This is a 0 to 1 design project ideating a AI-Powered research tool for internal usage.
My primary contributions included synthesizing information, creating models, and leading the prototyping and interface design. I played a critical role in owning the prototype development and leading expert reviews and usability testing sessions.
The Aether concept was highly praised by key stakeholders, and HRI@Ohio will continue its development confidentially.
Project Overview
How did we navigate through ambiguity? What efforts we took to surface the real problem behind?
Jan - Aug 2024
Week
1
2
3
4
5
6
7
8
9
10
11
12
13
14
🎨
Initial Research/ Background Research
📈
Stakeholder Map/ Concept Model
🎉
Literature Review
DESIGN
🗝️
Expert Interview/ Talking to Internal Stakeholders
💙
Synthesis & Storyboards
🔐
Spring Presentation
🗂️
Research Map Co-creation workshops
AUTH
🧢
Competitive Analysis
🗺️
Concept Validation & Design Metrics
💰
Prototype Creation & Initial Usability Testing
STRIPE
🏗️
Prototype Testing with Internal Stakeholders
🛠️
HRI-SV visit & Testing session
↗️
GPTs Co-creation Workshop
📤
Data Export
🚀
Final Pitch!
🧪
Handoff Documents
💪
Website Development
📈
Process documentation
PRODUCT
Jan - Aug2024
Month
Today
Spring
Summer
1
2
3
4
5
6
7
8
9
10
11
12
13
14
🎨
Initial Research/ Background Research
📈
Stakeholder Map/ Concept Model
🎉
Literature Review
DESIGN
🗝️
Expert Interview/ Talking to Internal Stakeholders
💙
Synthesis & Model Creation
🔐
Spring Presentation
🗂️
Wushiland Ideation Card Matching
AUTH
🧢
Research Road Map
🗺️
Concept Validation
💰
Mock Up Creation
STRIPE
🏗️
Prototype Testing with Internal Stakeholders
🛠️
HRI-SV visit & Testing session
↗️
Hi-Fi prototype with LLM GPTs creation
📤
Data Export
🚀
Final Pitch!
🧪
Handoff Documents
💪
Website Development
📈
Process documentation
PRODUCT
What's wrong
Repetitiveness slows down the researches.
Field-testing needs help.
An intelligent system is needed to guide decisions about deploying field-based experiments that may consider research goals and suggest scenarios, modalities, technologies and options for robustness, weight, and flexibility.
Two repeated manual efforts
Each step requires considerable time and effort, with team members often repeating the same actions without automated support. This not only slows down progress but also limits the team’s ability to focus on more strategic, value-adding aspects of research.
Background research is a long, tedious process
Writing planning proposal is time-consuming
Our Solution
Aether, a macro-level accelerator for experiments and a micro-level companion for individual researchers in HRI
Aether supercharges the background research
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.
Below are stories of how I lead the team during research, design, and prototype phases.I led a comprehensive approach that integrated innovative research, user-centered design, and rapid prototyping to translate complex concepts into tangible solutions.
User Research
Chat with Professors, PhDs, Researchers, and... a lot of people!
Coming Soon!
Design
Design AI features to optimize stakeholder workflows.
Prototype & Testing
4 Prototypes 5 Dimensions 1 Future
User Research Overview
Turning roadblocks into actionable solutions
How did I create models that helped my team navigate a meta, ambiguous project and narrow it down to a solvable problem?
We started with a long-standing, complex, and abstract problem involving both cultural and technical challenges, making it difficult to solve at first glance. However, through a series of user research, we were able to break down the complexity into manageable steps. As a visual thinker, I helped the team by creating conceptual models that synthesized research insights and served as a visual tool to align team communication and expectations.
15+ Literature Reviews
to understand the current HAIT research process, including common topics, testing methods, and typical research durations, all aimed at comprehending the problem space.
7+ Expert Interview
to gain a deep, micro-level understanding of the roadblocks in conducting tests and the detailed personal reflection on the process.
3+ Stakeholder Workshops
To document multiple research processes and create a unified blueprint that represents a general research workflow.
The challenge
Each Research Project Has a Unique Starting Point, Making Process Elicitation Challenging
From identifying initial objectives and gathering relevant background information to determining the appropriate methodologies, each stage demands a fresh perspective and careful planning. The variability in starting conditions can introduce complexity, necessitating flexibility and adaptability to address distinct project requirements. As a result, building a framework that supports a consistent yet customizable process becomes essential to streamline efforts and maximize research effectiveness across diverse projects.
This is an illustration of how we document the process for each researcher (Click to interact)
Key Findings
#1 To find plausible metrics, researcher spend excessive time in background research
A lack of a comprehensive theory or theoretical framework in HAIT is universal
Mini-App can pre-fill forms, personalize content, and streamline account-related actions without requiring repetitive data entry. This functionality not only reduces the risk of human error but also saves users valuable time, preventing the frustration that can arise from typing in redundant information.
"What variables should be assessed when an AI system encounters a group of individuals?"
Quote from one of our participants underscored a common obstacle in contemporary HRI research: the lack of clearly defined metrics for evaluating HAIT performance.
#2 "Background research is integral at every stage"
A lack of a comprehensive theory or theoretical framework in HAIT is universal
Background research, such as literature reviews, plays a crucial role at every phase. Researchers rely on their literature findings to guide decisions from study design to data analysis.
Study planning and approval are time-consuming
Developing a comprehensive study plan for real-world research can take anywhere from one week to a month, as it needs to align with the project timeline, budget, logistics, and safety regulations. Furthermore, the approval process may take an additional 2-3 months.
Initial Findings
#1: A lack of Framework
#2: Background Research
#3: Planning
"To find plausible metrics, researcher spend excessive time in background research"
A lack of a comprehensive theory or theoretical framework in HAIT is universal
An analogy: Just as there is a well-known, 'golden' recipe for making a classic Martini, ensuring consistency and expected outcomes every time, the same cannot be said for Human-AI Interaction (HAI) research. In this field, there isn't a single, universally accepted approach or formula that guarantees success,
"What variables should be assessed when an AI system encounters a group of individuals?"
Quote from one of our participants underscored a common obstacle in contemporary HRI research: the lack of clearly defined metrics for evaluating HAIT performance.
"Background research is integral at every stage"
Background research, such as literature reviews, plays a crucial role at every phase. Researchers rely on their literature findings to guide decisions from study design to data analysis.
"Study planning and approval are time-consuming"
Developing a comprehensive study plan for real-world research can take anywhere from one week to a month, as it needs to align with the project timeline, budget, logistics, and safety regulations. Furthermore, the approval process may take an additional 2-3 months.
Synthesize the findings
Two Key Weights Dragging Research Process
Logistical Weight
HRI researchers need to submit a planning proposal for research testing, which can take up to six months to complete.
Cognitive Weight
The extensive effort required to sift through numerous articles and studies delays their ability to focus on experimentation and data analysis, hindering overall progress.
Next Step,
We design to take the two Key weights off researcher's shoulder
If researchers need to sift through extensive literature to find reliable projects and articles related to their current experiments, why don't we create a product to help them quickly review the literature and conduct tests faster.
Next Chapter : Design
Next Chapter : Prototype
Design Iteration
How I iteratively designed AI features to optimize stakeholders' workflow.
The Two Challenges
Cognitive Weight
The extensive effort required to sift through numerous articles and studies delays their ability to focus on experimentation and data analysis, hindering overall progress.
Logistical Weight
HRI researchers need to submit a planning proposal for research testing, which can take up to six months to complete.
Elicit
An AI tool designed to assist with literature review and other research tasks.
Pro: Handles independent tasks efficiently.
Con: Lacks strong integration capabilities across different research activities.
DOT AI
A personal assistant that grows with the user.
Pros: Grows with the user, offering personalized assistance.
Cons: Their focus is different from ours; we emphasize enhancing the user's unique habits and capabilities.
Companies are serving only rice instead of meals.
Companies are offering users isolated use cases like small, individual peas, instead of integrating them into a comprehensive, substantial solution.
Our Vision
Envision Aether as a macro-level accelerator for experiments and a micro-level companion for individual researchers in HRI.
Previous Chapter : Research
Next Chapter : Prototype
My mission
I created 3 unique prototypes covering 5 dimensions of prototype
Ensures compelling, robust results.
If researchers need to sift through extensive literature to find reliable projects and articles related to their current experiments, why don't we create a product to help them quickly review the literature and conduct tests faster.
My mission
The 5 Dimensions of Prototype Fidelity
The 5 Dimensions of Prototype Fidelity provide a framework for evaluating and guiding the development of prototypes based on their intended purpose and level of detail. These dimensions help designers choose the appropriate depth and realism for each aspect of a prototype.
"The proper fidelity level will focus the feedback you receive on the proper aspect of the design"
-- Kathryn McElroy author of "Prototype for Designers"
Rapid Prototyping
Co-design Output with stakeholders for ideal model output
This process involved engaging in structured co-design workshops, gathering diverse perspectives, and incorporating iterative feedback to refine the model's output criteria. The result was a set of tailored and actionable model outputs that effectively met stakeholder needs and aligned with the project's overarching objectives.
Our Goal
Gathering feedback for a futuristic experience that researcher have never encountered
The Process of Crafting A GPT Model
Step 1. Set the Scene
We'll use a GPT builder, which lets us create conversational AI without coding.
We'll type our feedback and specifications into the left panel, and see the AI's responses live on the right.
Sounds Goood!
Step 2. Plug in the Plot
We gonna give you a Research question, and if you don’t mind please read it aloud first, and we will proceed from there.
Sure things!
03. Solicit & Craft Feedback
What are your immediate thoughts? What aspects of its output do you find helpful or unhelpful? Think out loud
What would you like to see the AI do differently?
Well, I think it would be great if Aether could......
Continue Iterate and Repeat Step 2 - 4...
Key insights
Good for newcomers, not useful for experts.
Prior Work Synthesis
For drafting initial research proposals, this tool could be helpful by synthesizing information from prior work.
Need for Specificity
Participants suggested adding details such as cost, participant numbers, available tools, and extracting common methods from literature reviews.
Usability Improvements
Allowing researchers to directly incorporate GPT suggestions into planning modules would enhance usability.
Idea Proposal
Ideally, the tool could generate tailored experiment design ideas based on research constraints and objectives.
.svg)
.svg)
.svg)
.svg)
.svg)
.svg){kind=small}
.webp)
.webp)
.webp)
