01
Aether
How might we design an AI research assistant that streamlines the research process while learning and evolving with researchers?
My Contribution
User Research / Product Design / UX Design / Prototyping / Usability Testing
Project Type
Master’s Capstone Project @ Carnegie Mellon University (Client: Honda Research Institute USA)
Team
1 UX Researcher, 2 Product Designer, 1 UX Engineer
Timeline
6.5 months
Project Goal:
Gaining Real-world Insights in Early Human-AI Teaming Research to Bridge the Gap between Research and Product
Honda Research Institute (HRI) envisions a future where human-AI teams are the norm, driving their focus on understanding these collaborations. To bridge the gap between lab research and real-world applications, HRI finds it important to account for unpredictable environmental factors in early concept testing. By conducting studies and gathering data in real-world settings sooner, they can better evaluate the impact of their research on future Honda products, supporting their mission of practical innovation.
However, transitioning to real-world studies presents challenges:
Challenge 01:
Unique context for each research project
Human-AI teaming research is highly qualitative, with each project having its unique background and context, making study planning a time-consuming process.
Challenge 02:
Complex planning for real-world testing
Compared to lab testing, real-world settings involve more unpredictability—such as weather conditions, location regulations, and logistics—adding complexity to implementing an in-field study.
Design Overview:
Aether: An AI research assistant that streamlines the research process while evolving with researchers
When optimizing human-AI teaming research, Aether learns from the content researchers provide, refining its responses through continuous interaction. Over time, Aether will evolve into an “expert research assistant,” boosting research efficiency across Honda Research Institute.
User Research:
Understanding Barriers to Real-World Studies
User Architype:
Researchers with Strong Academic Background
Our target users are researchers at Honda Research Institute focused on human-AI teaming. While they are well-trained in academic research and experienced in their fields’ methods, they have limited experience conducting studies in real-world settings.
Research Method:
Interviewing Researchers and Analyzing Their Workflow by Co-creating Research Maps
To better optimize the process of gaining real-world insights, our team interviewed 9 researchers at Honda Research Institute and 12 academic researchers at Carnegie Mellon University. We asked the participants to map out their research process with index cards to learned their research process, how they plan studies, and their pain points in the process.
Key Findings
Finding 01:
Background research is integral at every stage
Background research, including literature reviews, is integral at every stage. Researchers base decisions—ranging from study planning to data analysis—on their literature findings.
Finding 02:
Study planning and approval are time-consuming
Writing a comprehensive study plan for real-world studies can take one week to a month, as it must align with the project timeline, budget, logistics, and safety regulations. Additionally, approval may take up to 2-3 months.
Design Concept:
Labeling data in background research, using it to help study planning
Ideation:
A System that Helps Study Planning by Learning from Researchers
Our initial concept was to allow researchers to input background research findings and study details into the planning system, enabling the AI to generate a draft plan for them.
System Structure
1. Labeling data during background research
As researchers use Aether for background research, it learns from the content of papers and their behaviors, such as tagging, annotating, and clustering. This process allows Aether to understand the project’s context and store it in the project database.
2. Using the labeled data for study planning
Aether uses the knowledge from the project database to assist in study planning by generating draft plans and offering tailored suggestions that align with the researchers’ needs.
Challenge 01
How might we label data in the background research process seamlessly?
Design Strategy:
Labeling Data during Background Research for Future Use
Currently, researchers must review and synthesize complex research papers, absorbing key points, identifying relevant information, and making connections between different pieces of literature. With Aether, researchers can streamline this process, efficiently organizing and labeling information to make it more accessible and useful for later stages of their research.
Design Solution:
Streamlining Annotation and Synthesis to Boost Researcher Adoption
1. Annotate the Papers and Extract Key Findings
Researchers can select excerpts from the papers to annotate. They can also assign custom or AI-generated tags to mark specific excerpts that stood out to them. For example, these tags can remind the researchers about key concepts from the reading, or inform them of new methods they can use in their own research.
2. Synthesize Research Findings through Visualization
In addition, researchers can use the virtual whiteboard space to visualize any recurring themes or patterns to help them inform their project direction. They can also opt to share anything on their whiteboard with other team members, facilitating team collaboration.
“The ability to quickly annotate papers and extract key points is valuable [for me].”
— HRI Senior Scientist
Challenge 02
How might we use the labeled data to make study planning easier?
Design Strategy:
Planning Studies with Labeled Data from Background Research and the Variables
Before planning, researchers need to enter the variables to the modular component as the context of the study they are going to plan. The researchers don’t need to know every variables at this point, Aether can generate the unknown variables based on the knowledge researchers have collected and tagged during the background research phase.
Design Solution:
Automating the Logistical Considerations with LLM
1. Generate a Draft Plan with LLM
Based on the variables entered by the researchers and information gathered from background research, Aether can generate the first draft of the plan in a format that adheres to the HRI conventions.
2. Edit the Draft Plan with Aether’s Suggestion
When the researchers are not sure about a specific variable, Aether can make new recommendations based on the research context and inputs from the researchers. This ensures that the plan is well-informed and aligns with the specific needs of the project.
3. Comprehensive Consideration of the Interdependencies between the Variables
Aether also considers interdependencies between each variable, reminding researchers of how changes in one module might affect others (ex. changing the location might affect participant recruitment). This helps researchers plan comprehensively and ensures that all aspects of the plan are consistent and cohesive.
“It’s always good to have a starting point like [the AI generated draft plan].”
— HRI Senior Scientist
Future Steps
Expanding Product Scope Through Hardware Integration and Advanced Technologies
In its current design, Aether assists researchers with background research and study planning. In the future, it could become an end-to-end tool, helping to conduct studies and collect data. Aether would speed up data analysis by using AI to quickly and accurately parse large amounts of information. This context-aware process, where each step enhances the next, would ultimately support researchers in writing their papers with the help of Aether.
As Technology Advances, We Envision Aether Integrating Seamlessly into Every Stage of the Research Process in the Future:
1. Interactive Data Collection and Analysis
As Aether gains a full understanding of the research context, it can support researchers with real-time, interactive data collection and analysis at each stage of the process, enabling them to review data and extract insights more efficiently.
2. Gaining Real-world Insights through Virtual Space
In the future, Aether could become ubiquitous, deployed on wearable devices or sensor-equipped drones to autonomously gather data from participants and their environment, removing the need for researchers to be onsite. These devices might even livestream and recreate study settings in virtual spaces, allowing researchers to observe real-world scenarios from the lab—effectively breaking spatial limitations in field studies.
Reflection
The importance of system thinking in AI product design
Reflecting on this project, one of the most critical challenges was understanding why researchers would adopt Aether over their existing methods, as adoption would hinge on how effectively Aether could achieve the project goal of increasing real-world studies. To accomplish this, we prioritized understanding the workflow, stakeholders, and entire research system to ensure seamless implementation and maximize impact.
Our final design, a web application, provided immediate value in enhancing research efficiency. But what resonated most with Honda Research Institute was the framework we developed for implementing AI solutions internally and a roadmap for using AI to improve their workflow in the future.
From this experience, I learned the importance of system thinking in future-facing product design. By fully understanding the ecosystem, product teams can build adaptable, comprehensive solutions that remain effective in a constantly evolving world.
