At a recent gathering of AI innovators, researchers, and policymakers, a phrase cut through the noise of demos and data: “We are not going to deploy AI for the sake of deploying AI in healthcare. It has to be significant in increasing capacity, standardization, and quality of care.”
It is the kind of statement that sounds obvious on the surface. But it is also one that many organizations struggle to live by. The pressure to be seen as innovative, to ship products, to attract funding, all of it creates incentives to move fast and worry about real-world impact later.
We do not build that way at Ethnomet, and conversations like the ones we heard at this conference remind us why.
The Trust Gap is the Healthcare Gap.
One of the most compelling frameworks discussed was what researchers are calling the “trust paradox” in AI adoption: people’s willingness to use AI-enabled tools consistently outpaces their trust in those tools’ underlying capabilities. In other words, people are trying AI even when they are not sure they believe in it — driven by necessity, curiosity, or the absence of better options.
In healthcare, that paradox has consequences. A patient using an AI triage tool they do not fully trust may dismiss its guidance. A community health worker relying on a platform they cannot explain to their patients may lose their patients’ confidence altogether. A physician working with an AI diagnostic aid they cannot interrogate for reasoning may develop either blind dependence or reflexive skepticism — both dangerous.
The research presented was clear on what moves the needle: transparency about what an AI system can and cannot do, consistency in how it behaves and what values it upholds, and responsiveness to concerns when things go wrong.
These are not abstract principles. They are the design requirements we hold ourselves to across every Ethnomet product.
Accessibility is not a Feature— It is the Point.
The conference also surfaced something we think about every day: the danger of building AI tools that speak a language — literally or figuratively — that the communities most in need cannot understand.
A presentation on culturally inclusive AI for underrepresented languages made the case that most AI systems were designed by, for, and in the context of a handful of dominant cultures and languages. Even when these systems are extended to cover other languages, they carry embedded assumptions — about how knowledge is structured, what counts as credible, how decisions should be communicated — that can make them alienating or outright wrong for the people they claim to serve.
This resonates deeply with our work in Nigeria. When we bring AI-assisted healthcare to remote and rural communities, we are not simply translating an existing product. We are asking ourselves who was in the room when this was designed, what assumptions are baked in, and whether those assumptions hold for a grandmother in a village in Benue State or a community health volunteer operating in a region with limited connectivity.
The answer, more often than not, is that the assumptions do not hold — and that doing this work properly requires going back to first principles.
Care is a Continuum, and so Should be AI.
A panel on AI across the care pathway described a growing consensus: the biggest gains from AI in healthcare do not come from isolated breakthroughs — a faster diagnostic tool here, a smarter triage system there — but from connecting those capabilities so that a patient moves smoothly from risk detection to diagnosis to treatment without falling through the gaps.
This is the vision embedded in how Ethnomet thinks about its product suite. The EthnoRing monitors vital signs continuously. The Ethno Virtual Care Box brings diagnostic capacity to communities without clinics. The Ethnomet Vital Connect platform links patients to physicians for real-time consultation. The Digital Healthcare Platform supports the providers coordinating care behind the scenes.
Each of these is useful on its own. Together, they form something closer to what healthcare should look like: continuous, connected, and built around the patient rather than around institutional convenience.
The panel noted that one of the biggest barriers to realizing this vision is fragmentation — AI tools deployed in silos that cannot share data or coordinate handoffs. We are building against that tendency from the start.
What Trustworthy AI Actually Requires.
One speaker drew a key distinction: trust matters most in situations of vulnerability. When stakes are low, we do not need to trust — we can experiment, tolerate failure, and move on. But healthcare is almost never low-stakes. A misdiagnosis, a missed alert, a recommendation that does not account for the patient’s actual context — these carry real costs.
Building AI for healthcare therefore requires more than technical accuracy. It requires:
Transparency: being honest with patients, providers, and communities about what the system does, how it makes decisions, and where it can fail.
Consistency: behaving reliably across contexts — not performing well in trials and then degrading in the field, not applying different standards to different communities.
Responsiveness: listening closely to concerns and incorporating what we learn into a continuous improvement loop.
Human oversight: keeping clinicians and community health workers genuinely in the loop, not just nominally. The data presented at the conference showed that 78-79% of respondents across North America consider human oversight of AI systems essential — and 60% insist this should apply regardless of how low-risk the task appears.
These commitments shape how we train our systems, how we work with partner communities, and how we think about accountability when something does not work as intended.
The Real Benchmark is the Patient in Front of You.
A presentation on AI-augmented point-of-care ultrasound described a team in year three of bringing portable diagnostic tools to rural and Indigenous communities in Alberta — navigating real-world complexity around connectivity, community buy-in, clinical integration, and iterative improvement. The lesson was not that AI makes this easy. It was that with the right commitment and partnerships, it becomes possible.
We recognized ourselves in that story. The Ethnomet model was built for exactly this kind of context — where the last mile is long, where trust must be earned community by community, and where the measure of success is not the accuracy of an algorithm in a controlled trial but the quality of care experienced by a patient who previously had no access at all.
That is the benchmark we hold ourselves to, and it is why we keep building.