Designing trustworthy AI systems
Project Overview
With the advancement in AI technology, users often make decisions based on predictions suggested by the AI systems. However, the underlying models that produces AI system predictions to users are often opaque (also known as black-box model) and can contain flaws (e.g., bias in data set and prediction), thus leading to possible undesirable decisions made by users (e.g., unfair decisions). Thus, it is important to ensure AI technology and systems are trustworthy in the whole AI lifecycle, from the early design stage, development, deployment to final use.
In this project, I developed and led a AI design-thinking workshop that helped designers, engineers and product managers to understand how to identify user needs and potential AI features that users want. I created design-thinking activities that educate people who do not have technical background to understand what AI need and interact with users, thus supporting the product teams to integrate user’ perspectives in designing a AI features that users trust and willing to adopt.
Role and Time required
Role: Lead UX Researcher
Time: 3 months
Content of the workshop
The workshop will guide product teams to design user-centered AI technologies. Attendees will learn the following from the workshop: 1) Understand users’ AI needs 2) Align AI capabilities with users’ values 3) Identify feasible AI solutions that respect users’ data privacy and conform the Company’s data ethics principles
Activity 1: Understand users’ AI needs
- Identify who are the target users
- Using Jobs-to-be-done framework, identify painpoints that block users to work effectively in their workflow (i.e., get their job done)
- Leverage research findings about users perceived importance and task complexity to identify potential areas for AI to support users’ work (i.e., improvement of their job to be done)
Activity 2: Align AI capabilities with users’ values
- Create “How might we” statement to brainstorm what AI can support users to complete their job in their workflow
- Engage in an interactive role playing activity that help participants experience the interaction between AI and users, putting product teams into the users’ shoes
Activity 3: Identify feasible AI solutions that respect users’ data privacy and conform the Company’s data ethics principles
- Identify the common AI capabilities in the industry
- Mapping the AI idea (from Activity 2) with the AI capabilities
- Identify what kind of data that we will use to feed the AI (i.e., bridging the communication between product team and engineering team)
Deliverable
- Held workshops among 60+ stakeholders in cross-functional teams (Product, Design, Engineering)
- Created template that Product teams can use easily to run workshop before deciding what AI they want to develop
- Recognized for its impact, the workshop was selected to be presented at Autodesk Tech X Conference 2024