Bridging communication gaps with AI-powered accessibility
Access to information shapes whether people can participate fully in education, public services, culture, transportation, and everyday life. Yet for many people with disabilities, essential information is still delivered in formats that are difficult or impossible to use.
During the session AI and information accessibility: Bridging communication gaps for all at the AI for Good Global Summit 2026, Kevin Lee, Founder and CEO of EQ4ALL, explored how AI can help address two accessibility challenges: providing information in sign language and transforming complex written content into Easy Read formats.
Drawing on examples from Korea and the United States, Lee showed how AI can support more accessible communication across public services and digital media. His central message was that AI should be assessed not only by productivity, but also by whether it enables more people to access and understand information.
“Equal access to information is the starting point of equitable opportunity,” he said.
Making digital information available in sign language
Lee began with sign language accessibility, noting that around 430 million people worldwide require rehabilitation for disabling hearing loss, according to the World Health Organization. There is no official global count of deaf people, but the World Federation of the Deaf estimates that more than 70 million belong to signing deaf communities. For many sign language users, spoken and written languages may not be their first language, which can make text-heavy digital information difficult to access.
Simply adding written captions does not always remove the barrier. In some cases, information needs to be presented directly in sign language.
Lee described a workflow in which written text is processed through an AI-powered sign language translation system. Because the technology is not yet fully accurate, sign language users review and revise the output before the final version is delivered.
Human review remains essential. Sign language involves more than hand movements or vocabulary. Facial expressions, body movements, and other non-manual signals also contribute to meaning, making the creation and annotation of high-quality training data particularly demanding.
Lee explained that producing training data for sign language can take considerably more time than preparing datasets for written-language translation. Limited data availability remains one of the main barriers to improving the accuracy and scalability of these systems.
Combining AI with human expertise
Rather than presenting AI as a replacement for sign language professionals, Lee described a model in which technology and human expertise work together.
As more translated material is produced, previously reviewed sign language content can be converted into training data. This can improve the translation engine over time, reduce the revision required, and lower the cost of producing accessible content.
He also highlighted the importance of representation in digital sign language avatars. Diversity in avatars can help ensure that the technology better reflects the communities using it.
Bringing accessibility into public spaces
Lee presented several examples of how AI-generated sign language is being applied in practical settings.
In public transportation, emergency announcements may be delivered primarily through audio. Deaf passengers can therefore miss critical instructions, including evacuation notices. He described a system used on the Seoul metro in which screens display sign language information when an emergency announcement is made.
At Incheon Airport, a pilot project was designed to make gate changes and other time-sensitive updates available in sign language.
Sign language avatars have also been integrated into barrier-free kiosks, helping users navigate self-service interfaces. In museums and galleries, avatars on digital devices can provide accessible explanations of cultural, scientific, and historical content.
Supporting communication in everyday life
Lee also described a television-based application designed for deaf children and their families. More than 90 percent of deaf children, he explained, are born into hearing families. When family members do not share a common sign language, children may experience isolation within their own homes.
The application uses television because it is a screen that family members can use together. By supporting shared sign language learning, the technology is intended to strengthen communication between deaf children and their hearing relatives.
Lee also discussed ways to make online information more accessible. Sign language content can be delivered through webpages, tablets, QR codes, text-message links, and other digital formats.
One example involved QR codes placed on food packaging. Scanning the code can provide information in sign language about how a product should be stored, prepared, or consumed.
Making emergency information accessible
Emergency alerts illustrate why accessibility must be considered from the outset.
Mobile and television alerts are often distributed primarily as text. For sign language users who have difficulty reading written information, this can limit access to urgent warnings.
Lee described several approaches used in Korea. In one case, sign language avatars appear through television set-top boxes when emergency alerts are issued. Other projects display emergency information through large public screens or send mobile links that open sign language versions of the warning.
These examples show that accessibility is not an additional feature to introduce after information has been created. It must be integrated into the channels through which essential information is delivered.
Simplifying complex written information
The session then focused on Easy Read, a format designed to make written information clearer and easier to understand, supporting people with intellectual and developmental disabilities, people with low literacy, and those reading in a language that is not their first.
Traditionally, Easy Read content has been produced manually by specialists who rewrite complex information sentence by sentence. Despite its value, only a limited amount of content is currently available in Easy Read because producing it can be costly and time-consuming.
Lee explained that large language models can help accelerate this process. Historical examples of standard text and their Easy Read versions can be used to train systems that convert complex material into simpler language. Human review and continuous quality improvement remain part of the workflow.
The same approach can extend beyond written text. Lee presented an “Easy Cast” format using AI-generated dubbing to allow audiences to switch between an original commentary and a simplified version.
Designing AI for inclusion
Lee concluded with examples of how sign language tools can support participation in digital life. EQ4ALL has created sign language GIFs that people can use in social media posts and messages, allowing sign language users to communicate online without recording a new video each time.
The company has also developed digital sign names, reflecting the fact that many sign language users have both a written name and a sign name.
Together, the examples presented during the session demonstrated how AI can expand access across many contexts, from emergency warnings and public transportation to family communication and social media.
The broader opportunity lies not only in translating more information, but in recognizing that people engage with information in different ways. When accessibility is built into digital services, AI can help reduce communication barriers and enable more people to participate in society on more equal terms.
