FINALS. FACING DATA-DRIVEN MACHINE AGENCY In his magnificent double history of Face and mask, Xxxx Xxxxxxx (2017, 6) tells us that ‘In this book the face is the cynosure of all images, which are always subject to time and thus break down and lose the competition with the living face when confronted with the impossibility of representing it accurately’. As San Francisco, where I am finalising this chapter, is on the verge of banning the use of facial image recognition technologies by local police, the confrontation between human and machine agency is taken to a new level. The famed techniques of ‘deep learning’ (DL) that were once expected to result in general artificial intelligence (GAI), have made major leaps in terms of image and speech recognition. That is why this ban has been proposed. But it turns out that the accuracy of DL collapses with minor tweaks of their objects (based on adversarial machine learning, see x.x. Xxxxx 2018). This kind of machine agency does not recognize faces as we do, and functions best when trained against its own kind, given a set of rules that define a closed game (as demonstrated by DeepMinds’ AlphaZero, Wu 2018). This exemplifies the gap between subject and image, face and portrait, presence and representation, self and me; it also shows us the gap between a pixelated capture of a living face and its ephemeral embodied object. Though I still believe we should learn to take the intentional stance towards data-driven machine agency (The End(s), 6.5.1 and endnote 69), I would now emphasize a better understanding of how that agency differs from our own, of the limits of computer science and machine learning and of how we can learn to respect those limits without sug- gesting that we should reject either computer science or data-driven agency. NOTES
Appears in 1 contract
Samples: End User Agreement
