Case Study - Usability Testing of an AI Learning Partner

Jigsaw Interactive Agent (JIA) provides realtime support for small groups of middle and high school students in collaborative work.

Client
University of Colorado Boulder
Industry
Education, Technology
Year
  • Artificial intelligence
  • Vulnerable population
  • Product management

Overview

  • Jigsaw Interactive Agent (JIA) provides realtime support for small groups of middle and high school students in collaborative work
    • Led the user experience research throughout the entire product design and development process
  • Evaluated the usability of JIA to improve the UI and functionality
  • Stakeholders included the Project Manager, a UX Designer, Natural Language Processing (NLP) and Multimodal AI Experts, Software Engineers, Learning Scientists, and a Recruitment Manager

Rui’s early market research provided valuable insights that guided product development and improved student success.

Peter Foltz
Executive Director of iSAT Institute
School districts
5
Teachers
50+
Students
5,000+

Research questions

  • How do users perceive the design and functionality of the AI partner JIA?
    • How can we improve the trustworthiness of JIA?
  • What are users’ reactions to JIA's support?
    • What issues do users encounter when working collaboratively with JIA?

Research process

Managed two phases of usability testing:

  • Wizard of Oz to test AI partner JIA's interface and interaction design
  • Human-in-the-loop to evaluate JIA's collaborative functionality

Usability testing procedure

  • Group introductions and ice breakers
  • Individual learning work
  • Collaborative learning work with support from JIA
  • Post-survey feedback
  • Focus group feedback

Key metrics

  • Reaction to JIA
  • JIA’s trustworthiness
  • Collaborative task performance
  • Time spent on task

Many other metrics were also reported…

Positive feedback

  • Users indicated the AI partner JIA’s collaboration was helpful and supportive
    • JIA supporting users by asking relevant questions was perceived positively
  • Users perceived JIA to be trustworthy (4.2/5)
  • Users indicated that the JIA web app was simple and easy to use

Actionable feedback

  • 20% of JIA's message audio notifications were missed during active and rapid conversations
  • Ask for help button should accept detailed user input, rather than be a toggleable button
  • Personalization of the AI Partner

More case studies

A / B Testing of AI Communication Styles

I studied AI communication styles in Rocket League, an online, team-based, sports game.

Read more

COSA Booking App

The Car Online Service Appointment (COSA) booking app helps 700K users schedule appointments to service their cars at their local dealerships.

Read more

Let’s connect

Coffee chats about…

  • Work opportunities
  • Public speaking
  • Publications
  • Mentorship