What if a person with a visual impairment could receive audio assistance reading a map — and detailed instructions on how to navigate their local railway system? Or what if they could use image-to-text technology to quickly discern what’s in their fridge, along with recipe suggestions and a shopping list for their grocery delivery order?

AI-powered tools that do just that are now a reality thanks to Danish startup Be My Eyes, which uses the visual input capability of GPT-4 to create “virtual volunteers” for people who are blind or vision-impaired. It’s just one example of how advancements in AI are transforming the digital accessibility landscape.

The new frontier of digital accessibility

AI has been a hot topic this past year, and for good reason — easy-to-use large language models (LLMs) like ChatGPT have democratized access to this transformative new technology. And it’s become even more accessible with the integration of multimodal capabilities like vision, image creation, and text-to-speech. For example, users can now interact with ChatGPT through voice and receive spoken responses. The use of deep learning algorithms based on neural networks means AI-driven tools like real-time speech-to-text and enhanced predictive text will continue to become even more sophisticated over time, offering seamless support for users with different abilities.

AI is also being integrated with visual tasks and assistive technologies like the work being done to embed AI with NVDA (Non-Visual Desktop Access). Advanced screen-readers will be able to discern what’s most important on cluttered web pages (like e-commerce sites) rather than reading every line. At Miro, we developed tools to communicate spatial information to people with visual impairments, like Miro Assist, which can be used to automatically generate and explain complex visual diagrams.

Website and app accessibility is also improving. While the current iteration of accessibility overlays can actually hinder accessibility and current implementations of LLMs tend to hallucinate when it comes to recommending valid fixes, in the future machine learning algorithms may be able to automatically detect and correct accessibility issues. For example, Facebook previously conducted research that demonstrates the value visually impaired users receive from having access to alternative text for images. 

This is just the beginning. As AI advances, it can streamline various tasks that in turn make technology more accessible. For example, AI has already automated processes like captioning videos, providing audio descriptions, or summarizing websites for screen readers. As these processes become more established, AI will continue to help personalize accessibility features based on user behavior and preferences, offering a more tailored and efficient experience. AI's predictive capabilities could also spot potential accessibility issues during development.

Mitigate potential risks

That being said, there are ethical considerations and potential risks that come with the use of AI in accessibility, especially as it relates to privacy and bias.

Privacy is a major concern, as AI systems often require access to personal data to function optimally. Ensure your company and user data is applied responsibly and securely by adhering to industry-leading global standards and policies.

There’s also a risk of AI systems perpetuating biases, especially if it’s trained on non-inclusive datasets. It's crucial that AI systems are developed with diverse input and tested across a wide range of users so they don't inadvertently exclude or misinterpret the needs of certain groups. At Miro, we use rigorous testing, content filters, and diverse perspective integration to counteract the potential for bias.

Integrate AI and accessibility

AI and accessibility should go hand in hand. This means taking a holistic approach to integrating AI into your accessibility strategy and including accessibility in your company’s broader AI initiatives. Product innovation is a key focus area in Miro’s accessibility strategy not only because it’s inclusive, but because product development seen through the lens of accessibility catalyzes creative solutions. While there’s no silver bullet, you can put this into practice with the right priorities and investments:

• Foster a culture that values inclusivity and accessibility. From inclusive hiring practices that increase your talent pool and reflect the diversity of your customers to keeping accessibility top of mind during every stage of product development, inclusivity, and accessibility should permeate all aspects of your business.
• Invest in training and awareness around both AI and accessibility. Arm employees with knowledge and tools about what accessibility means, how it impacts people with disabilities, and how to think about their work through this lens. Ensure leaders and decision-makers understand not only how AI works, but its implications related to privacy, bias, accessibility and governance. Deep understanding of both will help you make informed decisions about how to wield AI responsibly and inclusively.
• Collaborate with experts. This includes leaders in AI and accessibility as well as diverse users with different needs. Bringing varied areas of expertise together will help you develop more innovative, inclusive solutions that give you a competitive advantage.
• Stay informed about evolving AI technologies and accessibility regulations. As AI tech continues to advance, new laws and regulations are trying to keep up with the rapid pace of change. Whether it’s accessibility regulations, privacy concerns, copyright lawsuits, or sweeping legislation like the EU’s Artificial Intelligence Act, ensure your leaders — and legal team — are closely watching the regulatory landscape.

AI has ushered in a new era of innovation and digital accessibility. With the right safeguards in place, these transformational technologies can empower people with disabilities to gain greater independence and participate more fully in society. Let’s harness AI to create a better, more inclusive future.