← Back to projects

Capstone Project

Calibrating Trust in AI: Designing the TRUST-CT Scale

A human–AI interaction capstone exploring how interface design and a new scale can support healthy skepticism, reduce overreliance, and promote critical thinking with AI-generated text.

Role: Primary researcher & designer
Course: HSE 477 – Capstone
Focus: Human–AI trust, automation bias, digital literacy
Human–AI Interaction Trust & Overreliance Experimental Design Scale Development

Project Overview

As large language models like ChatGPT, Copilot, and Gemini enter everyday workflows, people are increasingly relying on AI-generated text to make academic, professional, and personal decisions. My capstone project asks: what makes people trust AI responses, and when does that trust actually support critical thinking?

I designed TRUST-CT (Trust for Critical Thinking), a user-facing scale that measures why someone trusts an AI output and whether that trust is likely to lead to healthy verification behaviors instead of blind acceptance.

The Problem

Existing trust scales were largely created for traditional automation (like autopilot systems), not generative AI that can be confidently wrong and hallucinate. That mismatch creates two failure modes:

  • Over-trust / automation bias – users accept AI outputs without checking.
  • Algorithm aversion – users see one error and reject AI even when it would help.

Many current measures focus on “increasing trust,” but don’t ask whether that trust is appropriately calibrated or whether the user actually verifies claims.

Research Goals

Main goal: create a brief, validated scale that captures the reasons behind a user’s trust in AI and predicts whether they will double-check the information.

I focused on three questions:

  • What latent factors best explain user trust judgments that trigger verification?
  • Does TRUST-CT predict appropriate reliance (accepting correct AI, rejecting incorrect AI)?
  • Which interface cues (uncertainty, citations, explanations) most improve trust calibration?

TRUST-CT Scale Concept

TRUST-CT adapts information-literacy frameworks like CRAAP and SIFT to a human–AI trust setting. It operationalizes five user-centered factors:

  • Transparency – does the AI reveal limits, uncertainty, and gaps?
  • Reliability – does it feel consistent and accurate over time?
  • Understandability – can users follow the reasoning behind a response?
  • Source Credibility – are sources visible, checkable, and trustworthy?
  • Teleology (Purpose) – is the AI’s purpose clear and aligned with user goals?

The scale is intended not only to measure trust but also to nudge users into a brief moment of reflection when they read an AI-generated claim.