Solving the Single-Vehicle Self-Driving Car Trolley Problem Using Risk Theory and Vehicle Dynamics

Science and Engineering Ethics https://doi.org/10.1007/s11948-019-00102-6 ORIGINAL RESEARCH/SCHOLARSHIP Solving the Single‑Vehicle Self‑Driving Car Trolley Problem Using Risk Theory and Vehicle Dynamics Rebecca Davnall1 Received: 1 August 2018 / Accepted: 27 March 2019 © The Author(s) 2019 Abstract Questions of what a self-driving car ought to do if it encounters a situation analogous to the ‘trolley problem’ have dominated recent discussion of the ethics of self-driving cars. This paper argues that this interest is misplaced. If a trolley-style dilemma situation actually occurs, given the limits on what information will be available to the car, the dynamics of braking and tyre traction determine that, irrespective of outcome, it is always least risky for the car to brake in a straight line rather than swerve. Keywords Self-driving cars · The trolley problem · Automation · Ethics · Risk · Vehicle dynamics A Generic Self‑Driving Car Dilemma Recent high-profile advances in self-driving car technology have prompted a wave of interest in the ethical implications of the design and deployment of these systems. This attention has coalesced around a set of ethical dilemmas which, especially in the popular imagination,1 are held to be broadly analogous to the trolley problem (Foot 1967). In some cases this framing has been used to suggest that the development of road-ready self-driving car software will require ‘solving’ the trolley problem itself. The dilemmaic framing of these cases is a mistake born of failure to appreciate important and well-established empirical data from the field of vehicle dynamics. In effect, all such hypotheticals as they apply to self-driving cars in the real world will be optimally resolved if the car in question performs its best emergency stop procedure. The car will not actually face the kind of dilemma envisaged, because no 1 Nyholm and Smids collect some representative examples (2016: 1276, especially footnote 1). * Rebecca Davnall 1 Department of Philosophy, University of Liverpool, Brownlow Hill, Liverpool L69 7ZX, UK 13 Vol.:(0123456789) R. Davnall action available to the car, given the information it is reasonable to anticipate will be available to it, will be clearly morally preferable or equal to maximally-efficient braking in a straight line. This solution, set out in detail in the third section of this paper, is sound for cases which involve only a single vehicle; additional vehicles, regardless of whether they are self-driving or not, introduce complexities which are briefly addressed in the final section. This is an important limitation of the solution. In another sense, however, this solution is quite general: it will be established below—again on practical grounds—that very little information about the specific scenario apart from the number of vehicles involved is relevant to the question of what the vehicle ought to do. To that end, it will be useful to establish a minimally-detailed version of the hypothetical. Most of the clear examples in the literature are quite vivid, featuring details which must be examined later; at this point a representative sample of scenarios are presented only to draw out their common elements. A commonly-cited and clear example is Jan Gogoll and Julian F. Müller’s tunnel case: Imagine you are sitting in your autonomous car going at a steady pace entering a tunnel. In front of you is a school bus with children on board going at the same pace as you are. In the left lane there is a single car with two passengers overtaking you. For some reason the bus in front of you brakes and your car cannot brake to avoid crashing into the bus. There are three different strategies your car can follow: First, brake and crash into the bus, which will result in the loss of lives on the bus. Second, steer into the passing car on your left—pushing it into the wall, saving your life but killing the other car’s two passengers. Third, it can steer itself (and you) into the right hand sidewall of the tunnel, sacrificing you but sparing all other participants’ lives. (2017: 683) Patrick Lin offers the following version: Imagine in some distant future, your autonomous car encounters this terrible choice: it must either swerve left and strike an 8-year old girl, or swerve right and strike an 80-year old grandmother. Given the car’s velocity, either victim would surely be killed on impact. If you do not swerve, both victims will be struck and killed; so there is good reason to think that you ought to swerve one way or another. (2016: 69–70) In a Stapledon lecture at the University of Liverpool, Fiona Woollard offered two examples: Swerve: An autonomous car with two passengers (mother and child) is driving at the speed limit on a 40 mph road when three drunken pedestrians stumble into the road. The only way to avoid hitting the pedestrians is to swerve the car into a wall, risking the life of the mother and child. Tree: A tree suddenly falls into the road in front of a driverless car carrying 5 passengers. The only way to avoid hitting the tree is to swerve onto one of 13 Solving the Single-Vehicle Self-Driving Car Trolley Problem… the pavements. A single pedestrian is walking on the right hand pavement. A crowd of school children is waiting for a bus on the left hand pavement. (2017) And Sven Nyhom and Jilles Smids offer the truck case: A self-driving car with five passengers approaches a conventional car (e.g. a heavy truck) that for some reason suddenly departs from its lane and heads directly towards the self-driving car. In a split-second, the self-driving car senses the trajectory and the likely weight of the oncoming truck. It calculates that a high-impact collision is inevitable, which would kill the five passengers, unless the car swerves towards the pavement on its right-hand side. There, unfortunately, an elderly pedestrian happens to be walking, and he will die as a result if the self-driving car swerves to the right and hits him. This is the sort of situation in which the human passengers of a self-driving car cannot take control quickly enough. (2016: 1278) These examples can all be described as variants of the trolley problem because they all involve a decision between clearly-delineated options, each of which at least implicitly leads to certain harm (or in these cases, certain collisions), and where the chief difficulty arises from a lack of general consensus about which harms would be worse. Cases of this kind are supposed to raise questions of which option to take and which harm thereby to cause. The number of options said to be available to the car differs among the five cases. In the tunnel case, Lin’s case and Woollard’s tree case, there are three options—to continue straight, to swerve left or to swerve right. In Woollard’s swerve case and the truck case, there are only two: swerve, or continue straight on. It can be safely assumed in the latter two cases that the swerve is in the direction (...truncated)