AI for the Common Good?! Pitfalls, Challenges, and Ethics Pen-Testing

Recently, many AI researchers and practitionershave embarked on research visions that involve doing AIfor “Good”. This is part of a general drive towards infus-ing AI research and practice with ethical thinking. One fre-quent theme in current ethical guidelines is the require-ment that AI be good for all, or: contribute to the Com-mon Good. But what is the Common Good, and is it enoughto want to be good? Via four lead questions, I will illus-trate challenges and pitfalls when determining, from an AIpoint of view, what the Common Good is and how it can beenhanced by AI. The questions are: What is the problem /What is a problem?, Who defines the problem?, What isthe role of knowledge?, and What are important side ef-fects and dynamics? The illustration will use an examplefrom the domain of “AI for Social Good”, more specifically“Data Science for Social Good”. Even if the importance ofthese questions may be known at an abstract level, they donot get asked sufficiently in practice, as shown by an ex-ploratory study of 99 contributions to recent conferencesin the field. Turning these challenges and pitfalls into apositive recommendation, as a conclusion I will draw onanother characteristic of computer-science thinking andpractice to make these impediments visible and attenuatethem: “attacks” as a method for improving design. Thisresults in the proposal ofethics pen-testingas a methodfor helping AI designs to better contribute to the CommonGood