Artificial intelligence, data science and healthcare improvement: what has actually changed and has it changed enough?

Artificial intelligence (AI) and data science are reshaping how healthcare systems detect risks, generate insights and organise improvement. This paper asks whether these developments have changed healthcare improvement enough, conceptually and practically, or whether they have primarily provided faster tools to monitor and detect, with longstanding challenges remaining endemic. The argument advanced is that AI and data science have altered the means of improvement more than the actuality of improvement. Four propositions are developed. First, healthcare improvement has begun to shift from periodic measurement to continuous sensing, but continuous sensing does not automatically produce continuous improvement. Second, the field has moved from descriptive analytics and dashboards towards probabilistic prediction and generative decision support, but prediction is not prevention. Third, the unit of improvement has started to move from isolated projects towards learning health systems, yet learning systems are not built by software and data science alone. Fourth, the risks introduced by AI, including bias, opacity, automation errors, hallucination, deskilling and misaligned incentives, have pushed governance, and the values and biases encoded in AI, to the centre of improvement work. To succeed, we will need to continuously educate the workforce, build better-trained AI models, generate a databank of generalisable exemplar projects, ensure we embrace exchange of expertise, collaboration and sharing rather than competition and underpin progress with transparency, rigorous evaluation and well-crafted research. Governance and ethical oversight need considerable attention as they remain underdeveloped. The paper concludes that AI’s greatest contribution may be to strengthen the feedback loops through which health systems learn, but that outcomes will still depend on implementation capability, organisational learning, workflow design and accountability.