. Similarly, the recent EU AI Act addresses the balance of risks and benefits associated with AI use.

The path forward: AI must enhance, not replace, existing practices

The future of AI in creating accessible solutions and serving as assistive technology is not a question of possibility—it's already happening—but rather of timing and implementation. Advocates for people with disabilities are justifiably concerned that premature abandonment of current accessibility efforts in favor of unproven AI technologies could lead to gaps in accessibility. Historical precedents— where technology transitions have left temporary voids in accessibility — underscore the need for caution.

Thus, while AI continues to evolve, it is crucial to balance its integration with ongoing accessibility practices. We must manage the risks and impacts carefully: lower-risk applications might be expanded more freely, but higher-risk scenarios will require stringent oversight. The maxim "do no harm" must guide the transition to ensure that existing effective practices are not discarded prematurely. The overarching concern is that the allure of future AI solutions might tempt a shift away from proven methods before the new technology is fully ready and capable of providing equal or enhanced access. This cautious approach will help safeguard against regressions in accessibility as we embrace the potential of AI.

I’ll explore this potential in the second part of this series, which will focus on how teams can harness AI to create and maintain accessible digital experiences.