Human language reflects our social values, biases, and moral judgments. Large language models (LLMs) trained on extensive human texts may therefore learn or encode such information, allowing them to generate responses within moral and ethical domains. Investigating whether LLMs exhibit human-like (including potentially biased or skewed) moral judgments is therefore crucial. Recent moral psychology research suggests that humans tend to have stronger negative reactions toward, and attribute more blame to, intelligent autonomous machines than to fellow humans for identical harm. Here we examine whether LLMs (OpenAI’s GPT-3.5 and GPT-4) exhibit a similar bias against machines in the specific domain of driverless cars. We replicate experiments from two previous studies in the USA and China and find that GPT-4 (but not GPT-3.5), similar to human participants reported previously, consistently rates machine drivers as more blameworthy and causally responsible than human drivers for identical traffic harm (Study 1), while also rating machine versus human drivers’ identical actions as more harmful and morally wrong (preregistered Study 2). This asymmetry in moral judgments is replicated across both LLMs and human participants in a new crash scenario that is unlikely to have been included in the LLMs’ training sets (preregistered Study 3). We discuss whether the blame bias against machines might be morally justified, and also propose that its presence in humans and LLMs could be due to different mechanisms.

Anthropomorphic design in intelligent vehicle agents (IVAs) is crucial for driving safety and user experience. Cultural background may shape user preferences, as evidenced by Chinese car manufacturers offering more anthropomorphic IVAs (e.g., physical robots, human-like virtual agents) than their Western counterparts. While prior research has examined cross-cultural differences in visual anthropomorphism, behavioral anthropomorphism remains understudied. Here we evaluated the performance of eight IVAs (five from Chinese brands, three from Western brands) in responding to user requests and their social repair behaviors (e.g., apology and promise) following request failures. Overall, Chinese and Western IVAs did not differ in their corrective responses or likelihood of employing repair behaviors. However, Chinese IVAs were more likely to use combined behaviors rather than single ones and to incorporate intimacy expressions in their repair behaviors. Our findings highlight cultural design nuances in behavioral anthropomorphism, with implications for the culturally adaptive design of IVAs.

Automated agents’ errors will cause various negative influences on humans and their relationships with humans (e.g., reducing user experience). They are increasingly required to have social recovery strategies (e.g., human-like apology and explanation) to mitigate the negative impacts of their errors and maintain resilient human–automation relationships. However, the efficacy of these strategies in human–automation interaction (HAI) largely remains unknown, especially in less controlled environments. Here we conducted a test track experiment and designed a verbal recovery design (consisting of an apology, explanation, and promise) by an automated driving system (ADS) installed in a real automated vehicle after an ADS failure. We utilized a Wizard of Oz design to simulate the ADS’ failure and its verbal recovery attempt (through a voice by a human or Apple Siri). Participants (N = 389) were assigned to four groups: normal (without experiencing the ADS failure), fault (experiencing the ADS failure), Siri-voice-recovery, and human-voice-recovery. The major measures were positive experience and negative experience while riding in the automated vehicle and perceived ADS usability. Overall, we found that the human-voice-recovery can to some degree mitigate the negative impacts of the ADS failure on user experience. The Siri-voice-recovery worked on positive experience but cannot restore it to that in the normal group. It implies that more empirical efforts are needed to examine social recovery strategies in HAI in natural environments, develop strategies specific to HAI, and offer effective guidelines for social recovery design.

Introducing automated vehicles (AVs) on roads may challenge established norms as drivers of human-driven vehicles (HVs) interact with AVs. Our study explored drivers’ decisions in game-theoretical scenarios amid mixed traffic using an online survey study. We manipulated factors including interaction types (HV-HV vs. HV-AV), scenario types (chicken game vs. public goods game), vehicle driving styles (aggressive vs. conservative), and time constraints (high vs. low). The quantitative results showed that human drivers tended to “defect” more, that is, not cooperate, against vehicles with conservative driving styles. The effect of vehicle driving styles was pronounced when interacting with AVs and in chicken game scenarios. Drivers exhibited more “defection” in public goods game scenarios and the effect of scenario types was weakened under high time constraints. Only drivers with moderate driving styles “defected” more in HV-AV interaction. Our qualitative findings provide essential insights into how drivers perceived conditions and formulated strategies for decision-making.

Automation is replacing or working alongside human operators in various settings. However, its widespread implementation poses potential risks for operators, one of which is referred to as “automation complacency” or simply “complacency.” This concept originated in the field of aviation crash investigations and has since been developed and researched in human-automation interaction (HAI). It holds professional operators in safety-critical systems accountable for system failures. The increasing prevalence of vehicle automation has shed more light on this issue, as it has been identified as a probable cause of recent traffic crashes involving partial and conditional automation. Furthermore, it even plays a significant role in judicial decision-making. In this position paper, we delve into seven key questions surrounding this concept, including its conceptualization, operationalization, detectability beforehand, and prevention. We also examine its causal influence on traffic crashes and its potential to be perceived as nothing more than a blame game. When subjecting these questions to critical scrutiny, we argue that while it is an important phenomenon in HAI, there are valid concerns regarding its appropriate usage in crash investigations and liability litigation. Applying it uncritically in these fields may give rise to moral and ethical issues and result in new consumer harm.

Liability attribution for crashes involving automated vehicles (AVs), if applied improperly, is a factor which can potentially hinder acceptance. The present study investigated the impact of liability attribution on intention to buy an AV. A vignette-based survey was implemented with a hypothetical crash similar to the 2018 Uber crash (which was jointly caused by driver distraction and the malfunctioning of the automated system) leading to a pedestrian’s fatality. Respondents (N = 1524) chose their preferred liability attribution, ranging from human driver exclusively liable to AV manufacturer exclusively liable. Respondents were then randomly allocated to different conditions of actual liability attribution by the local authority. These conditions were then combined into, negative misattribution (the authority assigned more liability to the human driver, compared to the respondent), positive misattribution (the authority assigned less liability to the human driver), and no misattribution. Negative misattribution negatively affected intention to buy; however, positive misattribution did not have a significant impact. The results of a multiple-mediator model indicated that negative misattribution affects intention to buy through the mediating effects of trust, negative affect, and crash acceptability. Theoretical and practical implications of our results are discussed.

Failures of the automated driving system (ADS) in automated vehicles (AVs) can damage driver–ADS cooperation (e.g., causing trust damage) and traffic safety. Researchers suggest infusing a human-like ability, active trust repair, into automated systems, to mitigate broken trust and other negative impacts resulting from their failures. Trust repair is regarded as a key ergonomic design in automated systems. Trust repair strategies (e.g., apology) are examined and supported by some evidence in controlled environments, however, rarely subjected to empirical evaluations in more naturalistic environments. To fill this gap, we conducted a test track study, invited participants (N= 257) to experience an ADS failure, and tested the influence of the ADS’ trust repair on trust and other psychological responses. Half of participants (n= 128) received the ADS’ verbal message (consisting of apology, explanation, and promise) by a human voice (n= 63) or by Apple's Siri (n= 65) after its failure. We measured seven psychological responses to AVs and ADS [e.g., trust and behavioral intention (BI)]. We found that both strategies cannot repair damaged trust. The human-voice-repair strategy can to some degree mitigate other detrimental influences (e.g., reductions in BI) resulting from the ADS failure, but this effect is only notable among participants without substantial driving experience. It points to the importance of conducting ecologically valid and field studies for validating human-like trust repair strategies in human–automation interaction and of developing trust repair strategies specific to safety-critical situations.

COVID-19 pandemic may positively impact the transportation sector’s use of automated vehicles (AVs) and hasten AV adoption in areas such as contactless delivery. However, resistance or negative attitudes toward AVs and fear over safety creates unknowns about whether people are willing to adopt driverless AVs if available in similar pandemics. Two vignette-based studies (N = 1087) found that the pandemic can lead Chinese participants to be more willing to use driverless Robotaxis and exhibit more positive responses to Robotaxis than traditional, manned taxis. In Study 1, their preference for a Robotaxi was greater in the pandemic than in the normal condition. They also preferred the Robotaxi over the manned taxi in the pandemic when we manipulated the Robotaxi to be merely as safe as or superior to the manned taxi. Study 2 compared the Robotaxi and manned taxis with equal safety in both conditions. Participants expressed greater negative affect (fear and anxiety) and lower willingness to ride in a Robotaxi in the normal condition. However, in the pandemic condition, they expressed lower negative affect and greater willingness to ride in the Robotaxi. Trust in Robotaxi was lower than in the manned taxi in the normal condition, whereas there was no difference in trust in both taxis in the pandemic. There was no difference in risk acceptance across the two conditions. Our research provides unique and empirical evidence for supporting the adoption of AVs in similar situations and calls on policy support to develop and deploy AVs to meet the consumer need for driverless mobility services in these situations.

Automated vehicles (AVs) are ready to share public roads with humans in mixed traffic. How human drivers will interact with AVs is unclear. We assumed that AVs’ aberrant behaviors (e.g., driving slowly) provoke higher driving anger than identical ones by human drivers do and examined this assumption from the perspective of mind perception (i.e., perception of whether AVs have human-like minds or lack thereof). Mind perception has two dimensions: agency (capable of doing and planning) and experience (capable of feeling and sensing). Our survey (N = 622) supported this assumption. Agency attribution to AVs is negatively correlated with driving anger, suggesting that perceiving AVs as lacking agency might be associated with greater anger toward AVs’ aberrant behaviors. Attributing high agency to AVs eliminated the human-AV difference. The relationship between driving anger and experience attribution was U-shaped: Participants attributing “middle” experience to AVs reported lower anger than those attributing no or “high” experience to AVs. We speculate that perceiving AV as having high capability to feel and sense might induce the “uncanny valley” effect documented in the literature on human-robot interaction, enhancing more negative reactions. Our findings suggest AVs’ aberrant behaviors and their induced anger as emerging risks to future mixed-traffic flow.

This research involves a controversial topic in the public sphere: should automated vehicles (AVs) be programmed with selfish algorithms to protect their passengers at all costs or utilitarian algorithms to minimize social loss in crashes involving moral dilemmas? Among a growing number of studies on what AVs should do in sacrificial dilemmas from the perspective of laypeople, few have considered how laypeople respond to AVs programmed with these crash algorithms. Our survey collected participants’ deontological evaluation (i.e., evaluations of the moral righteousness of the decisions made by these AVs and of adopting these AVs), their perceived benefit and risk of these AVs, and their behavioral intention to use and willingness to pay (WTP) a premium for these AVs. The participants (N = 580) perceived greater benefits from selfish AVs and reported a greater intention to use and higher WTP a premium for selfish AVs than utilitarian AVs. Deontological evaluation and perceived risk were non-significantly different between these AVs. Overall, selfish AVs were more acceptable to our participants. Deontological evaluation, perceived benefit, and perceived risk were predictive of behavioral intention. Additionally, after controlling for them, vehicle type still exerted a direct influence on behavioral intention. Perceived benefit was the dominant predictor of WTP a premium. Remarkably, participants expressed an insufficient intention to adopt both AVs, probably indicating that in regards to AV deployment, non-positive public attitudes toward AVs are more pressing than the challenge of deciding upon their ethical behaviors in rare moral dilemmas.

We addressed a fundamental question in human reliability analysis (HRA), namely: What is the functional form of the combined effect of multiple performance shaping factors (PSFs) on human unreliability? The combined effect of multiple PSFs on human error probability (HEP) is typically treated as the product of the effects of single PSFs on HEP in a multiplicative model or a sum of the effects of single PSFs in an additive model. However, there is a paucity of studies as to which model is more appropriate. In the present study, we collected data on empirical combined effects of multiple PSFs (198 data points for two PSFs, 61 for three PSFs, and three for four PSFs) from the human performance literature and calculated their multiplicative and additive effects. Both the calculated multiplicative and additive effects were positively correlated with the empirical combined effect. However, the median of the multiplicative effect exceeded that of the empirical combined effect while the median of the additive effect did not significantly differ from that of the empirical combined effect. Thus, the multiplicative model yielded conservative estimates while the additive model produced accurate estimates. Thus, the additive form might be more appropriate to model the joint effect of multiple PSFs on human reliability.

The autonomous vehicle (AV) is expected to dramatically increase road safety. Understanding the public’s initial perceptions and acceptance of AV is imperative because these aspects are likely to determine the future evolution of AVs. This study focuses on public perceptions and acceptance of the two highest levels of vehicle automation – highly autonomous vehicle (HAV) and fully autonomous vehicle (FAV). We drew from the conversation on trust and developed a psychological model to explain three acceptance measures, namely, general acceptance, behavioral intention to use, and willingness to pay (WTP). Using a between-subject survey (N = 742), we determined that the respondents held a stronger belief of benefits from FAV than from HAV. Trust in AV retained a direct effect as well as an indirect effect (mainly through perceived benefit) on the three acceptance measures. In comparison with perceived risk, perceived benefit exerted a higher direct effect on AV acceptance and a higher mediating effect on the trust–acceptance relationship. A prediction analysis further demonstrated that the model exhibited acceptable predictive capability for public acceptance. We drew certain implications for increasing AV acceptance.