ChatGPT in the Generation of Explanations for Cyber-Physical Systems (#1588)

Authors

Peña-Cáceres, Oscar

Silva-Marchan, Henry

Espinoza-Nima, Rudy

Garay-Silupu, Elvis

Ricalde-Morán, Dania

Alvarado-Paiva, Douglas

Abstract

The research focused on designing a mechanism for the generation of explanations in the smart home environment through ChatGPT with the purpose of dynamizing the interaction between humans and cyber-physical systems. A simulator was developed that integrates different services and explanation objects related to the smart home environment. Additionally, the simulator interacts with an artificial intelligence model to determine the level of attention required by the user, the type of explanation, and the interaction mechanism. ChatGPT's text-davinci-003 model was used to generate explanations. For the evaluation, a checklist was applied to 20 professionals linked to the study area in order to determine the quality of the system and the effectiveness of ChatGPT in the generation of explanations for cyber-physical systems in the context of the smart home. The system provided favourable results, with outstanding scores in explanation quality (95%), user-system interaction (90%), and adaptability and robustness (95%). The explanations varied between 15 and 50 words, depending on the scenario. Each explanation generated by the system has demonstrated clarity and comprehension. The "what" explanations are shorter, and the "why" explanations are longer. It is suggested to explore as an alternative the gpt-3.5-turbo-instructions and Davinci-002 models for possible improvements in conciseness and empathy. As a future perspective, the integration of new types of explanation is considered, addressing aspects such as "why not" and providing details on "how" to carry out the tasks.