Calls for a more human-centred approach to AI and machine learning have been constant in recent years. People, companies, NGOs, and political bodies are all demanding that AI-powered systems and products account for individuals' as well as society's well being through responsible applications of these technologies, data and privacy protections, explainability of systems, prevention of bias, and equal access for all. While there has been early progress in meeting these demands, much more remains to be done. In this talk, we'll review the approach Google has taken on human-centred AI with a focus on the efforts of the People + AI Research group (PAIR). From that vantage point, we'll then examine why these efforts may not be enough and if a new approach to building these socio-technical systems - "relational design" - may be needed.