What I Do - James Anibal
I have a joint appointment for AI research between the University of Oxford and the NIH Clinical Center under the NIH Oxford Cambridge Scholars Program. The program sends researchers to Oxford and Cambridge for a PhD. I’ve been studying at Oxford for three years.
What inspired the work that I’m doing was the realization that just because artificial intelligence is high performing on paper doesn’t make it useful in the real world. I think voice AI has tremendous potential to change that. But even now, I think it’s being done in the same way as always, with sound booths and premium technology—this image of AI that doesn’t translate to the real world.
My research seeks to provide a more scalable solution, what I call multimodal audio data for health. We built a web/mobile application broadly themed around the idea of, “Tell me about your health.” We’re trying to reimagine voice AI into communication AI, where we understand all of the information that’s captured. We capture the inherent complexity of spoken human communication, which contains three parts: 1) voice, the way a person sounds; 2) speech, the way they put together their words; and 3) the meaning of the words that they say. All three contain information, which may relate to health if collected in the right way.
This data can reveal potentially important information about health conditions. For example, a specific type of voice sound paired with patient-reported symptoms may be a more robust biomarker, or indicator, of COVID-19 than just a recorded cough.
I look at deep learning as an exploratory process guided by a question. For example, if our objective is to screen for a certain range of infectious diseases, we don’t know what biomarkers necessarily will associate with those diseases. But we know what we want: We need the clinical question guided by experts and we need the data. The idea is to link those two with new, higher-order biomarkers that our AI models are able to discover.
To me, working in public service at the NIH is a pathway to building clinical AI systems that not only offer health benefits but are ethically designed to respect patients, privacy, individual rights, and democratic values.
The NIH is uniquely positioned in terms of funding and the ecosystem to build impactful technology that has far reaching implications. A great example that preceded our project is the Bridge2AI Initiative through the NIH Common Fund, which has allocated millions of dollars for the development of ethical, sustainable AI technology. One subset of that is the Voice Data Generation Project. We collaborate very closely with that group. They’re doing amazing work in the space of just understanding the fundamentals of voice biomarkers and identifying what’s even possible. We’re working to translate that into real-world environments.
Interview by Sean Markey. Photograph by Maria Maslennikov.