UX Research · Interaction Design · Prototype · 2026
FairLend
AI is making loan decisions that hurt people — not because it is malicious, but because it learned from data that was already discriminatory. I designed a dashboard to put the human back in the loop.
Role
Solo researcher and designer
Methods
Literature review · Cognitive forcing design · Figma prototyping
Tools
Figma · Zotero · Material Design
Course
Human-Centered AI — UNC Charlotte, Spring 2026
THE PROBLEM
Lending has never been an equitable system
Redlining — the federal government's practice of denying mortgage credit to residents of predominantly Black neighborhoods — created patterns of racial exclusion that lasted for generations. People from historically redlined communities still face higher loan denial rates today, even when they are financially qualified.
Now AI is being introduced into this system. These models are trained on historical data — the same data that reflects decades of discrimination. When a model learns from that data, it does not learn to be fair. It learns to replicate the patterns it was shown.
This creates something called proxy discrimination. The model never mentions race. Instead it uses variables like zip code, which correlates strongly with race due to housing segregation. A model that penalizes applicants for their zip code is effectively discriminating by race — quietly, at scale, and nearly impossible to see from the outside.
RESEARCH QUESTION
"How can a human-centered AI lending dashboard leverage cognitive forcing functions and augmented sense-making to help loan officers critically evaluate algorithmic recommendations and reduce overreliance on potentially biased AI outputs?"
WHAT THE RESEARCH TOLD ME
20 sources. Four findings that changed everything.
FINDING 01
People overtrust AI
Professionals follow AI recommendations without critical evaluation. In standard lending dashboards, officers follow the AI 96.8% of the time.
FINDING 03
Explainable AI makes it worse
Adding explanations to AI recommendations doesn't reduce overreliance — a persuasive explanation gives users more reasons to agree, not question.
FINDING 02
People don't see AI as biased
Algorithmic decisions producing racial disparities are less likely to be seen as discriminatory than identical human decisions made by humans.
FINDING 04
Cognitive forcing works
Design elements that require deliberate thinking before acting significantly reduce overreliance — but users prefer the easier design. This tension shaped everything.
THE PIVOT
I started with the wrong idea
My initial design direction was explainable AI. I planned to redesign the standard lending dashboard so it displayed clear explanations alongside the AI's recommendation. The thinking was that transparency would naturally prompt more critical evaluation.
The evidence said otherwise.
Research by Buçanca et al. showed that explanations can increase overreliance. A well-presented explanation gives a person more justification for deferring, not more reason to push back. I changed direction entirely — toward cognitive forcing functions and the four-step structure of FairLend.