CausaDisco: An LLM Dialogue-Based Autonomous Learning Enhancement System Using Aristotle's Four Causes

This post introduces CausaDisco, a dialogue-based autonomous learning system that integrates Aristotle's four causes (material, formal, efficient, final) into LLM prompts. The system aims to reduce learners' cognitive load, promote deep engagement, and foster multi-dimensional understanding. A controlled study shows it enhances interactive attractiveness, stimulates complex exploration, and supports multi-perspective thinking.

LLMs have advanced self-learning tools but face key challenges:

• Cognitive load issues: Information overload, lack of structure, shallow understanding.
• Dialogue dilemmas: Difficulty asking good questions, fragmented conversations, hard to follow up deeply.
• Complex information processing: Trouble connecting concepts, limited multi-angle understanding, application barriers. To address these, the study proposes using Aristotle's four causes as an epistemological framework to guide structured, cognitively supportive dialogues.

Aristotle's four causes explain a thing's existence and have clear learning applications:

• Material Cause: What it's made of (identify basic components, raw data, foundational concepts).
• Formal Cause: Its structure/form (understand definitions, classifications, logical structures).
• Efficient Cause: What causes it (analyze causal relationships, mechanisms, influencing factors).
• Final Cause: Its purpose (grasp learning goals, application scenarios, value connections). These provide a comprehensive, structured cognitive framework for deep learning.

CausaDisco's core is integrating four causes into LLM prompts. Key components:

1. Four Cause Prompt Generator: Selects appropriate causes based on context, generates follow-up questions, balances coverage.
2. Dialogue Manager: Maintains history, tracks explored causes, identifies knowledge gaps.
3. Learner Model: Models knowledge state, cognitive style, personalizes cause presentation. Prompt strategy: System prompt defines the role (four-cause learning assistant) and goals; dynamic questioning uses rotation (balance four causes), adaptation (adjust to learner responses), and coherence (link to prior dialogue).

• Formative Study: 26 self-learners; semi-structured interviews/observations. Findings: Learners have different four-cause preferences; successful learners switch between causes, struggling ones stick to single perspectives.
• Controlled Study: 36 learners split into control (standard LLM) and experimental (CausaDisco) groups. Evaluation metrics: engagement (dialogue turns, active questions), exploration depth (concept connections, multi-angle analysis), learning effects (test scores, application ability).

• Engagement: 42% more dialogue turns, 67% more active questions, 35% longer learning time; learners report deeper thinking and Socratic-style interaction.
• Exploration Complexity: 2.3x more concept connections, more multi-angle analysis, higher-quality follow-up questions.
• Multi-Perspective: Experimental group explores all four causes evenly, shows better cognitive flexibility and overall understanding vs control group (focused on 1-2 causes).

Contributions:

• Theoretical: Expands LLM as educational agents (cognitive scaffolding role of epistemological frameworks).
• Practical: Reusable design patterns (framework-driven dialogue, dynamic questioning) and implementation tips (choose matching frameworks, optimize prompts, multi-dimensional evaluation). Limitations: Small sample size, limited to general knowledge, short duration, fixed framework, limited personalization, text-only interaction. Future Directions: Explore other frameworks, enhance personalization, add multi-modal support, expand to professional fields (medicine/engineering), integrate cognitive science theories (cognitive load, motivation).