Co-Creative Sketching and User Agency in Drawing Apps

ISEF Category: Technology Enhances the Arts

Subcategory: Other  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

What happens when a drawing app starts finishing your sketch before you do? Your hand makes one choice, and the model makes the next one. That tug-of-war can change how much control you feel over the final picture. You can measure that feeling with a real questionnaire, not just opinions.

What Is It?

This project studies co-creative sketching, which means you and an AI tool create art together. The app sees your partial drawing, then suggests the next stroke based on a target style. Think of it like a smart sketch partner that keeps guessing your next move. The big question is whether that help feels useful or bossy.

Your key outcome is user-perceived agency, which means how much control people feel over their own actions. If the app fills in too much, users may feel like the drawing belongs to the machine. If the app gives just enough help, users may feel faster and more creative without losing ownership. You can compare that feeling with a full auto-completion version, where the system takes over more of the drawing.

Why This Is a Good Topic

This is a strong science fair topic because you can test a clear human response to a clear design change. You are not guessing whether the app feels better, you are measuring it. The project connects to human-computer interaction, creative tools, and AI design. You can also learn how to run a fair user study, collect survey data, and compare two interface styles with statistics.

Research Questions

• How does partial stroke suggestion affect user-perceived agency compared with full auto-completion?
• What is the effect of style strength on perceived control while drawing?
• Does the timing of stroke suggestions change how much ownership users report?
• To what extent does prior drawing experience change agency ratings in a co-creative sketch app?
• Which interface design leads to higher trust, partial suggestion, or full auto-completion?
• How does the amount of AI editing in the final image affect whether users say the sketch still feels like their own?

Basic Materials

• Tablet or touchscreen laptop with stylus support.
• Drawing app or prototype interface that can show partial stroke suggestions.
• Survey form for the Sense of Agency questionnaire.
• Consent form and participant instructions.
• Spreadsheet software such as Google Sheets or Excel.
• Timer or screen recording tool for session logs.
• Digital note-taking app for observation notes.

Advanced Materials

• Tablet or pen display with pressure sensitivity.
• Prototype app built in Unity, Flutter, or a web interface.
• Machine learning model access for stroke prediction or diffusion-based generation.
• Local server or cloud notebook for running model inference.
• Image storage system for versioned sketch outputs.
• Statistical software for mixed-effects models or ANOVA.
• Screen capture software for interaction analysis.
• Optional eye-tracking device for attention and hesitation measures.

Software & Tools

• Google Forms: Collects agency ratings and open-response feedback from participants.
• Google Sheets: Organizes survey data and basic summary statistics.
• R: Runs statistical tests and visualizes differences between interface conditions.
• Python: Supports prototype testing, data cleaning, and simple plotting.
• ImageJ: Helps measure sketch coverage, stroke density, and image change across conditions.

Experiment Steps

1. Define the exact feeling you want to measure, then choose one clear agency metric and one backup metric.
2. Decide the comparison you will make, such as partial stroke suggestion versus full auto-completion.
3. Design matched sketch tasks so each participant faces the same drawing challenge under each condition.
4. Plan controls that keep style, device, and instructions consistent across sessions.
5. Build a scoring plan that combines questionnaire results, task performance, and any interaction logs.
6. Predefine your statistics so you know how you will compare conditions before you collect data.

Common Pitfalls

• Letting the AI condition change more than one thing at once, which makes it hard to know what caused the agency shift.
• Using messy or inconsistent sketch prompts, which can change user feelings for reasons unrelated to the interface.
• Asking leading questions on the questionnaire, which pushes participants toward the answer you want.
• Mixing first-time doodlers with experienced artists without tracking skill level, which can hide a real effect.
• Measuring only final image quality and ignoring perceived control, which misses the main question of the project.

What Makes This Competitive

A stronger project goes beyond a simple preference survey. You can separate the effect of assistance level, style strength, and timing, then test whether those factors interact. You can also compare different user groups, such as novice and experienced drawers. Careful controls, preregistered analysis, and a clean questionnaire design can make the study feel much more serious.

Project Variations

• Test whether agency changes more for cartoon, realistic, or abstract target styles.
• Compare stylus input with finger input to see whether device precision affects control.
• Study whether users feel more ownership when the model suggests lines, colors, or full shapes.

Learn More

• MIT OpenCourseWare, Introduction to User Interface Design: Search MIT OpenCourseWare for lectures on human-computer interaction and interface evaluation.
• NIH PubMed: Search for review articles on sense of agency, human-AI collaboration, and creative support tools.
• ACM Digital Library: Search for papers on co-creative systems and human-computer interaction in drawing or design.
• Google Scholar: Search for recent studies on AI-assisted creativity, user agency, and mixed-initiative design.
• Stanford HCI Group publications: Look for free papers and project pages on human-AI interaction and user studies.