Taking Competitive Advantage: GANS and the Inherent Ethics of AI

• In Event: Media Studies Interruptions of STS - II

Abstract

Recent scholarship has convincingly demonstrated that artificial intelligence algorithms and other computational agents frequently, and perhaps inevitably, embed the social, cultural and political biases of their creators. While deep-learning systems cloak their conclusions in an apparent ability to self-correct and objectively iterate on their determinations, there are increasing indications that they ‘learn’ the mistakes of human systems all to well. Generative Adversarial Networks, or GANS, for example, utilize a pair of neural networks competing with one another to produce astonishing accuracy in previously difficult tasks from image and speech recognition to video analysis. In simple terms, they deploy a basic game scenario to develop techniques for decoding and manipulating images and other forms of media. Based on principles of competition and obfuscation, such systems have demonstrated themselves equally capable of both representation and misrepresentation, decoding the contents of real images and convincingly creating fake images that appear incredibly real. In the process they have also provided their creators and the companies who fund this research with an incredible, if opaque, competitive advantage.

This paper will argue that GANS and other ai technologies currently being developed for analyzing information and media are a direct product of the socio-technical assemblages at work in the research centers which develop them (Silicon Valley and elsewhere.) While the ethics of automation and unemployment have been laid at the feet of the industries who deploy these technologies, less discussed are the ethics reflected in the technologies themselves.

Author

• Kris Fallon, University of California, Davis