From The Turing Test To The Trolley Problem: Tests, Measures And The Standardisation Of Machine Intelligence.

• In Event: Critical Data Studies: Ethics and Human Contexts of Data Science II

Abstract

'Machine intelligence' has been known through tests and games, and being challenged by humans: from the Turing Test to Chess, Go, and Jeopardy; to driving cars and demonstrating ‘ethics’. My research takes the application of thought experiments such as the Trolley Problem, which frames a popular media discourse on the ‘ethics of autonomous driving’. But what does it mean for machine learning to pass or fail a [driving] test? And in relation to what standard? I theorise tests of machine learning as Barad-ian ‘apparatuses’, and follow the intra-actions through which tests are co-constitutive with machine learning as it is applied to autonomous driving. Automobility is a heavily regulated social activity, so my work is curious about the epistemic and socio-technical mechanisms by which tests intra-act with governmentality and regulation. I claim that despite anxieties about the need for regulation of AI through ‘ethics’, machine learning is in fact being trained to eventually become its own regulator through the very use of tests. Additionally, that there is a shift being made from ‘ethics’ to 'accountability' for crashes, and accountability is materialised through measures that enable standardisation. While standardisation manages a new and emerging technology, it also black-box-es it further within regulatory bureaucracies and codification under the law; and in doing so shapes 'objective' and normative notions of machine autonomy and agency. My paper will discuss these claims through the presentation of on-going fieldwork, and my own positionality as a researcher working with actors in the German auto industry.

Author

• Maya Indira Ganesh, Leuphana University, Germany