Improving AskPolly — Bringing Human-Centered Design to Artificial Intelligence
Posted on Tuesday, March 29th, 2022
Client | Advanced Symbolics Inc. |
Professor(s) | Jed Looker, |
Program | |
Students | Joyat Joseph Muhammad Khan Zhi Wang |
Project Description:
Introduction to Project
AskPolly is a self-service tool for market research created by Advanced Symbolics Inc. (ASI) that uses artificial intelligence to study what people are saying online on social media. A person enters a query of what they want to measure (e.g. trucker protest), and the population they want to measure (e.g. residents of Ottawa). Polly then returns what conversations people are having online, how people feel about each of the conversations, and the demographics of the people. Though a powerful tool on the back-end, AskPolly’s engineering-first interface design was in need of improvement. ASI approached our team to research and design a new interface for AskPolly that makes the product easier and more intuitive to use.
Our Roles
Muhammad Khan: UX Research
Joyat Joseph: Interaction Design
Zhi Wang: Data Visualization
What We Did
We began the project by conducting a literature review and a heuristic evaluation of the current product. We also conducted interviews with engineers and augmented research managers at ASI to gain a better understanding of AskPolly and its function. We organized the information architecture of AskPolly and researched data visualization techniques that we could use to output information on a dashboard.
Using what we learned during our initial research we created a low-fidelity prototype using Figma that makes several changes to the original product to fill gaps in the user experience. First, we simplified the search page to have a single entry for a query, rather than having to manually add keywords one by one in separate entry boxes. We also added a loading screen, to improve the experience during the 5 minutes that it takes Polly to gather information for the dashboard. On the dashboard, we introduced a new card-based interface to make the conversations easy to digest. We also introduced simplified graphs to quickly provide users with an overview of the information related to the conversations they select. We also added functionality to add and remove graphs so that the user can customize the dashboard to their needs.
Next Steps
The next step of this project is design validation. We will be conducting usability testing on our design, and iterating it over 4 weeks to ensure that it truly meets the needs of users. During the tests, we will also be collecting important information on how users enter queries into AskPolly, which will be used to improve Polly’s ability to understand user inputs. Once we have a validated design, we will present a high-fidelity prototype and our research findings to Advanced Symbolics Inc. so that the engineering team can implement our recommendations.
We would like to thank the team at Advanced Symbolics and Jed Looker for the opportunity to work on this project and for all their help!
Short Description:
Our team at the Human-Centered Design Lab explored how we could improve the user experience of AskPolly, a self-service tool for market research powered by Artificial Intelligence.