Core Concepts
Large language models can be leveraged to create scalable, accessible, and personalized social skill training environments through an AI Partner for simulated practice and an AI Mentor for tailored feedback.
Abstract
This paper proposes a framework for social skill training that leverages large language models (LLMs) to create an AI Partner for simulated practice and an AI Mentor for tailored feedback. The key insights are:
Simulation-based learning can effectively teach communication, cooperation, and leadership skills, but practice environments are often inaccessible, especially for underrepresented groups. LLMs can create scalable, flexible simulations for social skill practice.
The AI Partner vision uses LLMs to simulate believable characters and scenarios, allowing learners to practice social skills in a risk-free environment. The AI Mentor vision uses LLMs to provide personalized, theory-grounded feedback to enhance the learning process.
Integrating the AI Partner and AI Mentor (APAM) can merge experiential learning with realistic practice and tailored guidance, making social skill training more accessible and effective.
Deploying APAM systems requires addressing technical challenges around LLM consistency, grounding, and personalization, as well as carefully integrating expert frameworks and enabling user control.
Evaluating the impact of APAM systems should go beyond intrinsic NLP metrics to include extrinsic measures of behavioral change, self-efficacy, and long-term outcomes, drawing on expertise from education and economics.
Overall, the APAM framework presents a promising approach to democratize access to social skill training and development using the capabilities of large language models.