Approaching the social dilemma of autonomous vehicles with a general social welfare function. (September 2021)

• Record Type: Journal Article
• Authors: Ebina, Takeshi; Kinjo, Keita
• Is Part Of: Engineering applications of artificial intelligence. Volume 104(2021)
• Publication Date: 2021-09
• Subjects: Autonomous driving -- Social welfare functions -- Social dilemma; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2021.104390
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: We consider the social dilemma related to the trolley problem in ethics regarding the driving behavior of an autonomous vehicle (AV) when there are passengers in the AV and pedestrians on the street. To investigate the optimal driving behaviors in different concepts of values, we introduce a general social welfare function (SWF) into our model that encompasses the three types of Bentham- ( ρ = 0 ), Nash- ( ρ → 1 ), and Rawls-type ( ρ → ∞ ) SWFs as a special case. Parameter ρ is key and represents the inequality aversion. We first demonstrate that the optimal solution continuously and monotonically moves from ρ = 0 to ρ → ∞ . This result implies that the optimal solutions involving the Bentham- and Rawls-type SWFs are extreme (maximum or minimum value) solutions, whereas solutions involving the Nash-type SWF can be intermediate. Although some empirical studies have stated that it is better for a policymaker or AV manufacturer to employ Benthamism, our results suggest that care should be applied when choosing a driving behavior. Second, if the total utility of the passengers (or pedestrians) at the optimum is relatively large, our results indicate that the AV will likely go straight and hit the pedestrians (or swerve and crash into a wall) as ρ increases (or decreases). Finally, we discuss the applicability of our results in controlling AVs worldwide. We confirm that setting ρ = 0 is preferred in the U.S., Canada, Australia, and some Western European countries, whereas setting ρ > 0 is preferred in some African and Asian countries, such as China, India, and Japan.