Can AI Replace Doctors? Efficient Neural Networks for Response Classification in Health Consultations

Abstract

In the rapidly evolving landscape of healthcare, artificial intelligence (AI) is increasingly integrated into medical consultations. While AI offers the potential for scalable and efficient healthcare solutions, its dependency on large and accurate datasets raises concerns about reliability and patient safety. This research introduces MEDXNET, an advanced neural network designed to classify whether a response in a health consultation is generated by a medical professional or an AI system. MEDXNET leverages a hybrid architecture combining Bidirectional Long Short-Term Memory (BiLSTM), Transformers, and one-dimensional Convolutional Neural Networks (CNN1D) to capture both local and contextual dependencies in medical text data. TF-IDF is employed for effective vectorization. Trained on a custom-labeled dataset named MEDIC, the proposed model is benchmarked against traditional deep learning models including BiLSTM, GRU, and LSTM. MEDXNET demonstrated superior performance with an accuracy of 95%, significantly outperforming BiLSTM (91.78%) and GRU (91.16%). Moreover, an extended CNN2D variant further improved classification accuracy to 96.78%. This innovative tool empowers users to assess the origin of medical advice—human or AI—thereby fostering trust, accountability, and safety in digital healthcare interactions. The findings have substantial implications for the deployment of AI in sensitive medical domains.