From MYCIN to MedGemma: A Historical and Comparative Analysis of Healthcare AI Evolution

The evolution of artificial intelligence (AI) in healthcare has transitioned through distinct technological eras, each marked by unique advancements and challenges. This article provides a comprehensive histor-ical and comparative analysis of healthcare AI assistants, from early rule-based systems like MYCIN in the 1970s–1980s to contemporary large language models (LLMs) such as Med-PaLM and MedGemma, and explores emerging adaptive AI frameworks. Rule-based systems offered transparency and interpretability but were limited by brittleness and scalability. The machine learning (ML) era introduced data-driven approaches, improving predictive analytics but raising concerns about bias and explainability. The 2020s saw the rise of LLMs, enabling conversational AI for clinical triage and patient education, though halluci-nations and safety risks emerged. Future adaptive AI systems promise real-time personalization and con-tinual learning but lack empirical validation. The study synthesizes technical architectures, functional applications, and evaluation metrics across eras, highlighting gaps in cross-era benchmarking and inte-grated governance. Ethical and regulatory challenges have also evolved, from liability concerns in rule-based systems to bias and fairness in ML, and now to safety and alignment in LLMs. Despite progress, fragmentation persists in the literature, with limited comparative analyses and a focus on provider-facing tools over patient-oriented applications. This review underscores the need for unified frameworks to evaluate performance, ensure ethical compliance, and guide the development of next-generation AI in healthcare. By addressing these gaps, the field can better harness AI’s potential to transform clinical prac-tice while mitigating risks.