When Should Teachers Control AI Generation for Mathematics Visuals?
The study explores when teachers should control AI-generated math visuals, finding post-generation control best for accuracy and predictability.
Key Findings
Methodology
The study employs a mixed-methods approach, designing three control stages: pre-generation, mid-generation, and post-generation. It analyzes the impact of control timing on teacher workflows through experiments with 24 primary math teachers.
Key Results
- Post-generation control scored highest in accuracy and predictability, allowing teachers to verify and modify results directly.
- Pre-generation control supports rapid ideation but reduces perceived agency and predictability.
- Mid-generation control improves structural alignment but increases complexity.
Significance
The research offers new insights into the application of generative AI in education, particularly in math, highlighting the importance of teacher control in ensuring the accuracy of visual materials.
Technical Contribution
Introduces control timing as a design space in AI-assisted visual authoring, validating its impact on teacher workflows and content accuracy.
Novelty
First systematic study on control timing for AI-generated math visuals, filling a research gap in educational applications of generative AI.
Limitations
- The study is limited to primary math teachers, and results may not apply to other subjects or educational levels.
- The experimental setting may differ from real teaching environments, affecting external validity.
Future Work
Future research could extend to other subjects, exploring control timing impacts in diverse educational contexts and integrating more complex AI models.
AI Executive Summary
Generative AI holds great potential in education, especially in math where visual material accuracy is crucial. This study examines when teachers should control AI-generated math visuals, designing three control stages: pre-generation, mid-generation, and post-generation.
Through experiments with 24 primary math teachers, post-generation control was found to score highest in accuracy and predictability, allowing teachers to verify and modify results directly. Pre-generation control supports rapid ideation but reduces perceived agency and predictability, while mid-generation control improves structural alignment but increases complexity.
The findings suggest that in accuracy-sensitive educational tasks, generative tools should align with teacher intent, supporting workflows that combine automation with direct manipulation. This provides important guidance for future educational technology design.
Deep Analysis
Background
With advances in generative AI, particularly text-to-image models, teachers can quickly create customized educational visuals. However, existing tools mainly support prompting and post-hoc editing, which may not meet the high accuracy demands of math education.
Core Problem
In math education, visuals must accurately reflect quantities and relationships. Current generation tools perform poorly in handling counts and structural constraints, making fully automated generation unsuitable for large-scale deployment.
Innovation
This study systematically explores when teachers should control AI-generated math visuals, proposing a design space with pre-generation, mid-generation, and post-generation control stages, and validating their impact on teacher workflows.
Methodology
- �� Designed three control stages: pre-generation, mid-generation, post-generation.
- �� Conducted a mixed-methods study involving 24 primary math teachers.
- �� Analyzed the impact of control timing on teacher workflows and content accuracy.
Experiments
The experiment used a mixed-methods design with 24 primary math teachers, employing three control stages to create visuals for standardized arithmetic problems. Data collection included questionnaires, interaction logs, and semi-structured interviews.
Results
Post-generation control scored highest in accuracy and predictability, allowing teachers to verify and modify results directly. Pre-generation control supports rapid ideation but reduces perceived agency and predictability.
Applications
The findings can inform the design of more effective AI-assisted educational tools, particularly in math education, helping teachers create more accurate visual materials.
Limitations & Outlook
The study is limited to primary math teachers, and results may not apply to other subjects or educational levels. The experimental setting may differ from real teaching environments, affecting external validity.
Plain Language Accessible to non-experts
Imagine a teacher in a kitchen, with AI as her assistant. The teacher can tell the assistant what she wants to cook before starting (pre-generation control), check and adjust the ingredients during cooking (mid-generation control), or taste and adjust the dish after it's done (post-generation control). The study found that the last method makes the teacher more satisfied with the final result, as she can directly taste and adjust.
ELI14 Explained like you're 14
Imagine you're playing a game where you can set up your character before the game starts (pre-generation control), adjust the gear during the game (mid-generation control), or modify the character's appearance after the game ends (post-generation control). The study found that the last method makes players happier because they can directly see and modify the results.
Glossary
Generative AI
AI technology that generates content using algorithms, often used in image and text domains.
Used in the study to generate math visuals.
Text-to-image models
AI models that convert text descriptions into images.
Used to generate educational visuals.
Human–AI interaction
The study of interactions between humans and AI systems.
Analyzed the impact of different control timings on teachers.
Educational visuals
Visual aids used for teaching.
The math visuals generated in the study.
Mathematics education
The field of teaching and learning mathematics.
The study's application background.
Open Questions Unanswered questions from this research
- 1 How can similar control timing designs be applied in other subjects?
- 2 How can the study's findings be validated in real teaching environments?
Applications
Immediate Applications
Math Classroom
Teachers can use generative AI to quickly create accurate math visuals, enhancing classroom teaching effectiveness.
Long-term Vision
Cross-disciplinary Application
Extend control timing design to other subjects, improving the overall level of educational technology.
Abstract
Generative AI has the potential to help teachers rapidly create classroom-ready visual materials, particularly in mathematics where diagrams and visual representations must be pedagogically meaningful and instructionally correct. However, current generative tools primarily support prompting and post-hoc editing, leaving open a key question for correctness-sensitive educational authoring: when in the generation pipeline should teachers exert control? In this paper, we investigate how the timing of human control in AI-assisted generation shapes teachers' visual authoring practices in correctness-sensitive tasks. We introduce a design space of three stages of control: pre-generation control, where users specify intent solely through natural language prompts before generation; mid-generation control, where users inspect and confirm an explicit layout structure before the system completes generation; and post-generation control, where users directly modify AI-generated visuals after generation through object-level edits. In a within-subject, mixed-methods study with 24 primary mathematics teachers, post-generation control received higher ratings on predictability and correctness, while other subjective measures showed no reliable differences. Qualitative findings explain these differences by revealing workflow trade-offs: highly automated, pre-generation control supports rapid ideation but reduces perceived agency and predictability; mid-generation control improves structural alignment at the cost of additional effort; and post-generation control preserves user agency through low-cost, direct verification and correction. Together, these results suggest that in correctness-sensitive educational tasks, effective generative tools should align system behavior with teacher intent and support stage-dependent workflows that combine automation with direct manipulation.