Artificial intelligence (AI) has the potential to make health care more effective, efficient and equitable. AI applications are on the rise, from clinical decision-making and public health, to biomedical research and drug development, to health system administration and service redesign. The COVID-19 pandemic is serving as a catalyst, yet it is also a reality check, highlighting the limits of existing AI systems. Most AI in health is actually artificial narrow intelligence, designed to accomplish very specific tasks on previously curated data from single settings. In the real world, health data are not always available, standardised, or easily shared. Limited data hinders the ability of AI tools to generate accurate information for diverse populations with potentially very complex conditions. Having appropriate patient data is critical for AI tools because decisions based on models with skewed or incomplete data can put patients at risk. Policy makers should beware of the hype surrounding AI and identify and focus on real problems and opportunities that AI can help address. In setting the foundations for AI to help achieve health policy objectives, one key priority is to improve data quality, interoperability and access in a secure way through better data governance. More broadly, policy makers should work towards implementing and operationalising the OECD AI Principles, as well as investing in technology and human capital. Strong policy frameworks based on inclusive and extensive dialogue among all stakeholders are also key to ensure AI adds value to patients and to societies. AI that influences clinical and public health decisions should be introduced with care. Ultimately, high expectations must be managed, but real opportunities should be pursued.