.png)
There is an assumption that embodiment in AI is considered an improvement without substantiating evidence, while also tackling the absence of established benchmarks defining a quality collaborative drawing experience.
How do different interaction modalities with an AI agent influence the interaction dynamics and affect human perception of the co-creative process?
Drawcto: co-creative AI for drawing
Drawcto's agents take turns with the user to create a varied strokes based on the selected agent for non-representational line art
Interaction modalities for collaborative drawing
Compare two different interaction modalities for collaborative line drawing activities: a web-based software interface and a tangible robot interface
Wizard of Oz setup
Ensure consistent AI responses across its two interaction modalities
Drawcto: a multi-agent for collaborative drawing
We investigate the research question in the context of collaborative line drawing using Drawcto- a co-creative drawing system1. Drawcto facilitates open-ended, improvisational interactions where each new stroke is influenced by the selected AI agent and the most recent strokes.
For this study, we used the Grammar Agent of the Drawcto, which utilizes shape grammar theory and visual design principles for creating its strokes.
Interaction modalities for collaborative drawing
We explored how different modalities with AI agents affect perception of the co-creative process.
We included a control condition of human-human collaboration to create a benchmark for effective collaborative drawing since humans naturally collaborate well.
Human - Web
Human - Robot
Human - Human (control)
Wizard of Oz method
To ensure consistency across modalities, we used the Wizard of Oz method, where a human 'wizard' simulated the grammar agent's responses, ensuring controlled and comparable 'Al' responses across modalities.
Thus, conducting the study involved three individuals: a facilitator and two wizards (one for each interface), with roles rotated between studies to minimize bias.
We recruited 10 pairs of university students1 to conduct an empirical study of mixed methods which proceeded in two phases after the onboarding activities2
Key findings reveal an average OCSM curve indicative of typical human-human interactions, varied themes in human-AI collaboration, and a notable influence of AI embodiment on participant perceptions, with the robot interface resembling human-human collaboration more closely than the software interface.
Human-Human
Human-Robot
Human-Web
will be updated soon...
Based on the findings, we identified design implications for both the modalities of Drawcto to bring it closer to the human co-creative experience.
