AI Sandbox: A Secure and Controllable Human-AI Interactive Platform for Cross-Disciplinary Research and Education through Large Language Models

Recent advancements in Artificial Intelligence (AI) have made significant improvements in modeling existence and creating generalizations by learning natural patterns from the physical world and interacting with humans. It has become increasingly beneficial in many areas such as smart manufacturing, healthcare, intelligent transportation, and education. In this work, we present the design and development of an AI sandbox, a secure and controllable human-AI interactive platform, to advance cross-disciplinary research and education with large language models (LLMs). The developed AI sandbox features its integration and embodiment with generative AI, mainly including a multi-function graphical user interface (GUI), a reliable and security-oriented infrastructure, and a multi-LLM-enabled engine system. It can provide users with a scalable environment for developing, testing, and deploying AI applications without user-provided data being employed for public model training. With the incorporation of multiple LLMs, the friendly GUI allows users to easily access and switch between them. The system can effectively employ a variety of methods for balancing the security needs of future users while still providing rich learning opportunities that come with generative AI. Implementation results and analysis of the AI sandbox suggest that its human-centered design has promising potential in catalyzing AI-enabled creations and applications, advancing responsibility, explainability, trustworthiness, accessibility, and diversity in human-AI interaction and collaboration.