Toward HCXAI, Beyond XAI: Along with the Case of Referring Expression Comprehension Under the Personal Context

The goal of eXplainable AI (XAI) is to increase the transparency and trustworthiness of AI algorithms to humans by clarifying their internal decision-making process. As AI technology continues to permeate various aspects of our daily lives, including the workplace, the importance of XAI has grown and some XAI methodologies have increased our understanding of AI algorithm’s inner logic. However, the achievement of explainability of AI models did not make them user-centered in real life and users still need to be aware of the internal mechanism of AI models to interact with them effectively. In this regard, the paper argues there is a need to move beyond XAI towards Human-Centered eXplainable AI (HCXAI) to ensure human-centered usage in our daily lives. As the steps heading for HCXAI, the paper suggests researching intuitive human behaviors first and then training AI models to comprehend these natural human behaviors. When users interact with AI models trained in this way, they will not need to think consciously about the model’s working mechanism anymore. The paper elaborates on this approach further with a practical case that can be encountered in our daily lives; the vision and language model for referring expression comprehension under the personal context.