Unpacking the Expressed Consequences of AI Research in Broader Impact Statements

Abstract

The computer science research community and the broader public have become\nincreasingly aware of negative consequences of algorithmic systems. In\nresponse, the top-tier Neural Information Processing Systems (NeurIPS)\nconference for machine learning and artificial intelligence research required\nthat authors include a statement of broader impact to reflect on potential\npositive and negative consequences of their work. We present the results of a\nqualitative thematic analysis of a sample of statements written for the 2020\nconference. The themes we identify broadly fall into categories related to how\nconsequences are expressed (e.g., valence, specificity, uncertainty), areas of\nimpacts expressed (e.g., bias, the environment, labor, privacy), and\nresearchers' recommendations for mitigating negative consequences in the\nfuture. In light of our results, we offer perspectives on how the broader\nimpact statement can be implemented in future iterations to better align with\npotential goals.\n

Ähnliche Arbeiten