Faking it and Breaking it: Responsible AI is needed Now

Abstract

Frauds, fakes, forgeries… these are not new problems for the public to contend with. Purveyors of designer goods such as handbags and shoes have developed keen skills to spot clever knockoffs, though some copies are so good that their authenticity seems less important to folks buying them to wear. Forging currency is an incredibly serious crime, and yet it is the crux of many movies and stories. In antiquities and in art, making forgeries gets a different sort of respect as doing so effectively requires great skill akin to that practiced by the original artisans themselves. In academia, as well as politics [8], plagiarism is one of the worst offenses one can commit because faking one's intellectual achievements [7] threatens the attainment of credentials and offices while getting away with plagiarism can threaten the institution's reputation. So, with so much practice living life negotiating the real from the fake, why are the latest advancements in machine-based image and text generation so concerning? One reason might have to do with computing's penchant for outsourcing messy, qualitative evaluation methods to external, seemingly neutral metrics. Like all sciences, computing must constantly grapple with the myth of objectivity [15].