0→1 Problem Framing
Ethical AI
Guide users to make more ethical decisions with AI
ROLE
Project Manager (Jan - May)
UX Designer (May - Jul)
PROJECT DURATION
Jan - July 2025
TEAM
Project Manager
Product Designer
Engineer
Researcher
Creative Strategist
Project Manager
Product Designer
Engineer
Researcher
Creative Strategist
Video Creds: Ken Stewart
This is a Capstone Project sponsored by 99P Labs and Honda Research Institute at Carnegie Mellon University for the Master’s in Human-Computer Interaction program.
PROJECT BRIEF FROM CLIENTS
How might non-technical users reflect their evolving values in AI systems beyond the domains of transportation?
LEADING WITH CLARITY
With the initial project brief that Honda Research Institute’s 99P Labs had shared with us, I identified three key constraints: the medium of the solution (AI), the problem domain (outside of transportation), and the timeframe (10–15 years).
With the initial project brief that Honda Research Institute’s 99P Labs had shared with us, I identified three key constraints: the medium of the solution (AI), the problem domain (outside of transportation), and the timeframe (10–15 years).
With the constraints in mind, I defined areas of ambiguity and drove clarity by structuring the research into three targeted workstreams
With the constraints in mind, I defined areas of ambiguity and drove clarity by structuring the research into three targeted workstreams
REFRAMED PROBLEM STATEMENT
My strategy to split up the workstreams enabled us to gather enough information to build conviction around the problem we needed to explore.
My strategy to split up the workstreams enabled us to gather enough information to build conviction around the problem we needed to explore.
HMW guide high-stakes decision-makers to engage System 2 thinking so that they can better consider the uninteded consequences for others invovled?
HMW guide high-stakes decision-makers to engage System 2 thinking so that they can better consider the uninteded consequences for others invovled?
CONTEXT
There have been many cases where leadership decisions have negatively impacted people at scale
There have been many cases where leadership decisions have negatively impacted people at scale
Boeing 737 MAX Crisis
Boeing 737 MAX Crisis
Downplayed importance of pilot training on new MCAS software to stay competitive have led to:
Two crashes that killed 346 people
Downplayed importance of pilot training on new MCAS software to stay competitive have led to:
Two crashes that killed 346 people
Ford Pinto
Ford Pinto
Ford Pinto
Internal memos weighing the cost of fixing fuel tank rupture issue despite crash-test results led to:
27-180 deaths from fire-related crashes
Therac-25 Overdose
Reusing software components from older machines without error checking led to:
6+ patients receiving radiation overdoses
SHORTCOMINGS OF AI
And lack of objectivity in human decision-making has driven the use of AI systems to make decisions, which has, in cases like the COMPAS tool, resulted in biased outcomes
And lack of objectivity in human decision-making has driven the use of AI systems to make decisions, which has, in cases like the COMPAS tool, resulted in biased outcomes
An independent ProPublica analysis found it labeled Black defendants as “high risk” even when they didn’t reoffend, while predicting white defendants as “low risk” despite many committing future crimes.
An independent ProPublica analysis found it labeled Black defendants as “high risk” even when they didn’t reoffend, while predicting white defendants as “low risk” despite many committing future crimes.
Labeled higher risk by COMPAS, yet didn't reoffend
Labeled higher risk by COMPAS, yet didn't reoffend
White
0%
0%
White
0%
0%
White
0%
0%
African American
0%
0%
African American
0%
0%
African American
0%
0%
Labeled lower risk by COMPAS, yet did reoffend
Labeled lower risk by COMPAS, yet did reoffend
White
0%
0%
White
0%
0%
White
0%
0%
African American
0%
0%
African American
0%
0%
African American
0%
0%
PROBLEM SPACE
I interviewed high-stakes decision makers in the manufacturing and robotics domain and uncovered three key challenges in using AI to support their decision-making.
I interviewed high-stakes decision makers in the manufacturing and robotics domain and uncovered three key challenges in using AI to support their decision-making.
SKEPTICISM
TOOLS
With the existing AI tools, high-stakes decision makers felt the "black box" recommendations superficial
SKEPTICISM
TOOLS
With the existing AI tools, high-stakes decision makers felt the "black box" recommendations superficial
COMPLEXITY
DECISION SPACE
Problem contexts varied across domains, company values, their own values and specific moments
COMPLEXITY
DECISION SPACE
Problem contexts varied across domains, company values, their own values and specific moments
COMPLEXITY
DECISION SPACE
Problem contexts varied across domains, company values, their own values and specific moments
RISK
EXPENSE
The stakes of making a wrong decision came at a huge cost
RISK
EXPENSE
The stakes of making a wrong decision came at a huge cost
RISK
EXPENSE
The stakes of making a wrong decision came at a huge cost
User Research
I CONDUCTED
user research with 10+ users to create an intent-driven discovery product strategy that was implemented in 2025
User Research
I CONDUCTED
user research with 10+ users to create an intent-driven discovery product strategy that was implemented in 2025
User Research
I CONDUCTED
user research with 10+ users to create an intent-driven discovery product strategy that was implemented in 2025
User Research
I CONDUCTED
user research with 10+ users to create an intent-driven discovery product strategy that was implemented in 2025
SOLUTION
As of July 2025, I'm designing for a multi-agent AI system of ethical sub-agents that visualizes potential alternatives based on unintended consequences surfaced
As of July 2025, I'm designing for a multi-agent AI system of ethical sub-agents that visualizes potential alternatives based on unintended consequences surfaced
KEY UX CHALLENGES
As a UX designer, I am researching a new interaction paradigm for our multi-modal system that integrates user feedback seamlessly into the experience, making it unobtrusive
As a UX designer, I am researching a new interaction paradigm for our multi-modal system that integrates user feedback seamlessly into the experience, making it unobtrusive
In order for the multi-agent AI system to be contextual, it was essential for the system to prompt user feedback in ways that did not feel intrusive.
In order for the multi-agent AI system to be contextual, it was essential for the system to prompt user feedback in ways that did not feel intrusive.
Video Creds: Ken Stewart
My KEY TAKEAWAYS SO FAR…
Finding a starting point in ambiguity
Within the inherently subjective context of ethics with time constraint, my strategy was to go broad individually, then converge as a team.
We initially started with literature reviews, then shared out our findings to then align on a shared definition of “ethics” as personal principles and moral codes that guide decision-making.
Guiding the team through uncertainty
Gaining conviction through action, rather than overthinking the perfect next step, empowered the team to critically assess and refine the direction of our work, making it easier to pivot when necessary.
Balancing different ways of working within the team
Working on speculative projects under time constraints meant I had to learn how to effectively balance the team's creativity with grounded work. Deliberately planning activities like Crazy 8s and sci-fi movie nights while developing our research plans and survey questions helped the team uncover ambitious unknowns that was outside the box, while following up with grounded discussions helped surface new insights that directly supported our project goals.
Finding a starting point in ambiguity
Within the inherently subjective context of ethics with time constraint, my strategy was to go broad individually, then converge as a team.
We initially started with literature reviews, then shared out our findings to then align on a shared definition of “ethics” as personal principles and moral codes that guide decision-making.
Guiding the team through uncertainty
Gaining conviction through action, rather than overthinking the perfect next step, empowered the team to critically assess and refine the direction of our work, making it easier to pivot when necessary.
Balancing different ways of working within the team
Working on speculative projects under time constraints meant I had to learn how to effectively balance the team's creativity with grounded work. Deliberately planning activities like Crazy 8s and sci-fi movie nights while developing our research plans and survey questions helped the team uncover ambitious unknowns that was outside the box, while following up with grounded discussions helped surface new insights that directly supported our project goals.
Finding a starting point in ambiguity
Within the inherently subjective context of ethics with time constraint, my strategy was to go broad individually, then converge as a team.
We initially started with literature reviews, then shared out our findings to then align on a shared definition of “ethics” as personal principles and moral codes that guide decision-making.
Guiding the team through uncertainty
Gaining conviction through action, rather than overthinking the perfect next step, empowered the team to critically assess and refine the direction of our work, making it easier to pivot when necessary.
Balancing different ways of working within the team
Working on speculative projects under time constraints meant I had to learn how to effectively balance the team's creativity with grounded work. Deliberately planning activities like Crazy 8s and sci-fi movie nights while developing our research plans and survey questions helped the team uncover ambitious unknowns that was outside the box, while following up with grounded discussions helped surface new insights that directly supported our project goals.
Copyright © 2025 Lucy Ji Soo Choi
All rights reserved.
This is a Capstone Project sponsored by 99P Labs and Honda Research Institute at Carnegie Mellon University for the Master’s in Human-Computer Interaction program.