AI Transparency in Healthcare
IBM Service Design Challenge
TIMELINE
Nov 2024 - May 2025
SUMMARY
ECHO, a digital platform that fosters transparent communication between speech-language pathologists (SLPs) and families around the use of AI, addressing critical gaps in AI trust, cultural responsiveness, and explainability within pediatric speech therapy.
MY ROLES
UX Design Lead: I managed project planning and timelines, balanced critical decisions, and guided research and design efforts to keep the team aligned and collaborative, successfully meeting each milestone throughout the 8-month design challenge
User Researcher: Proposed scalable design solutions by leading workshops with patients, clinicians, and AI engineers to balance stakeholder needs with technical feasibility
KEY ACHIEVEMENTS
Awarded 2nd place out of 200+ international teams
CONTEXT
Mitigating bias in AI healthcare is one of the biggest task faced worldwide
During the 2025 Student service design challenge, IBM challenged teams to design equitable services that help human navigate bias. We chose to tackle the challenge in the complex space of AI healthcare, exploring and paving the way toward equitable and unbiased future.
$4.9 trillion
spend in U.S. healthcare, creating pressure for more efficient, tech-enabled solutions
> 40% underdiagnosis
are experienced by underserved populations
Among 11 high-income countries…
U.S. ranked last in health equity, reflecting deep disparities in care
AT A GLANCE
A trusted and transparent AI Healthcare
Echo serves as the first step of trust building between patients and AI Healthcare. It facilitates doctor patient communication on AI usage before, during, and after visits. For doctors, Echo is also a research hub to understand AI and build trust with it.
The application is streamlined online and includes dedicated section about AI usage information.
PROBLEM FRAMING
Understanding from the source: clinical research to speech-language pathologist
To deeply understand the problem space, I spoke with diverse range of professionals working in the healthcare system:
Seattle Clinical researchers focused on recruitment and trial design
UW Health equity experts studying access & bias
Speech-Language Pathologists (SLPs) working in various settings (hospitals, schools, private clinics)
AI/ML engineers building healthcare tools
Patients and caregivers with lived experience in speech therapy
I mapped out the system map of stakeholder relations in the clinical research context to understand the interactions and values exchange in between
KEY INSIGHT
Transparency gap in AI usage result from limited communication between doctors and patients
72%
Patients demanded transparency in how healthcare AI makes decisions
80%
Patients were comfortable with AI in care if clinicians understood and explained it
48%
Clinicians agreed utilization of AI tools can increase the risk to patient privacy
PROBLEM STATEMENT
How might we create a single source of transparency that helps patients communicate clearly, and support clinicians in using AI ethically and efficiently despite time constraints, cultural gaps, and limited system guidance?
IDEATION
Make ideas spark through complexity
I organized brainstorming sessions where we explored human-AI collaboration models, balancing automation with clinician oversight. While we had many great ideas, I made sure to prioritize concepts that would best bring out our design objectives of transparency, voice, and reflection.
Ideas brainstormed based on critical user needs
We didn't implement all of them because…
Could stigmatize patients by labeling them based on race or diagnosis, reducing trust instead of building it.
Information overload in waiting rooms could confuse patients and overwhelm clinicians.
Low technical feasibility, creating friction for patients and providers instead of integrating seamlessly into existing workflows
Patient side features focus on communication
AI Explanation Layer
Simple visuals and language showing what the AI analyzed and why
Voice of the Patient
A channel for patients to share feedback on AI-driven results
Research opportunity
Voluntary option for patients to contribute data to strengthen AI models
wireframe of patient side phone screen
Clinician side features focus on feedback and community
Bias Flags
Alerts for potential dataset mismatches (e.g., bilingual cases)
Feedback Loop
Patient feedback integrated into clinicians’ workflows for continuous improvement
Insight community
A collaborative space for clinicians to access research insights from the community to facilitate diagnosis
Wireframe of clinician side screen
ITERATION
Validate designs with real-world use case
I conducted a co-design session with a speech-language pathologist, who confirmed the value of contextual intake and AI explanation features. I iterated based on their feedback, simplifying workflows and clarifying client-facing language.
1
Adding summary on critical information of community resources helps clinicians utilize it quickly and effectively.
2
Showing personalized bias considerations of patient's overall file is more useful than listing out general bias considerations of AI tools.
FINAL DESIGN
Context-rich intake form that adds transparency in AI usage
Collects detailed patient information to ensure AI considers personal, cultural, and contextual factors
Patient's AI acceptance note that makes clinicians feel confident
Provides clinicians with a comprehensive patient profile to guide accurate, personalized decisions
Clinician networking to grow trust in AI
Connects clinicians to shared research, cases, and AI insights across their professional network
Community resources for continuous improvements in the healthcare system
Central hub for accessing datasets, guidelines, and tools to design, monitor, and refine AI systems
IMPACT METRICS
How I would evaluate impact
Increased satisfaction of application process
Parent users submit enriched intake form that enables them to add contextual and cultural details and compare the perceived usability or satisfaction score with previous form.
Reduction rate in clinician prep time
Time spent preparing for sessions, by using Echo’s AI-supported summaries and context views which flag potential biases.
Improved percieved transparency of AI use
Parents’ understanding on how AI is (or isn’t) used in their child’s care, as measured by a post-visit survey & clinicians’s perceived transparency after using Echo Insights for guidance.
LEARNINGS
0 to 1 is a process of deconstruct and rebuilding the mindset
Leading a team to explore an unfamiliar space pushed me to be even more proactive. A key challenge I faced was the need to quickly adjust our project direction based on emerging research. To keep the team aligned, I organized debriefing sessions after each interview and reflection meetings after every round, which were effective approaches that fostered shared understanding and agility.
I’m proud that through hard work and the courage to step outside our comfort zones, we achieved a well-deserved accomplishment!
The amazing Healthcare and Humanity team! What a journey we've been through together!!
Thanks for stopping by;)
Glad I was able to share a piece of me with you.
The story goes on…
@ Janet Chen, 2025