Manas Bedmutha | HXI - Human-centered eXtended Intelligence

Simulated Patients for Clinical Communication Training

Mon, 14 Sep 2026 00:00:00 +0000

Overview

How a clinician talks to a patient is part of the care, not a skill layered on top of it. Communication quality determines whether trust is built, whether a patient discloses what matters, and whether bias is enacted in the room. While that much is established, the opportunity to practice a difficult conversation, and to learn afterwards how it went, is scarce for clinicians and trainees alike.

Our UnBIASED work established that these dynamics are measurable: dominance, warmth, engagement, interactivity, and turn-taking can be extracted from real clinical conversations and modeled against how the interaction was experienced. This project turns that measurement into a training platform. Large language models and expressive speech synthesis drive simulated patients that behave like people rather than assistants: they hold information back, and their affect shifts in response to how they are treated, so trust has to be earned before a patient’s underlying needs surface. Each rehearsed conversation is analyzed with the UnBIASED pipeline and returned through ConverSense feedback, pairing time-aligned visualizations of the interaction with prompts that ask the learner to interpret their own behavior.

ConversHIVe applies the platform to HIV care teams, where implicit bias linked to sexual orientation and race affects access to care for people who have already faced stigma, and where clinics and community-based organizations work under limited time, staffing shortages, and turnover. Its second phase, ConversHIVe-Live, embeds the training in routine HIV service delivery.

EmpathIQ applies it to pre-clinical medical students, where empathy declines during training precisely as students meet the emotional load of clinical work, and where standardized patients and faculty-led workshops are effective but too resource-intensive to scale. EmpathIQ identifies the markers that distinguish empathic communication, varies patient affect within a single encounter, and evaluates whether the resulting feedback changes measured empathy and compassion.

Funding and External Collaborations

ConversHIVe is funded by the California HIV/AIDS Research Program (CHRP) under its Low Barrier Technology Interventions for HIV Prevention and Care program, with a second phase supporting real-world implementation. It is a collaboration between the HXI Lab and the Owen Clinic at UC San Diego, the AntiViral Research Center (AVRC) Community Advisory Board, and San Diego community-based organizations including Christie’s Place, San Ysidro Health Center, and Father Joe’s Villages.

EmpathIQ is supported by a Sanford Research Fellowship from the T. Denny Sanford Institute for Empathy and Compassion at UC San Diego, awarded to Manas Bedmutha, and is conducted with Drs. Lisa Eyler and Federica Klaus of the Sanford Institute and with Dr. Charles Goldberg, Associate Dean for Graduate Medical Education.

UNDERSTAND: Uplifting the New generation through DBT Education and Resilience for Social Triggers, Anxiety, Negativity, and Depression

Fri, 11 Jul 2025 00:00:00 +0000

Overview

Many college students experience depression and anxiety together, which compounds both symptom severity and functional impairment. Treatments like CBT and DBT work, but they are delivered on a schedule. The skill a student needs is taught in a session on Tuesday and required at 2am on Saturday, and the gap between those two moments is where the intervention fails.

UNDERSTAND closes that gap with three components.

Sensing. Wearables and phones carry signals that precede and accompany distress: physiological arousal, sleep and activity disruption, changes in social behavior and interaction patterns. We instrument these continuously and in daily life, so that the system is reading the situation as it develops rather than reconstructing it afterwards from self-report.

AI. Machine learning models turn those raw streams into an estimate of when someone is struggling and when they are receptive, which is the harder of the two questions. Large language models then generate the support itself, and our work constrains that generation so the model stays inside the skills and principles of DBT rather than producing generic encouragement that sounds therapeutic without being it.

Just-in-time adaptive intervention. The decision of what to deliver, when, and whether to deliver anything at all is the intervention. JITAI makes that decision continuously from the sensed context, offering a DBT skill at the moment it applies, and staying silent when an interruption would do more harm than good.

By targeting transdiagnostic mechanisms such as emotional dysregulation, interpersonal difficulty, and cognitive distortion, the approach is designed to generalize across populations with comorbid mood and anxiety disorders rather than to a single diagnosis.

Funding and External Collaborations

UNDERSTAND is funded by the National Institute of Mental Health (NIMH) at NIH, with the funded project starting 1 August 2026. It is a collaboration between the HXI Lab and the UC San Diego Department of Psychiatry and the Herbert Wertheim School of Public Health, bringing together expertise in DBT, clinical psychology, digital mental health, and ubiquitous sensing.

Student Mental Health and Well-Being: Research with WILLO

Mon, 14 Sep 2026 00:00:00 +0000

Overview

WILLO is UC San Diego’s well-being platform, available to every student on campus and run by the university. The HXI Lab does not operate WILLO. We collaborate with the teams behind it, and this page collects the research that has come out of that collaboration.

The standard response to student well-being needs is to add clinical capacity: hire more counselors, shorten the wait. Our work argues the binding constraint sits elsewhere. Campuses already hold far more resources than students use, and the students least likely to find them are often the ones who would benefit most. The problem is therefore discovery and connection rather than supply, and that claim is the through-line across everything below.

Three strands follow from it.

Reach and discovery. If resources exist but go unused, the design question is how a student encounters the right one at a moment when it is useful. We study this both at the platform level and inside clinical workflow, including EHR-embedded prompts that let a provider activate holistic campus resources during a visit rather than referring a student into a separate system they may never enter.

Willingness to share data. Any well-being technology that adapts to a student can only use data that student agrees to give. We study where students draw that line and why: which sensing modalities feel acceptable, how trust in the institution governs those choices, and what that implies for anyone building a system a student population would actually adopt.

Deployment at campus scale. WILLO is deployed to the entire student body, which makes it possible to study engagement and reach in the conditions where they actually matter, rather than in a study cohort assembled for the purpose.

The same questions drive the lab’s work on just-in-time adaptive interventions in UNDERSTAND.

Collaborations

This research is conducted in partnership with the Joan & Irwin Jacobs Center for Health Innovation (JCHI) at UC San Diego Health, our principal collaborator on this work.

It also involves UC San Diego Student Health and Well-Being and Counseling and Psychological Services (CAPS), together with the WILLO team.

CSE 217

Sat, 26 Oct 2024 00:00:00 +0000

HC4H - Human-Centered Computing for Health (Spring Quarter)

Background

The advent of new mobile and ubiquitous computing technology (tablets, smartphones, tracking devices, depth cameras, wearable devices, augmented reality devices, etc) has created new opportunities to design novel solutions that bring innovation to health and healthcare. The health and healthcare domains, however, are extremely challenging to research and develop technology for. Numerous regulations exist for the protection of patients as well as health data which can impact how easily new technologies might be used. These rules, such as the Health Insurance Portability and Accountability Act (HIPAA), come from federal administrations, such as the Food and Drug Administration (FDA), as well as institutional bodies, such as human subject protection programs and Institutional Review Board (IRB).

Nevertheless, it is possible to thoughtfully design technologies that improve healthcare experiences and address real health problems. Human-Centered Computing has the potential for clear and important impact - enhancing the workflows of healthcare professionals and improving the health of all people.

In this class students will be exposed to the health domain at large through presentations, remote visits and discussions with experts in emergency rooms, trauma rooms, operating rooms, radiology clinics, sleep clinics, outpatient medical offices, the Simulation Training Center (STC), the Professional Development Center (PDC), the Center for the Future of Surgery (CFS), the Exercise and Physical Activity Resource Center (EPARC), and the West Health Institute.

The HC4H class will first learn about health regulations and human protections, covering both the legal mandates and ethical implications of working in and around health. We then embark on healthcare virtual visits to experience first hand the front lines of modern health and healthcare. The class will conclude with the creation of design proposals that offer a specific solution to address a specific problem relevant to HC4H. Students are invited to apply cutting-edge interactive technologies that are currently being used (or could be used in the near future) to support their proposals. Example technologies include but are not limited to Azure Kinect, Google Glass, Fitbit, Withings, Apple Watch , Microsoft HoloLens, etc.

In groups, students will create and deliver a visual prototype where functionality of the proposed solution is demonstrated through mockups, videos, a website, and a final presentation. Based on both feasibility and refinement of proposal, students may be offered the opportunity to continue their work in a collaborative research project after the course concludes. Independent research credit can be provided in subsequent quarters as CSE 198 / CSE 199 (for undergraduate students) or CSE 293 / CSE 298 / CSE 299 (for graduate students).

Course Description

HC4H an interdisciplinary course that brings together students from Engineering, Design, and Medicine, and exposes them to designing technology for health and healthcare.

The course is focused on studying how technology is currently used in healthcare and identify opportunities for novel technology to be developed for specific health and healthcare settings.

Successful students in this class often follow up on their design projects with actual development of an HC4H project and its deployment within the healthcare setting in the following quarters

More information

See CSE 217 (Human-Centered Computing for Health) on Canvas: https://canvas.ucsd.edu

UnBIASED: Understanding Biased patient-provider Interactions And Supporting Enhanced Discourse

Thu, 23 Sep 2021 00:00:00 +0000

Overview

Healthcare bias, based on patients’ race, gender, sexual orientation, and other factors, leads to health disparities, such as lack of appropriate treatment and inadequate pain support. Such biases are often unintentional and “hidden” in communication between patients and doctors.

Existing approaches to address hidden bias are limited because they are removed from actual patient-doctor interactions in which bias hides. Technology offers an opportunity to design new approaches that can make den bias more visible and thus addressable.

We are investigating a new approach to address hidden healthcare bias by improving patient-doctor communication in primary care. This approach monitors body language for signs of bias and provides feedback to raise awareness of patients and doctors for opportunities to adjust their communication style.

We are partnering closely with patients and doctors to ensure this approach is guided by their experiences and needs. Through this collaborative effort, we expect to gain a deep understanding of how hidden bias is experienced and how we can address it better in the future.

More Info here: http://unbiased.health

ConverSense

Measurement is only useful if a provider can see it. ConverSense turns the pipeline’s output into feedback: a web application that visualizes how dominance, interactivity, engagement and warmth moved through a visit, and across visits over time. A study with five clinicians across ten patient visits showed the feedback lands only when it is tied to the specific interaction it came from, which is why the visualizations are time-aligned to the conversation rather than reported as scores.

SocialLM

Assessing communication at scale is the bottleneck. SocialLM asks whether large language models can track social behaviors directly from clinical transcripts without fine-tuning, and finds that they can, but unevenly: performance varies by patient race and by segment of the visit. Because that variability is itself an equity problem, the work introduces an agreement-weighted ensemble that improves both accuracy and stability, giving a practical route to social signal tracking at scale.

Funding and External Collaborations

UnBIASED ia a 5-year project, funded by the National Library of Medicine (NLMR01LM013301), and it is a collaboration between the University of Washington and the HXI Lab at UC San Diego. Our ultimate goal is to create tools to support patients and the next generation of doctors to have bias-free interactions that promote healthcare access, quality, and equity.

Designing Smart and Autonomous Vehicles

Thu, 25 May 2023 00:00:00 +0000

Overview

Adoption of autonomous vehicles is held back less by the driving than by the passenger’s judgment of it. Trust has to be calibrated: too little and the technology goes unused, too much and it is relied on where it should not be. We study that calibration at the level of the individual, using virtual reality, biometric measurement, human-centered design, and data science, rather than treating drivers as a single population.

Trust is personal. A survey of 1,457 young adults, modeled with machine learning and SHAP, showed that perceptions of AV risks and benefits, attitudes toward feasibility and usability, institutional trust, prior experience, and mental models predict trust, while psychosocial traits and driving style contribute far less.

Explanations carry that trust, and they can fail. In a simulated driving study with 232 participants, errors in an AV’s explanations reduced comfort in relying on the vehicle, confidence in its ability, and explanation satisfaction, even though the driving itself was identical. Perceived harm and driving difficulty amplified the damage, so the contexts where an explanation matters most are where getting it wrong costs most.

Studying any of this requires observing the driver. DriveSimQuest is a VR driving simulator and research platform on the Meta Quest Pro, capturing gaze, facial expression, hand activity, and full-body gesture in real time, so that studying a driver’s affective state is a matter of designing the study rather than building the rig.

The same setting looks different when the driver is a person. Fleet drivers are involved in collisions that impose severe financial costs and endanger lives, and fleet companies rely on one-to-one coaching to prepare them. We characterize that coaching from both sides, surveying coaches and interviewing drivers, and find that manager-led coaching outperforms self-coaching across experiential outcomes.

Funding and External Collaborations

The autonomous vehicle trust work is a collaboration between the UCSD departments of Cognitive Science, Computer Science and Engineering, and external industry partners. The fleet driver coaching research is conducted with Lytx.

AffAdapt: AFFect-driven ADAPTive AI Personas for Seamless Conversations

Mon, 02 Nov 2026 00:00:00 +0000

Depression Detection at the Point of Care: Automated Analysis of Linguistic Signals from Routine Primary Care Encounters

Wed, 08 Apr 2026 00:00:00 +0000

Designing for College Mental Health: We Have Resources; We Lack the Reach