At a Glance
Challenge: Museums increasingly are exploring the use of AI technologies for learning, but lack frameworks to understand how visitors engage with these exhibits and what design features support different learning pathways.
Outcome: Developed a research-based framework identifying seven core engagement behaviors (questioning, consuming, creating, explaining, connecting, predicting, evaluating) and mapped them to participation genres (Hanging Out, Messing Around, Geeking Out), providing actionable design guidance.
The Challenge
The Exploratorium's 'Adventures in AI' exhibition featured 30+ interactive technological exhibits designed to teach visitors about artificial intelligence concepts, such as word embeddings, neural networks, bias in data, and more. AI literacy in informal learning is a new and emerging field. As such, we have a:
- Limited understanding of how museum visitors interact with AI exhibits in social, self-directed contexts
- Lack of frameworks to characterize the quality of engagement in relation to learning content
- Unclear design guidance on which features support which types of engagement
- Need for accessible language to communicate findings across stakeholders (designers, educators, researchers)
This led us to ask:
Research Question:How do visitors engage with interactive technological exhibits in museums, and what design features support diverse engagement types?
My Role
- Data cleanup and preliminary analysis
- Interaction analysis of 20+ hours of visitor audio
- Collaboration with exhibit designers to identify design features
- Framework development and synthesis
Research Process
Theoretical Framework
I grounded this study in the concept of 'genres of participation' developed by Ito et al., which identifies three constellations of learning behaviors:
- Hanging Out (HO): Social, peer-driven engagement focused on being together
- Messing Around (MA): Experimental, interest-driven exploration
- Geeking Out (GO): Deep, intensive engagement with specialized knowledge
This framework has proven central to designing informal learning environments like makerspaces, but hasn't been applied to free-choice, self-directed learning with museum exhibits.
Phase 1: Audio-Recorded Visitor Study (7 weeks)
Participants: 17 visitor groups (52 individuals total)
Method: Visitors wore microphones as they explored the 'Adventures in AI' exhibition while researchers observed from a distance
Visit duration: 20-80 minutes per group
Data collected: ~20 hours of naturalistic audio of visitor conversations and interactions
Note: This data collection was led by the broader team at the Exploratorium.
Phase 2: Ethnographic Field Work (3 days)
Participants: 28 visitor groups
Methods:
- Direct observation of visitor interactions with exhibits
- Unstructured interviews with visitors about their experiences
- Field notes documenting engagement patterns
Phase 3: Interaction Analysis
I analyzed the data using interaction analysis techniques adapted from prior museum research:
Step 1: Identify Significant Events
- Reviewed all audio transcripts and observation notes
- Identified moments of meaningful engagement with learning content
- Focused on social conversation, experimentation, and rule-breaking behaviors
Step 2: Code Engagement Behaviors
- Inductively coded visitor interactions with each other and exhibits
- Drew on existing museum engagement literature and empirical observations
- Identified distinct behavior patterns
Step 3: Map to Participation Genres
- Categorized behaviors as Hanging Out, Messing Around, or Geeking Out
- Determined categorization based on subject and complexity of behaviors
Step 4: Identify Design Features
- Collaborated with exhibit designers to map behaviors to specific design features
- Documented which features supported which participation genres
Method Reflections
What worked:
- Naturalistic audio data captured authentic visitor conversations without researcher influence
- Captured interactions across several exhibits with diverse design features allowing us to see transitions
- Collaborative analysis with designers ensured actionable insights
- Familiar participation framework bridged learning science literature and museum practice
Challenges:
- Audio-only approach missed non-verbal engagement
- Analysis ongoing; preliminary findings may shift as dataset completes
Preliminary Findings (WIP)
Finding 1: Seven Core Engagement Behaviors
Identified distinct ways visitors engage with learning content at technological exhibits:
Questioning
Consuming
Creating
Explaining
Connecting
Predicting
Evaluating
The subject and complexity of these behaviors determine how they map to participation genres (Hanging Out, Messing Around, or Geeking Out).
Finding 2: Design Features Shape Engagement
Specific design features support different participation genres. Example from "This or That" exhibit:
| Participation Genre |
Example Behaviors |
Supporting Design Features |
| Hanging Out |
Taking pictures, laughing, consuming non-subject-matter information |
Familiar activity (photos) that reduces barrier to entry |
| Messing Around |
Assessing where pictures fall on spectrum, evaluating subject-matter information |
Quick repetition, experimentation, immediate feedback allows easy testing |
| Geeking Out |
Testing hypotheses about what the AI detects, predicting, questioning, explaining |
Keeping log of results allows pattern identification |
Impact and Outcomes
For HCI Community:
- Provides framework to design and evaluate interactive AI exhibits in museums
- Extends understanding of Human-AI interaction in informal learning contexts
- Paper submitted to HCI conference (under review)
For Exploratorium:
- Provides evidence-based framework for future AI exhibit development
- Offers accessible language for discussing visitor engagement across teams
Ongoing Work and Next Steps
Current Analysis:
- Completing detailed coding of all 20+ hours of audio data
- Refining rubric that maps behaviors to participation genres
- Documenting design features across 6 primary exhibits
- Preparing findings for publication
Related Publications
Paper under review for HCI conference (2026)
Additional publications forthcoming