Artificial intelligence (AI) has the potential to solve important societal problems and has already found broad application, but there remain substantial challenges. Problems, datasets and algorithms are often selected and developed without sufficient consultation with stakeholders and domain experts, and without sufficient engagement with system engineers, social scientists, and legal and public policy experts. As a result, important societal problems may be overlooked, resulting in unforeseen real-world complications that limit scalability and performance. Resulting technologies may raise data ownership issues and have negative impacts on society, such as unacceptable incursions on privacy or perpetuation of bias and inequity.