Transforming Human-ai Interactions through Reinforcement Learning from Human Feedback and Ai Feedback: a Human-ai Classification Report

AI systems are progressively being implemented across diverse disciplines and application areas. This increase has intensified scientific emphasis and public apprehension regarding the active involvement of humans in the development, operation, and adoption of these systems. Notwithstanding this apprehension, the majority of current scholarship on AI and Human–Computer Interaction (HCI) predominantly on elucidating the functionality of AI systems and, occasionally, enabling users to challenge AI determinations. This research aims to assess the efficacy and dependability of a hybrid feedback-driven learning methodology utilizing a classification model trained on multi-class human-labelled data. The methodology entails encoding diagnostic labels into numerical classes via LabelEncoder and implementing a reinforcement learning framework that incorporates both human-curated and AI-generated feedback. The classification report demonstrates exceptional performance across all categories, with an overall accuracy of 0.99. Precision, recall, and F1-score metrics typically approach 1.00, indicating negligible classification errors and robust generalizability. Class 2 has a somewhat lower precision of 0.94 but 100% recall, which means that there are false positives but no missed real events. The macro and weighted averages for all metrics are 0.99 or higher, which shows that the method works effectively even though the classes are not evenly distributed. The results showed that RLHF and RLAIF make AI decision-making better when there are more than one class. These results have an effect on AI systems that work with people in healthcare, self-driving cars, and personalised decision-making, where accuracy and ethics are very important.