Healthcare systems are in crisis around the world. They face greater demand for increasingly expensive treatments from a population that is ageing and has numerous unhealthy habits. After many decades of improvement, the statistics for healthspan—the number of years people live in good health—seem poised to go into reverse. This is a complex problem and there is no “magic bullet” solution. However, artificial intelligence (AI) can play many roles in reversing these unhappy trends and in boosting population health. It can assist diagnosis, treatment selection, treatment compliance, the development and testing of new therapeutics, and, by identifying the special features of each individual patient, enable greater adoption of personalized medicine. More fundamentally, AI can identify earlier signs of the accumulation of underlying biochemical damage, and recommend various “detox” treatments that, in contrast to many products currently marketed under that description, actually do restore vitality. As in the age-old proverb, a stitch in time will save nine. This will benefit populations, not only in terms of individual medical health but in terms of the health of the economy and society as a whole. Nevertheless, over-hasty imposition of untested AI solutions could backfire, sickening populations, and turning the public against the whole field, preventing the attainment of the many benefits it could bring. Accordingly, the application of AI within healthcare needs to be guided and governed by an overarching set of principles, explained in this article—principles that can avoid seven specific failure modes of AI healthcare solutions.