Large Language Models have attracted global attention due to their capabilities in understanding, knowledge synthesis, and generating contextually relevant responses, mimicking certain aspects of human reasoning. Although LLMs have demonstrated feasibility in performing autonomous driving tasks in simulated and real-world environments, little is known about their safety and ethical decision-making. To address these questions, we propose a novel framework for evaluating and interpreting the ethical decision-making mechanism of LLM-driven autonomous vehicles. Our study investigates the ethical dilemma of prioritizing saving pedestrians or passengers inspired by the Moral Machine Experiment. We used a stated preference survey to include factors of group size, age, gender, fatality risk, and pedestrian behavior to create 13,122 choice scenarios (a full factorial design) to analyze responses from advanced LLMs, including the GPT-4 series models from OpenAI and Mistral-Large from Mistral AI. Our findings reveal significant differences in the decision-making process and preferences for saving road users among these LLMs. Using a binary logit model to interpret GPT-4′s decisions, we found that the estimated number of deaths, age, and gender significantly affect the model's choices. The decision tree method was also applied to analyze LLMs’ decision-making processes, uncovering potential ethical standards and conditions considered by the models. This study provides valuable insights into ethical considerations in AI systems and thus facilitates the responsible development of AI in autonomous vehicles.

With the increasing use of phone-based ride-hailing apps, concerns have arisen regarding road safety and driver distraction. Despite the recognized safety risks of driver distraction, limited research has explored how distractions from various ride-hailing systems affect drivers in the taxi industry. To close this gap, the current research utilized a driving simulator experiment involving 51 male taxi drivers in two road environments (urban street and motorway) and three distracted driving conditions (no distraction, auditory distraction via radio dispatching system, and visual-manual distraction via mobile application). A car-following scenario with sudden brake events was incorporated into the experiments because this is a typical safety–critical situation where attention will determine the outcome. The collected performance indicators include brake reaction time, time headway, and car-following distance. The grouped random parameters Weibull accelerated failure time model was applied to model the duration data under different road conditions. The brake reaction time and time headway are dependent variables, while the car-following distance is a covariate in the models. The results indicate that although taxi drivers show longer brake reaction time when distracted by mobile app and radio system, this does not necessarily equate with greater risk or reduced safety since they compensate for the risk of rear-end crashes by maintaining a longer time headway. In general, taxi drivers’ brake reaction time and time headway are more profoundly affected by mobile apps when distracted in both urban and motorway scenarios. This highlights the elevated risks associated with such technologies. In addition, significant interaction effects revealed the observed heterogeneity, which suggests that drivers’ personal characteristics influence the relationship between distraction type and driving performance. This research provides valuable insights for designing safer ride-hailing operations and systems.

With the increasing use of ride-hailing apps, concerns have arisen regarding road safety and driver distraction. Despite the recognized safety risks of driver distraction, limited research has explored how distractions from various ride-hailing systems affect drivers in the taxi industry. The research utilized a driving simulator experiment involving 51 taxi drivers in two road environments (urban street and motorway) and three distracted driving conditions (no distraction, auditory distraction via radio dispatching system, and visual-manual distraction via mobile application). A car-following scenario with sudden brake events was incorporated into the experiments. The collected performance indicators include brake reaction time (BRT), time headway (THW), and car-following distance (CFD). The random parameters Weibull accelerated failure time (AFT) model was applied to model the duration data under different road conditions. The results indicate that although taxi drivers show longer BRT when distracted by mobile app and radio system, this does not necessarily equate with greater risk or reduced safety performance since they compensate for the risk of rear-end crashes by maintaining a longer THW. In general, taxi drivers' BRT and THW are more profoundly affected by mobile apps than by radio systems when distracted in both urban and motorway scenarios. In addition, significant interaction effects revealed the observed heterogeneity, which suggests that drivers' personal characteristics influence the relationship between distraction type and driving performance. This research provides valuable insights for designing safer ride-hailing systems and implementing effective driver training and management systems for transport operators.

Use of ride-hailing mobile apps has surged and reshaped the taxi industry. These apps allow real-time taxi-customer matching of taxi dispatch system. However, there are also increasing concerns for driver distractions as a result of these ride-hailing systems. This study aims to investigate the effects of distractions by different ride-hailing systems on the driving performance of taxi drivers using the driving simulator experiment. In this investigation, fifty-one male taxi drivers were recruited. During the experiment, the road environment (urban street versus motorway), driving task (free-flow driving versus car-following), and distraction type (no distraction, auditory distraction by radio system, and visual-manual distraction by mobile app) were varied. Repeated measures ANOVA and random parameter generalized linear models were adopted to evaluate the distracted driving performance accounting for correlations among different observations of a same driver. Results indicate that distraction by mobile app impairs driving performance to a larger extent than traditional radio systems, in terms of the lateral control in the free-flow motorway condition and the speed control in the free-flow urban condition. In addition, for car-following task on urban street, compensatory behaviour (speed reduction) is more prevalent when distracted by mobile app while driving, compared to that of radio system. Additionally, no significant difference in subjective workload between distractions by mobile app and radio system were found. Several driver characteristics such as experience, driving records, and perception variables also influence driving performances. The findings are expected to facilitate the development of safer ride-hailing systems, as well as driver training and road safety policy.