What does your car think about utilitarianism?
Reflecting upon “Is Your Autonomous Vehicle as Smart as You Expected? Suggesting a better route to evaluate autonomous vehicles’ smartness.” by Yu-Ju (Tony) Tu, Shari S. Shang, and Junyi Wu in the Association for Computing Machinery’s 2022 publication.
Check out the full research paper here and be sure to Tweet your takeaways at me, @UrszulaSolarz.
As I enter the CS industry, I’ve been looking for ways to learn more about ongoing software research. Therefore, I was super excited to discover the Communications of the ACM magazine and even more excited when I found an article about autonomous vehicle software in it. The article combined a lot of my interests at the intersection of humanities and STEM, specifically using computer science to solve problems in urban studies and physics. Autonomous vehicles (AVs) signify the future of transportation (both personal, public, and industrial), so urban studies answer which socio-technical networks AVs will change and how, while physics gives us the tools to know what goes on under the hood.
Before reading the article, I knew some basics about AVs. In a class on The Technological World, we covered how motors change electrical energy into mechanical energy. This is key in understanding how autonomous electric vehicles work. Furthermore, in high school, I did some projects on LiDAR, AKA Light Detection and Ranging. Some producers of AV use this technology to locate objects in the path of AVs and adjust their course if necessary. However, LiDAR has applications beyond that and cars are by far not the first transportation technologies undergoing automation. For instance, autopilot is a long-recognized tool on board planes, as are autonomous metro systems in Dubai and Miami. Going back to the automobile industry, the current point of AVs is not to completely replace human drivers, but rather supplement them. In fact, the main motivation for further AV development is not human laziness, but the appeal of precise mathematics guiding motion, preventing accidents, and navigating traffic. To this end, both governments and private companies have to be part of policy deliberations regarding the use of and safety regulations regarding AVs because they are used in public and private spaces.
The article presents a three-tiered framework for rating how smart AV features are based on their intelligence quotient (IQ), ethical quotient (EQ), and adversity quotient (AQ). This framework combines both technical and ethical measures to help consumers determine the smartest, and thus safest, AV to purchase. This is especially useful in a world where “Mobility-as-aService” (MaaS) is rapidly expanding and advertised information may not be veritable.
The article begins by arguing that the most straightforward way to decide how smart an AV is relies on comparing the kinds of driving decisions it makes relative to the ones a human would have done. It then explains how IQ rates an AV’s ability to “see and understand.” Seeing is based upon detection via sensor and processors, while understanding regards identification, risk analysis, and reaction. Then, EQ breaks down how an AV makes ethical decisions regarding “pedestrians’ or other vehicle occupants’ lives and deaths.” This may involve the development of software for the “utilitarianism principle, the legitimacy principle, and the social responsibility principle.” For instance, Mercedes-Benz’s AVs “always prioritize their own passengers regardless of any road situation” which differs from the economic utility functions other AVs use to “establish a priority in deciding what can be sacrificed.” Finally, AQ decides how well an AV reacts when information and communications technology (ICT) infrastructure is limited, like with a weak GPS signal in a tunnel shaky surface conditions affecting LiDAR accuracy.
The article has a useful chart summarizing these principles as:
In the future, I would like to learn more about the intersection of cybersecurity and driving. Currently, it seems the main issues drivers face regarding security deals with carjackings, stolen car keys, and GPS manipulation. With the expansion of AVs, the amount of software vulnerabilities undoubtedly will rise.
