Work

AI Ethics: A Study of Large Language Models in Autonomous Driving Scenarios

Year: 2025
Type: dissertation
Abstract: As autonomous vehicles become more prevalent in the transportation sector, integrating advanced artificial intelligence technologies, such as GPT-4 and other large language models (LLMs), into their decision-making processes is becoming a tangible possibility. In the absence of human in- tervention, these models could play a crucial role in autonomous driving, raising fundamental questions about the principles guiding their decisions, particularly regarding passenger and pedes- trian safety. Given public skepticism, ensuring ethical alignment from the outset is essential to promoting trust and promoting the acceptance of these technologies. To investigate the ethical reasoning of LLMs, this research proposes applying moral dilemmas related to autonomous vehicle based on the Moral Machine platform from MIT o different lan- guage models. The models' responses will be analyzed and compared with human decisions to identify potential discrepancies and emerging ethical patterns. Additionally, a systematic prompt engineering approach will be used to examine how different input strategies influence model re- sponses. A subsequent stage will involve hallucination detection, assessing whether and how these inconsistencies affect decision-making and whether current LLMs can grasp the complex- ity and nuances of scenarios involving moral dilemmas similar to those in the Moral Machine Experiment. Imagine a future where autonomous cars not only comply with established regulations but also align with moral principles that reflect societal values. This research seeks to evaluate whether LLMs can contribute to this vision by exploring their capabilities and limitations in making ethical decisions within autonomous vehicles
Author: Amanda de Oliveira Silva
Language: Portuguese
Open Access status: green
OpenAlex ID: W6989154016
Cites: 79
Cited by: 1
Related to: 10