2026 HPM Healthcare Conference

AI with a Soul: Guardrails, Access, and the Future of Healthcare

This year’s 2026 HPM Healthcare Conference was held on Friday, April 17th at The Forum on 125th street and Broadway. The morning began with a welcome from HPM Interim Chair Larry Brown and was followed by the introduction of Keynote Speaker Renée Cummings by student and conference volunteer Aspen Schwind MHA'27. Professor Cummings began by expressing her joy in returning to Harlem, having lived nearby in the 90s. Her work as a criminologist first introduced her to AI through risk assessment algorithms, which she described as creating "zombie predictions" as they overestimated the risks of black and brown children and poor defendants. When she asked herself what was wrong with these assessments, she came "face to face with the word algorithm." It was then that Cummings began to understand that the data used to create these assessment tools were "traumatized data sets...that created an extraordinary amount of trauma in the neighborhoods [they] were serving." She explained that her work in tracing the history of this data as a criminologist is what led her to become a data activist. As Cummings studied the challenges that data was creating for specific communities, she began to think about how we could reimagine our relationship to this information and the tools that we use to organize it.

Cummings isn't new to Columbia. Seven years ago, she was part of a community scholars’ program, researching the relationship between digital forces, AI, algorithms, and populations returning home from incarceration- with an interest in how surveillance technology was impacting communities in Upper Manhattan. She is now the Data Activist in Residence at the University of Virgina School of Data Science. 

In thinking about “AI with a Soul”, Cummings began with the depth of consciousness that is required to interact with this technology. She defined AI as, "the simulation of Human Intelligence in machines enabling them to perform tasks like learning, reasoning, problem-solving, and adapting, based on data". She then explained that highly autonomous AI systems or "superintelligence" is where many industries have set their goals for technological expansion. 

Cummings went on to emphasize that, "AI cannot think beyond the data. Data can encode and perpetuate past injustices, turning yesterday's bias into today's algorithmic harms." She stated that the quality, depth, and diversity of data sets the limits of what AI can do. She gives Large Language Models underperforming indigenous dialects as an example of the kind of harm that comes from lack of representation in data. As data shapes our reality, the historical biases that go into algorithms will likely reproduce injustices at "size and speed."

So how do we design AI with soul? Cummings explained that we must understand how injustice is automated into our data sets and how AI both amplifies and hides inequities. Those who generate data are rarely the ones that own and extract it- institutions benefit while patients are left exposed. She highlighted "data colonialism" as a critical term, stating that vulnerable communities have always experienced data differently as they are mined for their information and often exploited with no return.

To incorporate a more “soulful” AI into healthcare, Cummings urged that this technology must be based in ethics rather than just efficiency. She advises us to build for performance and people, govern by values and not just velocity, and deal with the invisible challenge of what is missing in the data. What does it mean to be an ethical leader in the age of AI? Cummings says we must ensure that the technology is human-centered from the point of data collection with bias in mind at every stage. Each system must be designed with equity, transparency, and accountability built in. Humans then must perform "wellness checks" on AI periodically to ensure that the technology continues to properly collect, encode, and aggregate data in accordance with its carbon-minded programming. Professor Cummings took questions from conference attendees before concluding her time at the podium.

After a short break, student and conference volunteer Sabah Bari MPH'26, introduced our afternoon panel, which included Deepak Chandra, Dr. Remle Newton-Dame, and Divya Pathak. HPM Professor Thalia Porteny moderated the conversation, first asking the panelists for concrete examples of AI at each of their respective organizations.

Divya Pathak, formerly at New York Health + Hospitals and now Senior Vice President and Chief AI Officer at Regeneron, pointed to AI’s ambient listening leading to less gaps in patient care. “Better documentation leads to better quality of clinical notes and better coding of services provided." She gives biotech as another example, with AI reducing the lift of clinical trials and increasing access as it accelerates drugs to the market. Language translation is another area of AI implementation. Within the diverse population of New York City, 72 languages are currently supported by New York Health + Hospitals' technology, allowing for better communication with patients.

Dr. Remle Newton-Dame, Assistant Vice President of Healthcare Analytics in New York City Health + Hospitals’ Office of Population Health, spoke about including safety net patients in the AI conversation, people who have historically experienced bias and therefore have gaps in trust. She states that though patient perspectives on AI are different, most want AI to increase access to care and improve the quality of doctor visits, not replace doctors. She emphasizes using AI to leverage providers’ time: programming automated response technology for answers as simple as “I’ll get that for you right away” can free up nurses to call patients and explain treatment plans for managing illnesses or clarifying diagnoses. 

Deepesh Chandra, Senior Vice President and Chief Digital and Information Officer at Montefiore Einstein, comes from an academic medical center, which serves a large population of patients in the Bronx and is also a leading medical school. Because of this, he states that his institution has a responsibility to create the next generation of care delivery and science for world to follow. At Montefiore, AI implementation currently falls into three buckets: co-pilot mode (assisting with clerical work), co-worker technology (“silicon workers alongside carbon workers”), and areas that will push science forward (within operating rooms, clinical rooms, and in training students). Chandra gave us an example of the Edu version of ChatGPT, which offers one to one tutoring to every medical student. How does it work? The AI program presents itself as a patient and allows the student to start asking questions as the medical provider. At the end, the AI gives the student an assessment of how well they did or what they may have left out.  

Moderator Porteny then asked the panel about challenges and ethical dilemmas around AI. Newton-Dame spoke to the difficulty of incorporating AI into delicate encounters where information is purposefully left out of clinical notes for the patient's safety, like reports of domestic violence or adolescent medicine. Pathak offered the example of regulatory intelligence, “keeping up with evolving guidelines is a huge challenge." She also stated that it’s a “huge problem” that people who have little to no experience writing code can build apps. Democratized AI brings security vulnerabilities that go against HIPAA patient regulations and data privacy. “We need to be proactive with guardrails before doling tools out. There’s an enthusiastic workforce that wants to leverage AI, but there’s so many vulnerabilities within organizations."

Professor Porteny closed the panel with the following question: what can students do to meaningfully engage with tools now in both deployment and development of AI technology?

Pathak says she views AI to be “more like oxygen, it’s already there- not an option.” She stated that adopting the mindset of AI fluency within our culture shouldn’t be an option.  Instead, AI must be embedded in everything we do on a regular basis. She urged students to build intellectual curiosity around AI tools and how it could impact their future workflows.

Newton- Dame agreed that AI is “going to be in everything,” but that people should still be able to explain their work. Her advice is to pick up a coding language and additional context to be able to properly evaluate the AI. “Look at the output and figure out where you need to drill in more”.

Chandra recommended reading up on two recent reports: the World Economic Forum’s “AI at Work” and McKinsey’s “State of Organizations”.

Before the panel took questions from the audience, Pathak offered one more critical point: “Build the workforce to be more AI fluent and more AI responsible”.

After the panel, conference goers were led to the Forum’s West Atrium for the reception, which included lunch and an inaugural Student Poster Showcase.

Attendees were able to eat, mingle, and view the students’ research at their leisure. HPM’s social media leads and the Mailman Journalism Club collaborated with conference volunteers by interviewing students, professors, and speakers about their views on AI. The day concluded with Professor Michael Sparer announcing raffle winners. Attendees with the winning tickets cheered and picked from a variety of HPM swag prizes.

Overall, it was a fantastic day of learning and connecting. Join us for more at next year’s conference on Friday, April 16, 2027.

Photos by Ellen Barroso