It’s Time to Rethink Levels of Automation for Self-Driving Vehicles
In 2017, Waymo CEO John Krafcik told tech conferences that “Fully self-driving cars are here.” The next year, however, at a Wall Street Journal panel with the title “Are we there yet?” Krafcik said that so-called “Level…
In 2017, Waymo CEO John Krafcik told tech conferences that “Fully self-driving cars are here.” The next year, however, at a Wall Street Journal panel with the title “Are we there yet?” Krafcik said that so-called “Level Five” self-driving cars were impossible: “I’m not sure that we’re ever… going to achieve an L5 level of automation… I think it’s sort of silly that we think about it. And it’s important I think for all of us to be really clear on the language around self-driving because it does end up confusing people… autonomy I think is always going to have some constraint on it” [1]. Survey evidence suggests that consumers are indeed confused about whether they can currently buy a vehicle that is “self-driving” [2]. This muddle is not because the public are ignorant. It is because one of the major ways in which the development of self-driving cars has been discussed — the levels of automation drawn up by the Society of Automotive Engineers (SAE) — is misleading. A typology originally developed to provide some engineering clarity now benefits technology developers far more than it serves the public interest. We are social researchers who have over the last three years worked with and interviewed the developers of this technology. We argue that the levels of automation need a rethink. The SAE levels, by emphasizing autonomy and implying that progress means more autonomy, do little to inform public decision-making about the conditions in which these technologies might have meaningful benefits. Self-driving cars could be a transformative technology in both good and bad ways. The important questions are not to do with when they will arrive but where , for whom, and in what forms they will appear. If we want a clearer sense of the possibilities from automated vehicle systems, we need to broaden our gaze [3]. Rather than emphasizing the autonomy of self-driving vehicles, we should instead be talking about their conditionality. We need to know about the circumstances in which different systems could have an impact on our lives. Self-driving vehicle systems will serve different purposes and take on different shapes in different places. A schema for innovation that points in one direction and says nothing about the desirability of the destination makes for a poor roadmap. Where Did the SAE Levels Come From? In order to represent and talk about new technologies, we need ways to describe them, compare them, categorize them, and keep in mind their risks and faults. As new waves of sociotechnical change have impacted daily life — electricity, radio, computerization — they have brought with them new terms. So it has been with automated vehicles. As the terminology of automation has entered public and policy discourse, it has reinforced some myths of autonomy. The taxonomy offered by SAE has provided a much-needed language by which to describe and compare the automation of driving. The six-rung ladder ( Figure 1 ), ranging from Level Zero (no automation) to Level Five (full, unconditional automation) has enabled engineers to think about the technical differences between systems. However, as this terminology has entered public and policy discourse, it has served to reinforce some myths of autonomy: that automation increases linearly, directly displaces human work, and will continue until automation is total and humans are completely eliminated from the system [4]. The levels of automation have been treated as waymarks along that seemingly self-eviden…