Send message to Author

Paper short abstract

This paper draws on historical debates about AI and rules to interpret some of the dilemmas of modern AI. If the practical ambiguity of rules once was thought to proliferate loopholes and preclude machines from thinking, today this ambiguity might present serious barriers to AI governance.

Paper long abstract

Sometime in 1949, Alan Turing, gave the philosopher Michael Polanyi a newspaper clipping of a horse race photo finish. In the photo, the two horses are in a ‘dead heat,’ both appearing to cross the line at the same moment, meaning that the (human) judges ultimately had to make the call. According to Polanyi, Turing gave him the photo as a memento of their friendly disagreement over whether rule-based systems could mimic human judgement or if there would always be a remainder, an excess to rules which could only be filled with tacit knowledge.

This ambiguity of rules was long seen as an insurmountable barrier to machine intelligence. It also made possible the easy-to-exploit loopholes and “computer says no” moments which persist in rule-based systems. Modern AI, in the form of machine learning, is no longer strictly rule-based or logical, instead its decision-making power arises from spotting patterns in large data sets. Now like Polanyi’s tacit knowledge, AI accomplishes tasks through repetition and feedback without needing to articulate the rules. But this begs the question: what happens when we need rules to govern them?

This paper draws on these historical debates about AI and rules and the example of sports technologies like the photo finish (Finn 2022) and goal line technologies (Collins and Evans 2013) to understand how rules are practically, materially and socially implemented as well as how they are circumvented and exploited, in order to shed new light on debates about AI Ethics, Fair ML and Value Alignment.