Introduction
While virtual companions have made significant strides in mimicking human conversation, emotion, and memory, they often lack "social motivation." Most current AI assistants are reactive, waiting for commands or responding to a user's mood. In an educational context, this passivity is a missed opportunity. Human learning is driven by curiosity—the internal urge to resolve uncertainty or explore the unknown. For a virtual companion to be truly socially intelligent and effective over the long term, it must possess its own sense of curiosity to proactively drive active learning and engagement.
Modeling the Curiosity Cycle
The researchers base their work on a psychological model of curiosity that distinguishes between "diversive curiosity" (the search for variety to alleviate boredom) and "specific curiosity" (the drive to acquire information to resolve a specific knowledge gap).
As illustrated in Figure 1, the process begins with a propensity to be curious, which leads to the discovery of novelty. This discovery stimulates specific curiosity, prompting in-depth exploration that ultimately results in enhanced understanding. The Curious Agent acts as a catalyst in this cycle, helping the user discover novelty and stimulating the transition to specific curiosity.
System Architecture
The proposed "Curious Companion" (CC) architecture is divided into three functional modules designed to perceive the environment and stimulate the user.
- Perception Module: This monitors the "User Context" (learning progress, personality traits, and mood) and the "Environment Context" (available learning activities in the virtual world).
- Cognition Module (The Novelty Detector): This serves as the brain of the curiosity engine. It uses Fuzzy Cognitive Maps (FCMs)—graph-based structures representing concepts and causal relationships—to model both the learner’s understanding and the expert knowledge base. By comparing these maps, the CC identifies concepts the learner hasn't encountered or knowledge that contradicts the learner's current understanding.
- Curiosity Stimulation Module: Once a "surprise" is detected, the CC waits for a heuristic time interval to see if the user explores it naturally. If not, it generates a persuasive prompt, such as a thought-provoking question, tailored to the user’s curiosity profile.
Methodology and Results
The effectiveness of the CC was evaluated through a field study involving Grade 8 students at Catholic High School in Singapore. The study utilized a custom-built knowledge base regarding plant transport systems, represented via FCMs, within a 3D Virtual Learning Environment (VLE).
Key findings include:
- Interest Retention: The group with the Curious Companion reported a 25.8% improvement in interest retention compared to the control group ($t = 2.897, p < 0.05$).
- User Preference: Students preferred using FCMs to record their knowledge over traditional rule-based systems, finding the symbolic representation more intuitive.
- Qualitative Impact: Students reported being genuinely surprised by the CC’s proactive prompts, with some following the agent to different locations specifically to see what it would suggest next.
Conclusion and Future Work
This research demonstrates that curiosity is a vital social skill for AI, allowing agents to influence user behavior and interest proactively. While the current system relies on heuristic timing and text-only prompts, future iterations aim to incorporate dynamic mood-to-behavior mapping and multi-modal stimulation (audio and animation) to further enhance the companion's social intelligence and scalability.
Summary updated