Artificial intelligence (AI) is both an object of governance and an actant which governs i.e., it is both the governor and the governed. The paper develops the growing field of AI auditing as an instance of the former and a symptom of the latter. By limiting the premise to the ‘how’ of governance, the most common tactic of non-profit, think-tank discourse on tech policy is to reduce the problem to ‘who will govern’ wherein ‘community governance’ becomes the most obvious radical demand. The paper builds on an ongoing project to build an AI auditing tool and an ontology of AI auditing for non-expert auditors, to stress the limits of what is constituted as ‘auditing’ or ‘governing’. By delving into the notion of AI auditing as governance, the paper argues that the ‘interesting’ challenge is not how AI will be governed, but rather why AI needs to be governed at all. Once the question is posed in a manner of ‘why govern?’, accepted and common-sense notions such as transparency, fairness, and accountability, the key ‘ethical’ framings that are currently in circulation in the field of AI ethics open up as categories which need to be unpacked. We argue that these ‘ethical’ frames which have garnered large traction in computer science and adjacent conferences and publications in recent years bracket off the social in a particular way and obfuscate the political to construct and produce AI as an object of governance.
