As governments and technology companies invest heavily in semiconductors, cloud computing and data centers, accessibility is still often treated as a secondary feature rather than part of digital infrastructure. In a written interview with Korea IT Times, Paul Lee, CEO of InnoCaption and Mezmo Technologies, argues that inclusive design determines who can participate in the AI economy and should therefore be built into digital systems from the beginning.
InnoCaption has surpassed 30 million captioned calls and expanded into AI-powered text-to-speech, multilingual captioning and accessible video communication. Lee discusses why human expertise remains essential alongside automated captions, how AI could help people restore their own voices, and why Korea has an opportunity to become a global model for accessible technology as it enters a super-aged society.
Q1. InnoCaption has now surpassed 30 million captioned calls. Looking back on this journey, what has been the most important lesson about how technology can improve not only communication, but also people's independence and quality of life?
The most important lesson is that communication access and independence are deeply intertwined. When we say 30 million captioned calls, what I really see is 30 million moments in people's daily lives: calling a doctor's office, contacting a child's school, or getting through a first-round job interview. These are everyday moments that, for our users with hearing loss, used to mean depending on someone else or avoiding the call entirely.
We commissioned a white paper surveying new InnoCaption users on their calling experience before and after our service: on average they reported a nearly 50% reduction in listening effort and a 55% reduction in listening fatigue, and their emotions around phone calls shifted from frustrated, anxious, and discouraged to confident, happy, and content. These figures, and that shift in emotion, represent the level of impact our service can have on someone's quality of life.
What we build is not a convenience feature, but an integral tool for leading an empowered, independent life.
Q2. Generative AI has dramatically improved speech recognition and real-time translation. Yet accessibility is still often treated as an optional feature rather than essential infrastructure. Why do you believe accessibility should become part of the world's digital infrastructure?
I believe accessibility belongs in digital infrastructure because infrastructure is whatever determines who can participate. We treat roads, power, and broadband as infrastructure because being cut off from them means being cut off from modern life. Communication access works the same way. If digital systems are not made accessible, part of the population is simply excluded. In the US, accessibility is considered a core part of the telecommunications industry, with the FCC regulating and funding the provision of captioning and text-to-speech services to ensure this vital infrastructure includes and benefits everyone.
Infrastructure built for accessibility also ends up serving far more people than intended. Curb cuts, where roads and sidewalks meet, were built for wheelchair users, but soon everyone with a stroller, luggage, or a bike was able to benefit from them. Captions are the same story: built for people with hearing loss, now used by an entire generation. Around 80% of adults aged 18 to 25 watch TV with subtitles on, even though only a small share have any hearing difficulty. Accessibility built as infrastructure ends up serving everyone.
Q3. As AI-generated captions become increasingly common, accuracy and trust are becoming just as important as speed. How does InnoCaption balance AI automation with human expertise, and where do you believe human judgment will continue to play an essential role?
We've lived this balance longer than most. We started captioning every call with live human stenographers—trained court reporters. In 2019, based on user feedback, we became the first in our industry to offer fully automated captioning as an option. The keyword is option. We didn't rip out human captioning and replace it, even though going fully automated would have been more profitable. Each mode has real strengths, so we built the model around user choice, including switching between AI and a live stenographer in the middle of a call.
This is because human judgment stays essential, especially on the hard calls: background noise, crosstalk, heavy accents, poor audio. AI always produces its best guess, even when the audio is too poor to transcribe reliably. A trained human captioner can focus on what is most important and relevant to the user, and can do something AI cannot: let the user know the audio is bad so they can ask the other person to speak up, rather than passing along a guess that may be completely inaccurate.
And by empowering our users with the best of both human and AI captioning, we've built a service whose demand keeps growing. Interestingly, even with significant adoption of automated captions, we have more human stenographers on our platform today than before we introduced them. When AI is deployed to serve users rather than replace people, everyone ends up busier, and better served.
Q4. Governments around the world are investing billions of dollars in AI infrastructure, including semiconductors, cloud computing, and data centers. Should accessibility also be regarded as critical AI infrastructure rather than simply a social policy initiative? Why?
Yes. I understand why accessibility usually gets sorted into the social policy column, but I'd argue the distinction is smaller than it looks. The best social policy and the best infrastructure investments often turn out to be the same thing.
The billions going into semiconductors, cloud, and data centers build what AI can do. Accessibility is part of how those capabilities actually reach people. It is a comparatively small investment, but treating it as infrastructure helps ensure the returns on all that spending are shared across the whole population, including people with disabilities and a fast-growing older population.
And those returns are real. As I described above, communication access leads directly to independence. With it, people stay in the workforce and build careers instead of being pushed out by a barrier technology can now remove. Multiple InnoCaption users have told us how they successfully navigated phone interviews because captions gave them the confidence and ability to communicate effectively without being excluded because of their hearing loss. These are stories about personal dignity, and they are also stories about productive workers an economy would otherwise have lost. When accessibility is built in, people with disabilities and older adults keep contributing their talent and experience. Society gains that productivity instead of paying the costs of exclusion.
Q5. InnoCaption has expanded beyond captioned phone calls into AI-powered text-to-speech, multilingual captioning, and accessible video communication. Which of these innovations do you believe will have the greatest long-term impact over the next five years, and why?
Each of these matters, but the one I believe will have the deepest impact over the next five years is AI-powered text-to-speech, which helps people be heard. This technology serves those who cannot rely on their own voice in a conversation: someone who has had a laryngectomy or throat cancer, someone living with ALS or aphasia, or Deaf individuals who do not voice. Our text-to-speech features let a user type instead of speak, with the app voicing their words naturally on the call and AI helping them keep pace with live dialogue instead of typing out every word. That was our first innovation on this front, and there is more to come.
One area we are working on now is custom voices. Not long ago, text-to-speech meant a robotic monotone. Today's AI voices sound natural and human, and the next step is even more remarkable: helping someone who has lost their voice revive their own, so they can speak as themselves again on phone calls and in virtual meetings. Our hope is that AI will keep expanding what these users can do, until having a speech disability no longer decides whether someone can hold a conversation. Restoring a person's voice is about as meaningful as technology gets.
Q6. As populations age and AI becomes increasingly embedded in everyday life, how do you see AI-powered accessibility benefiting not only Deaf and Hard of Hearing users, but also older adults and society as a whole?
Aging is where accessibility becomes everyone's story. About 1 in 3 people aged 65 to 74 has hearing loss, and nearly half of those over 75. So the technologies we have spent years building for the Deaf and Hard of Hearing community, starting with captioning, will benefit the older population broadly. Hearing loss due to aging brings a higher risk of social isolation, and research suggests this isolation contributes to cognitive decline. That means keeping the lines of communication open does double duty: it preserves day-to-day independence, and it helps people stay connected and cognitively healthy as they age.
But AI can help older adults well beyond captioning, and an area I find exciting is AI moving into the physical world. Human caregiving simply cannot scale at the pace our societies are aging: the OECD projects a shortage of more than 13 million care workers by 2040. Personal AI and robotics can help fill that gap. Picture an older person with a companion that remembers things on their behalf, keeps their medications on schedule, and holds natural conversations that keep the mind engaged and push back on that isolation. A robot that carries heavy groceries home, helps someone in and out of a wheelchair, or guides a person with low vision down an unfamiliar street. The personal level of accessibility assistance that physical AI could provide is incredible to think about.
Q7. Looking toward 2030, what advice would you offer to technology companies, policymakers, and AI leaders who want to build a future where accessibility is designed into digital systems from the beginning rather than added afterward?
For technology companies and AI leaders, my advice is to incorporate the accessibility user journey into product development from the start. AI has made that more achievable than ever: software development is faster, iteration is cheaper, and digital experiences can be customized in ways that were not practical before. We feel this ourselves, shipping features at a pace we could not have imagined a few years ago.
In UX design, you put on the hat of your core user and walk their journey through your product. Let's do the same for users with accessibility needs. Build real personas of people with hearing loss, speech disabilities, or limited vision, walk through your digital service with that lens, and ask where the experience breaks down and how to fix it. I'm the first to acknowledge that this is not always as easy as it sounds. We work in accessibility and even we are constantly striving to do this better. But if industry takes up this mantle in earnest, accessible design becomes a normal part of building and will benefit people in so many ways.
Policymakers should encourage industry to make this effort, while resisting the urge to be too prescriptive. Technology is changing so quickly that rigid rules go out of date almost as soon as they are written. Flexible policy that enables companies to build accessibility into their products will age far better. And if builders do their part, there is less need for heavy-handed rules at all.
Q8. As Korea continues to strengthen its role in AI and digital innovation, what opportunities do you see for the country to contribute to advancing inclusive AI and accessible digital communication globally?
Korea already has a strong accessibility foundation, and I notice it every time I visit: tactile paving along sidewalks and through subway stations, barrier-free buildings with accessible elevators, and mandatory Korean Sign Language interpretation for major government announcements. These are accomplishments many countries have not matched.
The opportunity I see is to extend that same commitment to captioning. Sign language serves an important community, but the population with hearing loss is far larger than the community that signs, and most of them communicate in spoken language. As Korea becomes a super-aged society, that population will grow dramatically. Captioning, on calls, in videos, in public announcements, and eventually in everyday settings, is the accessibility infrastructure that the next generation will need.
Korea has the technology and infrastructure capability to become a leading example showing what an accessible world can look like for an aging society.
Profile | Paul Lee, CEO of InnoCaption
Lee spent the first 12 years of his career in finance and investment, working with Merrill Lynch, FFL and the Ontario Teachers’ Pension Plan across New York, San Francisco, Hong Kong and London. During that period, he also served as a financial adviser to InnoCaption, which was founded by his father.
He formally joined InnoCaption as chief financial officer in 2020 and was promoted to chief operating officer in 2023 before becoming CEO. When Mezmo Technologies was established in 2025, Lee became its founding CEO. He now leads product development strategy across both companies alongside their mission-driven teams.
Lee holds a bachelor’s degree in economics from Cornell University and an MBA from INSEAD in France.
