Innovation’s highest purpose is removing barriers to care

First, ask the right question

Healthcare innovation often gets framed in terms of speed, efficiency, and new technical capability. Those things matter as we work to remove burden from clinicians and practice staff, and explore the potential of tools such as AI. But some of the most meaningful innovation starts with a more patient-centered question: Who is still struggling to access care, and why?

This question guides us every day. Which is why, at our recent Hackathon event, teams from across the company came together to generate concepts that drive better access and outcomes for providers, patients, payers, partners, and developers alike.

After a week of innovation and collaboration, the team generated a record-breaking 126 unique, creative ideas that can meaningfully impact the way we deliver care. Of them, one finalist stood apart, precisely because of the way the idea aligns with our vision of creating a thriving ecosystem that delivers accessible, high-quality, and sustainable healthcare for all.

This concept was straightforward but powerful: use AI-assisted sign language interpretation to help make telehealth appointments more accessible for patients who are deaf or hard of hearing. It is exactly the kind of idea that reminds us why we innovate in the first place: to help remove barriers that prevent people from fully participating in their care.

Fulfilling the promise of telehealth, thoughtfully

Telehealth has already expanded access in important ways. It can reduce travel burdens, make follow-up care more convenient, and help patients connect with clinicians from wherever they are. But digital access is not the same thing as equitable access. If a patient can technically join a video visit but still struggles to communicate effectively once they are there, the promise of telehealth is still incomplete.

For patients who use American Sign Language (ASL), that communication gap can be significant. Today, access may depend on the availability of a human interpreter. That support is important, but it can also introduce logistical challenges. Interpreter availability may delay an appointment. A new interpreter may need time to get up to speed on a patient’s history or the context of a visit. In highly specialized settings, interpretation itself can become more complex.

This is where thoughtfully applied AI can help. In the Hackathon concept, AI sits between patient and provider as an assistive technology – supporting communication without changing the structure of the visit itself. During the demonstration, the team illustrated how the experience could unfold within the familiar context of a telehealth visit. As the patient signs on, computer vision models interpret movement in real time and generate written captions within the video interface, translating ASL into live captions without requiring a third-party interpreter. The technology works quietly in the background, generating a shared communication layer that supports a direct connection between patient and provider.

In this way, AI does not replace interpretation; rather, it helps make communication more immediate, inclusive, and accessible within the care experience itself. It helps translate the appointment in real time so that a patient can participate more directly in a telehealth visit.

That distinction matters. In healthcare, conversations about AI can quickly become polarized. Some people see only risk. Others focus only on possibility. In reality, the most responsible uses of AI are often the ones that are the least flashy. They are practical. They are assistive. They help reduce friction in moments where communication or workflow barriers get in the way of care. This is one of those cases.

When AI joins the call, the feedback loop must be human

The Hackathon team explored how computer vision and AI models could recognize sign language in a telehealth setting and convert it into text for the provider. Just as important, the concept also pointed toward safeguards: giving the provider a way to confirm with the patient whether the interpretation is correct, and creating a feedback loop so the patient can verify that what they signed was accurately communicated.

That kind of human-in-the-loop design is essential. When AI is used to support communication in care, there must be transparency around what the system is doing and opportunities for humans to correct it. The role of the technology is to assist, not to assume authority.

I think that is a useful way to think about responsible AI more broadly. The goal should not be to remove agency from patients or clinicians; rather, the goal should be to increase it. If technology helps a patient communicate more easily, helps a provider understand more clearly, and preserves human judgment where it belongs, that is a meaningful step forward.

This project also points to a larger truth about accessibility: it cannot be treated as a side feature. Too often in healthcare technology, accessibility is considered late in the design process or narrowly understood as a compliance exercise. In reality, it is central to whether people can engage with care at all. And accessibility means more than one thing. It can mean making patient-facing tools work well with screen readers for people with vision impairments. It can mean improving provider workflows, so clinicians with disabilities can use health IT more effectively. In this case, it means recognizing that communication access is foundational to care access.

The opportunity ahead: inviting the community to collaborate

What makes this particular Hackathon idea especially promising is that it also highlights an important challenge the healthcare industry will need to solve together: data. AI systems improve through training, and training depends on having a “golden data set” of high-quality, representative examples. For sign language interpretation in a healthcare setting, that means not just general sign language data, but data that reflects the specialized vocabulary, context, and nuance of medical conversations. Such a massive project is not something any one company can –or should – undertake in isolation.

There is real opportunity here for collaboration with advocacy groups and with the communities these tools are meant to serve. Building better training data, validating outputs, and understanding where models perform well or fall short all require partnership. Inclusive innovation works best when it is informed by the lived experience of the people it is trying to help.

That may be the most important lesson from this Hackathon project. The future of healthcare innovation will not be defined only by what AI can do. It will be defined by whether we apply technology in ways that are inclusive, transparent, and grounded in real human needs.

Some of the best ideas do not begin with a technology looking for a use case. They begin with identifying and taking aim at a barrier that should not be there in the first place. If we want healthcare innovation to live up to its promise, this higher purpose is what we should continually prioritize.