Research question
Bagaimana visual presence dari pedagogical agent memengaruhi learning outcomes pada task yang berbeda? Fokusnya: apakah agent visibility lebih membantu untuk task yang butuh imitation/motor modeling dibanding task yang lebih abstract/cognitive.
“Visual agent works best when the task needs modelling, not when the task needs headspace.”
Study design (2×2 mixed factorial)
Studi ini memakai 2×2 mixed factorial design:
- Agent visibility: visible vs absent
- Task type: psychomotor vs cognitive
Dependent variables: task performance, self-efficacy, recall, affective beliefs.
Participants: 157 total participants.
Tasks & materials
Learning material berbasis Tetris, dibuat dua varian:
- Psychomotor task: kontrol game via hand gestures
- Cognitive task: kontrol game via mathematical problem-solving
Agent dibuat sebagai 3D model, visibility diatur (agent visible vs no-agent).
[INSERT IMAGE] Figure “Screenshots of Tetris tasks” (psychomotor vs cognitive)
[INSERT IMAGE] Figure “Agent model & visibility settings”
AI-based scoring
Psychomotor performance dinilai via AI-based gesture scoring. Validasi dilakukan dengan membandingkan AI scoring vs manual scoring:
- Pearson correlation r = .8418 (N = 152)
- Tidak ada perbedaan signifikan antara mean manual vs AI scores
- MAE & RMSE: 0.684 (AI scores aligned dengan manual)
Key findings (what changed, where it matters)
1) Psychomotor tasks benefit from visible agent
Temuan utama di abstract dan conclusion: visual agent meningkatkan performance & self-efficacy pada psychomotor tasks, serta memperbaiki affective responses.
2) Cognitive tasks: mixed / potentially overload
Untuk cognitive tasks, efek visual presence mixed dan secara teori bisa menambah cognitive load—agent tidak selalu membantu recall atau performance.
3) Recall: not significantly affected by agent visibility
Tidak ada significant interaction effect untuk recall antara visibility × task type.
Practical implications (design rule-of-thumb)
- Pakai visible agent untuk learning environment yang butuh physical imitation / motor skills (psychomotor).
- Untuk cognitive tasks, agent visibility perlu dibatasi agar tidak menambah cognitive load.
“Use visual agents when the task needs imitation. Minimize them when the task needs thinking.”
What I’m proud of
- 2×2 factorial design yang jelas dan measurable outcome.
- AI-based scoring dengan validation against manual scoring (reducing bias).
- Translating theory (social-cognitive, multimedia learning, cognitive load) menjadi keputusan desain yang actionable.