PLoS ONE plos plosone PLOS ONE 1932-6203 Public Library of Science San Francisco, CA USA 10.1371/journal.pone.0275812 PONE-D-22-20081 Research Article Research and analysis methodsResearch designSurvey researchSurveys Science policyResearch integrityResearch ethics Physical sciencesMathematicsApplied mathematicsAlgorithms Research and analysis methodsSimulation and modelingAlgorithms Engineering and technologyCivil engineeringTransportation infrastructureRoads Engineering and technologyTransportationTransportation infrastructureRoads Biology and life sciencesPsychologyPsychological attitudes Social sciencesPsychologyPsychological attitudes People and placesGeographical locationsAsiaJapan Engineering and technologyElectronics engineeringComputer engineeringMan-computer interfaceVirtual reality Computer and information sciencesComputer architectureUser interfacesVirtual reality Research and analysis methodsSimulation and modeling Personal ethical settings for driverless cars and the utility paradox: An ethical analysis of public attitudes in UK and Japan Personal ethical setting for driverless cars Takaguchi Kazuya Conceptualization Formal analysis Investigation Visualization Writing – review & editing 1 2 Kappes Andreas Data curation Formal analysis Methodology Software 3 Yearsley James M. Formal analysis 3 Sawai Tsutomu Investigation Resources 4 5 https://orcid.org/0000-0003-3958-8633 Wilkinson Dominic J. C. Project administration Supervision Validation Writing – original draft Writing – review & editing 1 6 7 * Savulescu Julian Funding acquisition Supervision Writing – review & editing 1 8 Oxford Uehiro Centre for Practical Ethics, Faculty of Philosophy, University of Oxford, Oxford, United Kingdom The Department of Ethics, Kyoto University, Kyoto, Japan City, University of London, London, United Kingdom Graduate School of Humanities and Social Sciences, Hiroshima University, Hiroshima, Japan Institute for the Advanced Study of Human Biology (ASHBi), Kyoto University, Kyoto, Japan John Radcliffe Hospital, Oxford, United Kingdom Murdoch Children’s Research Institute, Melbourne, Australia Melbourne Law School, Melbourne, Australia Kong Xiaoqiang ‘Jack’ Editor Texas A&M Transportation Institute, UNITED STATES

I have read the journal’s policy and the authors of this manuscript have the following competing interests: Julian Savulescu is a Partner Investigator on an Australian Research Council grant which involves industry partnership from Illumina (project unrelated to this study). He does not personally receive any funds from Illumina. He is an ethics consultant for Avon Cosmetics Ltd (2021-2023) and a Bioethics Committee consultant for Bayer. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

 * E-mail: dominic.wilkinson@philosophy.ox.ac.uk 15 11 2022 2022 17 11 e0275812 18 7 2022 25 9 2022 2022 Takaguchi et al This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Driverless cars are predicted to dramatically reduce collisions and casualties on the roads. However, there has been controversy about how they should be programmed to respond in the event of an unavoidable collision. Should they aim to save the most lives, prioritise the lives of pedestrians, or occupants of the vehicle? Some have argued that driverless cars should all be programmed to minimise total casualties. While this would appear to have wide international public support, previous work has also suggested regional variation and public reluctance to purchase driverless cars with such a mandated ethical setting. The possibility that algorithms designed to minimise collision fatalities would lead to reduced consumer uptake of driverless cars and thereby to higher overall road deaths, represents a potential “utility paradox”. To investigate this paradox further, we examined the views of the general public about driverless cars in two online surveys in the UK and Japan, examining the influence of choice of a “personal ethical setting” as well as of framing on hypothetical purchase decisions. The personal ethical setting would allow respondents to choose between a programme which would save the most lives, save occupants or save pedestrians. We found striking differences between UK and Japanese respondents. While a majority of UK respondents wished to buy driverless cars that prioritise the most lives or their family members’ lives, Japanese survey participants preferred to save pedestrians. We observed reduced willingness to purchase driverless cars with a mandated ethical setting (compared to offering choice) in both countries. It appears that the public values relevant to programming of driverless cars differ between UK and Japan. The highest uptake of driverless cars in both countries can be achieved by providing a personal ethical setting. Since uptake of driverless cars (rather than specific algorithm used) is potentially the biggest factor in reducing in traffic related accidents, providing some choice of ethical settings may be optimal for driverless cars according to a range of plausible ethical theories.

 http://dx.doi.org/10.13039/100010269 Wellcome Trust 203132/Z/16/Z JSPS KAKENHI 21K12908 Sawai Tsutomu This research was funded in whole, or in part, by the Wellcome Trust [203132/Z/16/Z]. For the purpose of open access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission. Julian Savulescu is Visiting Toh Chin Chye Professor in Molecular Biology and Medicine in the Centre for Biomedical Ethics at the National University of Singapore. Tsutomu Sawai is funded by JSPS KAKENHI Grant (21K12908).The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Data Availability All research data (questionnaire, anonymized result, analysis) are available from the OSF database (osf.io/kv6nu). 1. Introduction 1.1 Driverless cars

Advances in motor vehicle design have reduced the devastating harm associated with traffic collisions. For example it was estimated that forward-collision warning and autonomous braking system prevented about 14% of crash fatalities in 2016 in the US [1]. However, as many as 1.35 million deaths and 50 million injuries still occur every year worldwide from road traffic accidents [2]. This is the 8th highest cause of death, and the leading cause of death for young people aged 5–29. Further technological advances, particularly the advent of automated driving technology (driverless cars) could dramatically reduce this. In Germany, the US and UK, between 90–95% of car accidents are estimated to be caused by human error or misconduct (for example, speeding, inattention, failing to give way) [3–5]. Due to the high rates of accidents caused by human error, driverless cars are believed to have a positive impact on road safety. For example, one study estimated up to 73% reduction in pedestrian crashes in Finland [6], while a US study estimated up to 90% reduction [7]. Another survey suggests that if all human-driven cars were replaced by fully automated driverless cars this could in theory prevent 30,000 lives per year in the US [8]. Considering the great reduction in number of injuries and fatalities, this would be a massive benefit for road users [9] and the wider community by saving social expenditure [10]. These advantages have motivated governments worldwide to facilitate the adoption of driverless cars, and car manufacturers and leading tech companies to compete in their development [11].

However, the development of fully automated cars raises a number of ethical questions. One such question is how such cars should be programmed to respond in the event of a collision. Although automation will potentially eradicate human error, accidents will continue to occur (for example, due to environmental factors, technological failure, or following unexpected behaviours by other road users) albeit with reduced frequency. Faced with an imminent, unavoidable collision, human drivers have limited time or ability to respond. In contrast, fully automated vehicles can be pre-programmed to respond in one or more specific ways taking into account information from the environment and context of the accident. For example, a driverless car detecting an imminent crash could seek to protect occupants of the vehicle, it could seek to avoid harming pedestrians (or other innocent “bystanders” such as cyclists), or it could aim to minimise overall casualties (save the most lives).

 1.2 Ethical programming of driverless cars

These different programming alternatives have been debated from the point of view of ethical theory. For example, utilitarian or consequentialist approaches typically support decisions that would minimise overall numbers of deaths or injuries (i.e., save the most lives). Other approaches (often drawing on variations of the philosophical thought experiment “the trolley problem” [12]) have questioned the idea that it would be ethical to deliberately direct a vehicle in a way that would kill pedestrians or other innocent parties. The German Ethics Commission produced a report indicating (rule 9) that driverless cars should not sacrifice pedestrians to save occupants [13]. On the other hand, prospective users of driverless cars may wish to protect themselves and other passengers (particularly if the passengers are friends or family members). There is a question about whether such partiality to occupants of the vehicle could be justified.

In an attempt to help guide the development of driverless cars, some empirical work has examined the views of the general public on how they think such cars should respond to collisions (their “Moral Algorithm Preference”). For example, in the Moral Machine Experiment (MME), online responses were obtained from almost 40 million decisions from participants in more than 200 countries [14]. Respondents judged a series of hypothetical collisions. The strongest preferences were for sparing humans over non-human animals, saving more lives, and sparing the young over the old [14]. Other studies appear to confirm widespread support for saving the most lives in a collision (though often choosing to save pedestrians if there are similar numbers of occupants/pedestrians at risk) (Table 1). A follow-up study to the MME found that about 40% of respondents chose to treat groups of potential accident casualties equally when given that option (rather than saving the greatest number) [15]. Other studies have explored the way that question framing influences responses [16–18]. When instructed to adopt the perspective of a pedestrian, respondents were more likely to support driverless car responses that would endanger occupants rather than themselves [17, 18]. However, overall these studies suggest that the public may find it acceptable for driverless cars to be programmed with algorithms designed to minimise overall casualties, leading some to suggest that cars should be programmed with an algorithm saving the most lives as a mandatory ethical setting taking this into account [9].

10.1371/journal.pone.0275812.t001 Studies investigating views of the public about ethical response to collision scenarios.
Author Type of study Style of study (methodology) Participants Location Dominant preference
Awad et al., 2018 [14] Moral Algorithm Preference Online survey >500,000 (40 million decisions) 233 countries ‘Save the Most’, (but some global variation)
Some preference for pedestrians over occupants
Awad et al., 2020 [21] Moral Algorithm Preference Online survey 585,531 (alternative part of MME) 233 countries ‘Save the Most’, no preference for protecting passengers
Bergmann et al., 2018 [22] Moral Algorithm Preference Virtual reality simulation 189 Germany ‘Save the Most’
Bigman and Gray, 2020 [15] Moral Algorithm Preference Online survey 2352+843+993 US, UK ‘Save the Most’ (but 40% chose to treat equally)
Bonnefon et al., 2016 [19] Moral Algorithm Preference Online survey 1928 (total) US ‘Save the Most’
Purchase Preferences ‘Save the Occupants’
Faulhaber et al., 2019 [23] Moral Algorithm Preference Virtual reality simulation 189 Germany ‘Save the Most’
Frank et al., 2019 [16] Moral Algorithm Preference Online survey 12,000 US, Denmark Save pedestrian if low numbers in car (1, or 2), Otherwise ‘Save the Most’
(some influence of framing/perspective)
Kallionen et al., 2019 [17] Moral Algorithm Preference (virtual reality/animation) Virtual reality simulation + online survey 184 + 368 Germany ‘Save the Most’
(some influence of framing)
Li et al., 2019 [24] Moral Algorithm Preference Virtual reality simulation 60 China ‘Save the Most’
Liu, P and Liu, J, 2021 [20] Moral Algorithm Preference Online survey 580 China No preference for ‘Save the Most’ vs ‘Save the Occupants’
Purchase Preferences Willingness to pay is higher with ‘Save the Occupants’
Mayer et al., 2021 [18] Moral Algorithm Preference Online survey 1380 Germany Save pedestrian if equal numbers, otherwise ‘Save the Most’
Pugnetti and Schläpfer 2018 [25] Moral Algorithm Preference Online survey 107 Swiss ‘Save the Most’. Equal preference for pedestrians/occupants.
Wintersberger et al., 2017 [26] Moral Algorithm Preference (driving simulator) Driving simulator 40 Germany ‘Save the Most’
Survival rate influences their preference. From occupants’ perspective, higher personal survival rates motivate to choose altruistic decisions

* The table indicates preferences for how respondents think driverless cars should respond (or the choice they personally would make) in the event of a collision (Moral Algorithm Preference), and which car they would actually purchase (Purchase Preference).

**We searched The National Centre for Biotechnology Information PubMed in March 2022 for papers using the keywords ("Autonomous" OR "self-driving" or "driverless") AND ("public opinion" OR "preference") AND ethics. Additional papers were identified from reference lists, related papers and the authors’ libraries. Only studies reporting trade-off scenarios/preferences between saving larger/smaller numbers of people, and between saving occupants vs pedestrians are included.

One challenge is that there may be significant variations between communities in the values that they would apply to driverless car collisions. In the MME study, respondents from a “Western” cluster of countries (including North America and many European countries) had a much stronger preference for saving the most lives than respondents from “Eastern” countries (including Japan, Taiwan, China, India, and many Middle Eastern countries) [14]. The Eastern cluster had a stronger preference for saving pedestrians. Thus, the above suggestion may not be appropriate to a country where public moral preferences are different.

 1.3 Utility paradox

A further challenge is how the public’s views about the moral acceptability of driverless car programming would translate into their actual behaviour. Bonnefon et al. investigated not only public moral intuition about different driverless car algorithms, but also US consumers’ willingness to buy those driverless cars (“Purchase Preferences”). Three quarters of their participants indicated that driverless cars should be programmed to minimise the number of victims (preferences were particularly strong where this would save multiple lives such as 10 people) [19]. This tendency remained (though was weaker) when participants were asked to imagine a family member in the vehicle. However, when participants were asked about purchasing driverless cars, a strikingly different response was obtained–with low support for actually buying a utilitarian-programmed vehicle. This hesitation was even greater when participants were asked to imagine that their family members might use their driverless cars. Bonnefon and colleagues concluded that although people share moral intuitions supporting utilitarian algorithms, they do not want to buy such driverless cars because of concerns for their and their family’s safety [19]. Likewise, a study of 580 Chinese participants indicated a preference for purchasing driverless cars that would save occupants over pedestrians, including a willingness to pay more for such a vehicle [20] (Table 1).

This suggests a potential utility paradox for driverless cars: driverless car algorithms that are designed to minimise collision fatalities may lead to reduced consumer uptake, higher use of non-autonomous cars and higher overall road deaths. Purchase choices may be particularly important to consider because of the very large potential difference in risk of fatal accidents with driverless cars (compared to non-autonomous vehicles).

Because of the significant potential benefit to the community by introducing driverless cars, it would be important to assess which approach would lead to the greatest uptake. It may be preferable to either adopt a different mandatory ethical setting (i.e. other than ‘save the most lives’) or to allow prospective consumers to choose which they would prefer (a so-called “personal ethical setting” [9]). Empirical research is needed that would compare the expected uptake rates of different forms of mandatory ethical setting, with that of a personal ethical setting. As purchase choices may be sensitive to the price of the case we also need to assess consumers’ willingness to pay for their preferred algorithm. No previous studies to our knowledge have evaluated how providing a choice of driverless car algorithm would influence expected uptake of driverless cars, nor how this might vary between cultures.

 1.4 This study

We aimed to study how the potential consumers in two countries believe driverless cars should be programmed to respond in collisions (we call this “Moral Algorithm Preference”), as well as their hypothetical willingness to purchase these cars (we call this “Purchase Preference”). We chose to compare the UK with Japan, since previous research suggests different ethical values impacting on driverless car preferences. We compared purchasing preferences for different programming algorithms (whether mandated or provided as an option) and examined how these choices are influenced by framing effects and purchase price. Based on these data of purchasing behaviours, we attempt to seek what algorithm or which choice of algorithm would lead to the highest driverless car uptake if made available.

 2. Methods 2.1 Participants

The UK Survey was conducted online from April 7 to May 15, 2021. UK residents aged over 18 were recruited via Prolific, using convenience sampling [27]. We performed the same survey with Japanese participants from December 6, 2021, to January 5, 2022, using Crowdworks (crowdworks.jp), again using convenience sampling. This project received ethical approval from the University of Oxford (the Society and Humanities Interdivisional Research Ethics Committee, IDREC9).

For the UK survey, using G*Power [28], we calculated that a sample size of 200 participants would have sufficient power (>80%) to detect small to medium differences in price sensitivity between the driverless car models. Crowdworks has an online Japanese worker pool of 4.1 million and has been previously validated for psychology experiments [29]. We sought a larger sample (300) in Japan to take into account the possibility of higher drop-out with the broad online pool of freelance workers. Participants were reimbursed pro-rata at £7.50/hour (¥180 total for survey in Japan). An attention check was included at the beginning of the survey; those who failed were excluded from analysis. In the UK survey, 190 participants took part, of whom 186 passed the attention check and provided full data were included in the analysis. In the Japanese survey, 360 participated (346 valid responses were included in analysis). All materials and the complete data can be found at: https://osf.io/kv6nu/. Regarding demographic information, In the UK sample (N = 186), 27% were male and 64% were female, while 9% were unknown. 47% were in the range of age 20–40 and 42% were aged between 40–49 or over 50, while 11% were unknown. In the Japanese sample (N = 346). 43% were male and 49% were female, while 9% were unknown. 25% were in the range of age 18–40 and 27% were aged between 40–49 or over 50, while 48% were unknown (S1 File). Demographic information was collected from the survey providers, and not linked to individual responses.

 2.2 Purchase preference: Personal ethical setting

The questionnaire was created in English and translated into Japanese (available at: osf.io/kv6nu) (For survey flow see S4 Text in S1 File). At the beginning, we provided background information: participants were informed that driverless cars were estimated to reduce traffic accidents by 90–94% but would need to be programmed in advance how to respond to any collisions. For the purposes of the survey, they were asked to imagine that in ten years’ time, driverless cars would have the same cost and same features as a regular car but would be safer. They were asked to ignore any concerns about data privacy or liability. Their initial perceived likelihood of purchasing a driverless car (on a Likert scale from 1—Not likely to 7—Extremely likely) was assessed as a baseline measure.

Subsequently, participants were given information about programming options for driverless cars’ responses to collisions. Three programming algorithms were described: ‘Save the Pedestrians’ [always save pedestrians in a collision between driverless car and pedestrian]; ‘Save the Occupants’ [always save occupants of the car]; and, ‘Save the Most’ [always save the greater number of people]. We included edited images from the Moral Machine study [14] to help participants understand the different algorithms. The survey then asked them which of these programming models they would prefer if they were to purchase such a car. They were given the three programming algorithms (named, and briefly described), as well as the option of ‘Random choice’, wherein the driverless car would randomly choose to save either occupants or pedestrians.

 2.3 Framing effects and Purchase Preference

Next, they were asked to imagine (and to indicate their purchasing preference), first if they were buying a car in a situation in which they had a young family member who would often be a passenger, and second in a situation in which they had a young family member who would often be a pedestrian. As before, they were given the option of the three driverless car algorithms (‘Save the Pedestrians’, ‘Save the Occupants’, ‘Save the Most’) or ‘Random Choice’.

 2.4 Mandatory ethical setting: Purchase and Moral Algorithm Preference

In the next section, participants were asked to indicate how likely (on a scale from 1 (Not very likely) to 7 (extremely likely)), they were to buy a driverless car if all such cars were programmed with the same algorithm (i.e., mandatory ethical setting). For example, they were told that all cars were programmed to ‘Save the Most’. This question was repeated to assess likelihood of purchase for each of the three algorithms (‘Save the Pedestrians’,’ Save the Occupants’, ‘Save the Most’).

They were lastly asked which programming model they believed should be adopted if all driverless cars were going to be programmed in the same way (their Moral Algorithm Preference), (they were given the three algorithms as well as the option of ‘Random Choice’). They were reminded that they might be either occupants or pedestrians.

 2.5 Price sensitivity

Finally, a discrete choice experiment was conducted. Participants were presented with a series of choices between two driverless cars that feature different programming models but have different price tags and asked which car they would buy. It has been estimated that most new cars in the UK sell from about £12,000 to £23,000 [30]. We used that price range, starting with £15,000. There were 36 combinations for the price tests due to the three different models and three different prices (£15,000, £19,000 and £23,000; ¥2,250,000, ¥2,850,000, and ¥3,450,000). Fig 1 shows an example question:

10.1371/journal.pone.0275812.g001 An example question for the discrete choice experiment evaluating price sensitivity and preference for driverless car algorithms. 2.6 Model description—willingness to pay for the purchase

In our model, the utility difference between Car A and Car B can be described as the difference in price and the difference in safety programming: ΔU(A,B)=(PriceB−PriceA)+(ProgA−ProgB) Where PriceA is the price of Car A (e.g., £15000) and ProgA is the utility associated with the programming of Car A (e.g., Utilitarian).

There are three programming options: ‘Save the Occupants’, ‘Save the Pedestrians’, ‘Save the Most’ (utilitarian), resulting in two independent parameters. We set the utility of the ‘Save the Most’ option to zero, making it the reference point for the other two programming options. And we expressed the preferences for each of these programming options (‘Save the Occupants’ = Occ; ‘Save the Pedestrians’ = Ped) in units of thousands of pounds.

How did we turn a utility difference ΔU(A, B) into a choice probability? Roughly, if the utility difference is large and positive, participants ought to choose A with probability ~1. And if the utility difference is large and negative, participants ought to choose B with probability ~1. To model the choice probability, we scaled the utility by multiplying by a parameter τ, and then applied an inverse probit transformation. This allowed us to go from the utility difference (Eq 1) to the probability of choosing A given options A and B, pAB(A)=probit−1(τΔU(A,B))=∫−∞ΔU(A,B)τ2πeτx22dx The parameter τ represents how sensitive participants were to changes in utility, so how quickly choices switched between definitely A and definitely B as the utility varied. Lower values mean participants spent more time in the ‘unsure’ region.

At the individual level the choices are binary, so our choice probability that came from the model resulted in just a single yes/no response. We can model this by assuming the data comes from a single trial where the choice probability is given by pAB(A), in other words, p(Data|ΔU)∼Bernouli(pAB(A))

To model this in a hierarchical way means a) every individual has their own values for the parameters Ped, Occ, τ, b) These individual values can be assumed to be drawn from some population level distributions. Each of these population level distributions has an associated mean, HPed, HOcc, Hτ, and variance. More precisely, Ped(i)∼ϕ(HPed,Ht),Occ(i)∼ϕ(HOcc,Ht),τ(i)∼ϕ(Hτ,Ht2) Where ϕ(μ, τ) is the pdf of the normal distribution with mean μ and variance 1/τ. The idea here is that we allow for individual variation in utilities and sensitivity, but we constrain these differences so that information about the preferences of one participant is still weakly informative about the preferences of the others. HPed, HOcc, Hτ, Ht, and Ht2 are population level parameters (hyperparameters), which govern the distribution of utilities and sensitivity in the population. These are the things to be determined by the model fitting, and we assign them the following priors, HPed∼ϕ(0,110),HOcc∼ϕ(0,110),Hτ∼γ(12,12),Ht∼γ(12,12),Ht2∼γ(12,12). where γ(a, b) is the pdf of the gamma distribution with shape parameters, a, and b.

 2.7 Model implementation

The model was fit to the data via Bayesian methods using JAGS (Plummer, 2003 [31]), using a form of Markov Chain Monte Carlo (MCMC) [32]. Fits used three MCMC chains and 50000 MCMC samples, with a burn in of 5000 samples. Chain convergence was assessed using the R^ statistic, and all chains had good convergence by this metric. We report means and HDIs for the posteriors of the hyperparameters, and distributions of the Ped and Occ parameters across participants and trials.

 3. Results

Statistical analysis was conducted using SPSS, Version 28, and computational model analysis were performed with Matlab.

3.1 Algorithm and Purchase Preference–which algorithm should cars have?

The largest group of participants indicated the same response when asked which algorithm they believed should be programmed into all driverless cars (Moral Algorithm Preference), and which algorithm they would prefer to purchase if able to choose (purchasing preference; personal ethical setting) in both countries (Fig 2). In the UK, the largest group selected ‘Save the Most’ (55.3% answered ‘Save the Most’ should be programmed and 45.7% answered it was their preferred algorithm), while in Japan, the largest group selected ‘‘Save the Pedestrians’ (54.6% and 56.9% respectively). A minority of participants wished to purchase the other models (UK: ‘Save the Occupants’ 23.4%, ‘Save the Pedestrians’, 23.4%; Japan: ‘Save the Occupants’ 15.6%, ‘Save the Most’ 24.6%).

10.1371/journal.pone.0275812.g002 <p>Participant preference among UK and Japanese participants for driverless car algorithms when asked: a. which algorithm should be programmed (if all cars programmed identically) b. which they would personally prefer to purchase if able to choose c. Their purchase preference if they imagined having young family who would often be passengers d. Their purchase preference if they imagined having young family who would often be pedestrians.</p> </caption> <graphic mimetype="image" position="float" xlink:href="info:doi/10.1371/journal.pone.0275812.g002" xlink:type="simple"/> </fig> <p>While Moral Algorithm Preferences aligned with purchasing preference in each country, there was a statistically significant difference between Purchase Preferences and Moral Algorithm Preference: Moral Algorithm Preference for the UK (McNemar-Bowker (6 N = 186) = 17.60, p = .007) and for Japan (McNemar-Bowker (6) = 23.655, p = .001, N = 346).</p> </sec> <sec id="sec016"> <title>3.2 The effect of framing on Purchase Preference

Next, we examined the effects of different question framing. When asked to imagine that one’s family would often be in the car, a higher proportion of respondents in both countries selected to purchase ‘Save the Occupants’ compared to when the question was presented without such a frame (UK: 57%, McNemar-Bowker (6, N = 186) = 68.415, p < .0001; Japan: 40.2%, McNemar-Bowker (6, N = 346) = 103.814, p < .001). Similarly, there was a significant increase in preference for ‘Save the Pedestrians’ when asked to imagine one’s family often being pedestrians compared to when the question was presented without such a frame (UK: 50%, McNemar-Bowker (6, N = 186) = 50.509, p < .0001; Japan: 74.3% McNemar-Bowker (6, N = 346) = 63.800, p < .001). Responses to both framings were different between the UK and Japan (p < .001).

 3.3 The influence of mandatory ethical settings on willingness to purchase

As a baseline question, we asked participants how likely it is that they would purchase a driverless car in ten years’ time (if their current car needed replacing) without specifying any safety programming. In the UK, the median response on the 1 (not at all likely) to extremely likely (7) scale was 5, with a Mean of 5.13 (SE = .119). In Japan, the median response was 6, with a Mean of 5.66 (SE = .068). Overall, 74% of the UK participants, and 82% of the Japanese indicated that they were likely (response >4) to buy a driverless car (neither likely nor unlikely: UK 10%, Japan 9%). Japanese participants had a higher willingness to purchase a driverless car than UK participants, t(530) = 4.166, p < .001.

Next, we examined how consumers’ willingness to purchase driverless cars changes when a single algorithm is programmed for all driverless cars. For all three mandated ethical settings, there was reduced willingness to purchase compared to the baseline question indicating that some participants would only purchase if their preferred programming algorithm were available (when ‘Save the Most’ algorithm is only available—UK participants (M = 4.14, SE = .128, t(185) = 7.918, p < .001) and Japanese participants (M = 4.17, SE = .07, t(345) = 16.69, p < .001); when ‘Save the Pedestrians’ algorithm is only available—UK participants (M = 3.58, SE = .1249, t(185) = 10.925, p < .001) and Japanese participants (M = 4.77, SE = .07, t(345) = 10.72, p < .001); when ‘Save the Occupants’ algorithm is only available—UK (M = 4.134, SE = .123, t(185) = 7.59, p < .001) and Japanese participants (M = 4.12, SE = .093, t(345) = 17.505, p < .001)).

If a mandatory ethical setting were adopted, 47% of UK respondents would buy a car programmed to save occupants (a drop from 74% baseline) (Fig 3). 60% of Japanese respondents would purchase a car programmed to save pedestrians (a drop from 82% baseline non-specified). Overall, participants’ willingness of the purchase was affected by the type of algorithm in both countries.

10.1371/journal.pone.0275812.g003 The overall likelihood of purchasing a driverless car and the likelihood of purchasing a driverless car if a mandatory ethical setting were used and particular algorithms were the only option available.

By tracking each participant’s purchasing preference on the three algorithms, a Venn diagram was created to illustrate the distribution of consumers’ preferences for those three algorithms (Fig 4). A small proportion of participants were likely to buy a driverless car whichever algorithm was available (UK 11.8%, Japan 13.6%), and a minority were not willing to purchase regardless of algorithm (UK 23.1%, Japan 21.1%). In the UK, two thirds of respondents (67.8%) were willing to purchase either ‘Save the Occupants’ or ‘Save the Most’ models. In Japan, 72.3% were willing to purchase either ‘Save Pedestrians’ or ‘Save the Occupants’.

10.1371/journal.pone.0275812.g004 Allocation of consumers’ preferences for three different algorithms.

Respondents who were “unlikely” or “neither likely/unlikely” to purchase any model are indicated outside the Venn diagram (the figures in parentheses refer to the actual number of participants).

 3.4 Willingness to pay: Purchase Preferences

The model was able to predict well which decisions UK and Japanese participants would make on each of the 36 scenarios (see S1 File). Examining the estimates for the population level parameters for the preference strength, we observed that for the UK population, going from a utilitarian program to one favouring pedestrians was equivalent to a price increase of £10,700, while going from a utilitarian option to one favouring occupants was equivalent to a price increase of around £3,170 (Fig 5). A different picture emerged for the Japanese population. Here, the model estimated that going from a utilitarian program to one favouring pedestrians was equivalent to a price decrease of £4,180, while going from a utilitarian option to one favouring occupants was equivalent to a price increase of around £6,470.

10.1371/journal.pone.0275812.g005 Average preference strength of ‘Save the Pedestrians’ and ‘Save the Occupants’ over ‘Save the Most’.

(A positive price changes means that participants preferred ‘Save the Most’, and would require a price discount to choose the alternative).

Comparing the population estimates (S2 Table in S1 File), the largest difference between the UK and the Japanese population was the utility assigned to ‘Save the Pedestrians’ programming. For the UK sample this was assigned much lower utility than the ‘Save the Most’ or ‘Save the Occupants’ options. For the Japanese sample in contrast, ‘Save the Pedestrians’ was significantly more attractive than either ‘Save the Occupants’ or ‘Save the Most’ options. There was also a significant difference in the utility assigned to the ‘Save the Occupants’ programming, with the Japanese sample rating this as significantly worse than the UK sample. Finally, we also found that Japanese participants showed a greater consistency in their utility judgments (t parameter: Japan .0045 versus UK .0034) significantly higher, and a less noisy decision-making process (Tau parameter: Japan: .388 versus UK.291), indicating that Japanese preferences were more stable than the UK ones.

In each country, large individual differences existed in the UK and the Japanese Sample (S2 Fig in S1 File). When we examined the distribution of preferences values within our UK sample for ‘Save the Pedestrians’ over the ‘Save the Most’, we can see that a minority of 25 participants preferred a ‘Save the Pedestrian’ programming to a ‘Save the Most’ programming. Twenty-five participants would have paid up to £40,000 pounds more to get a car with ‘Save the Most’, and avoid a car with ‘Save the Pedestrian’ programming (S2 Fig in S1 File). In the Japanese sample, 149 participants preferred the ‘Save the Most’ over the ‘Save the Pedestrian’ safety programming. Here, 42 participants would have paid up to £10,000 to get a ‘Save the Pedestrian’ over a ‘Save the Most’ programming, and 10 participants up to £20,000 more. The most popular safety programming in each country was not endorse by everyone, and mandatory settings might deter large minorities from purchasing driverless cars. Even more pronounced is this effect when we examine the distribution of preference values in both samples for ‘Save the Occupants’ over ‘Save the Most’. Here, we can see that in the UK sample, 65 participants preferred such programming over ‘Save the Most’, and in the Japanese sample, 105 participants preferred ‘Save the Occupants’ over ‘Save the Most’.

 4. Discussion

In this international online survey, we found striking differences between potential consumers in the UK and Japan in their preferences for the programming of driverless cars and their hypothetical purchase. A majority of UK participants (55.3%) supported the programming of driverless cars to ‘Save the Most’ lives in a collision. In contrast, a majority of Japanese respondents (54.6%) supported algorithms that would prioritise the saving of pedestrians. Purchasing preferences in both countries were highly sensitive to framing with shifts (in favour of saving family members) when asked to imagine that their family members would often be either occupants or pedestrians. Importantly, in both countries, willingness to purchase was lower when ethical settings were mandated to a single algorithm, compared to baseline willingness to purchase and compared to the proportion willing to purchase when given options of driverless car algorithms. Respondents from the UK were divided (a similar proportion being willing to buy a car programmed to ‘Save the Most’ and to ‘Save the Occupants’), while a majority in Japan were willing to purchase a driverless car that would prioritise pedestrians. In both countries, price sensitivity tracked general preferences, and participants generally placed a significant price premium on their personally preferred algorithm.

4.1 Cultural variation in values applied to driverless cars

The overall Moral Algorithm Preference identified in our study is consistent with previous studies. Surveys (largely in European and North American populations) have generally supported driverless algorithms that would save the most lives (Table 1), and this was described as a globally shared preference in the MME [14]. We observed a similar pattern in the responses from our UK respondents. However, within the MME, there was a significant difference between respondents from a ‘Western’ cluster of countries and those from an ‘Eastern’ cluster [14]. Our finding that Japanese respondents had a much stronger preference for prioritising pedestrians is consistent with that prior global survey. It may reflect communitarian values in Japan and strong senses of social responsibility and conformity [33]. It is also possible that the Japanese preference reflects recent media attention and social concern about accidents involving elderly drivers and pedestrians [34], or lower levels of vehicle ownership in Japan. Further studies are needed to know the reasons for Japanese purchasing behaviours.

 4.2 Algorithm versus Purchase Preferences

We found a difference between Moral Algorithm Preferences (which algorithm they think should be programmed) and purchasing preferences. However, such purchasing preferences (in the absence of family framing) generally tracked participants’ views about what should be programmed overall (Fig 2). This is somewhat in contrast to Bonnefon et al., who found a marked fall in support for a utilitarian algorithm when participants were asking about purchase [19]. However, we found that Purchase Preferences were highly sensitive to question framing. Past studies asking about general Moral Algorithm Preference have found some effect of framing perspective on responses (i.e., asked to imagine or respond as pedestrian vs occupant of car), but generally preserved utilitarian responses. We found much stronger preferences when participants were asked to imagine family members involved in accidents. It may be that our questions about Purchase Preferences are more individually directed, and therefore more sensitive to personal circumstances. Alternatively, it may be that the wording of questions in our survey primed respondents to alter their responses (possibly in a perceived socially desirable direction).

 4.3 Mandatory ethical setting

We were interested to explore the impact of a mandatory ethical setting on overall willingness to purchase driverless cars. In Bonnefon et al.’s study, US consumers indicated aversion to governmental regulation on what algorithm to be programmed into driverless cars, which was pointed out as a concern for mandatory ethical setting. Although we did not ask the UK and Japanese consumers’ attitude towards governmental regulation, our data suggest that mandatory ethical settings discourage consumer uptake since for many consumers’, their decisions whether or not to buy driverless cars were contingent on which accident algorithm is available. For example, from our UK data, a mandatory ethical setting with any type of algorithm reduced by 20% the number of consumers willing to buy driverless cars. It appeared that the largest number of respondents would buy a driverless car if given the option of choosing between the three different algorithms, or what we call a personal ethical setting. Very few respondents indicated a preference for abdicating responsibility by having the car randomly choose to save either pedestrians or occupants.

 4.4 The utility paradox and the ethics of personal ethical settings

Overall, our results appear to support what we have labelled the ‘utility paradox’ in driverless car algorithms. In our sample, across two countries, algorithms designed to ‘Save the Most’ lives would actually lead to lower uptake of driverless cars and thus save fewer lives in practice, given the ability of driverless cars to reduce trade-off accidents in total. In Japan, that was because most respondents preferred driverless cars to prioritise pedestrians. In the UK, respondents were divided as to which algorithm they would prefer. Any mandated ethical setting attracted a smaller proportion of hypothetical consumers than those who indicated (without an algorithm specified) they were willing to purchase a driverless car.

This finding suggests that making driverless cars available with a personal ethical setting may yield a higher uptake and consequently have the greatest overall impact on road casualties, at least in the first instance. This is a libertarian option [35], but would also be predicted to have the greatest utility (and therefore be supported by utilitarianism). As some of the authors have argued elsewhere, from behind a veil of ignorance, it is rational to select the policy which saves the most lives (including those of the individual and family members) [36]. Thus, a broad contractualist approach would also support making driverless cars available with a personal ethical setting when they are introduced into general circulation. From this point of view, a personal ethical setting has the advantage that while it would be supported by utilitarians and contractualists by maximising driverless car uptake, it would also be aligned with liberalism by providing individual value-based choices.

This argument in favour of a personal ethical setting is dependent on purchasing choices, and our survey suggests that such choices may be susceptible to context and framing effects. It may be more of an advantage for some countries than others (in our Japanese participants, a ‘Save the Pedestrians’ mandatory setting appeared to have high support). One way for governments to increase uptake may be to subsidise the cost of driverless cars in general, or of particular algorithms (for example, in a way similar to subsidies for electric cars). For example, our willingness to pay analysis suggests that in Japan, a price subsidy of £4000 might be required to shift preferences to a ‘Save the Most’ algorithm. Another possibility (which would merit further study) may be to adapt algorithms to include a combination of values (for example, prioritising car occupants if passengers in the car, and similar numbers to pedestrians, while otherwise aiming to ‘Save the Most’). Yet another option would be to introduce driverless cars with a personal ethical setting in the initial stage, then to shift to a (e.g. ‘Save the Most’) mandated ethical setting once non-automated cars have been largely or entirely replaced.

There are other ethical considerations for a personal ethical setting. One concern may be that use of such an option would lead owners of cars to be morally responsible (and potentially liable for damages) in the event of collision fatalities, perhaps analogously to driving beyond the speed limit [37]. It may be that insurance costs would be higher for those who do not choose to purchase ‘Save the Most’ algorithms [37]. Allowing individuals to choose their ethical setting might be socially divisive or lead to stigma. On the other hand, a personal ethical setting allows consumers to exercise their freedom and autonomy. Given a range of different plausible ethical responses (particularly where there is not a large difference in the number of people under threat), a libertarian response [38] might allow individuals to incorporate their personal values (within limits) into their vehicle programming. It is potentially important to allow individuals to take moral responsibility for their actions when using artificial intelligence. The Japanese Cabinet Secretariat accepted this concept and pointed out that “when using AI, people must judge and decide for themselves how to use it” [39]. That would allow individuals to choose to sacrifice their own (and their passengers) to save the greatest number of lives in a collision. Individual autonomy arguably should be respected and is especially crucial in a situation where the person’s life is at stake.

One objection is that autonomy can be curtailed in the public interest or for the sake of public health. Yet our research indicates a happy convergence of personal autonomy and public interest in the UK and Japan. Based on plausible assumptions about the high-safety level of driverless car system, the rarity of trade-off accidents and contingent on purchasing behaviors, the most lives would be saved if people were allowed to express their autonomy (unless governments banned non-autonomous vehicles and mandated an algorithm of ‘Save the Most Lives’).

 4.5 Limitations and future directions

There are some limitations to the generalisability of these results. Online surveys have inherent selection biases. Although we used participants pools that have previously been validated in behavioural research, those who completed our surveys may not be representative of the wider population. (On the other hand, a younger, internet-using audience may overlap with the market for driverless cars). While the sample size was similar to many previous studies in this area (Table 1), a larger sample would have allowed more confident estimates of population attitudes. Only those who passed an attention check were included in the analysis, but, some of those included in the analysis may not have given careful thought to each question. This might be a problem especially in the discrete choice experiment.

As we found in our survey, respondents may be susceptible to the framing of the questions, and the specific wording of our study may have influenced results. (However, the fact that the global preferences among UK and Japanese respondents were consistent with those seen in the MME [14] and other related studies [40], supports the validity of responses). We were interested in the impact of driverless car algorithms on purchase preferences. However, stated intentions may not represent actual purchase choices. Participants may change their mind in the future or fail to imagine what they will actually do in the real situation.

 5. Conclusions

Fully automated driverless cars are not yet commercially available. But the results of this study provide some insights that may be important for their regulation and development. Public views about the safety and ethics of driverless cars will be crucial for their acceptance. It may be that how such cars are programmed to respond may need to differ between countries–depending on the prevailing values of the community. It may also be that permitting some ethical choice in driverless car programming would be valuable.

The programming of driverless cars is fundamentally an ethical decision. We must take responsibility for the goals of such programming, whether it is saving the greatest number, or considering other factors such as age, responsibility, etc [40]. Some kind of democratic process preferably informed by ethical procedures [41] is necessary to arrive at how lives should be valued. Regulation and mandating of programming and purchase of driverless vehicles to achieve these ethical goals (once they have been derived) may be justifiable. However, in the absence of mandated replacement of non-automated vehicles, it is essential to explore the psychosocial effects of policy on purchasing behaviour because the most important factor affecting the well-being and autonomy of road users and those affected by road use is likely to be uptake of driverless cars. If we were to minimise casualties overall, it might not be wise to adopt a ‘Save the Most’ algorithm as a mandated ethical setting because of the potential to decrease driverless car uptake (the utility paradox). We have argued that allowing freedom of choice of ethical setting in driverless cars would potentially be supported by utilitarian, contractualist and libertarian ethical theories. It may be the most ethical policy.

 Supporting information It consists of all the supporting files for this manuscript.

(DOCX)

We would like to show our gratitude to Professor Tony Hope (an emeritus fellow of St Cross college at University of Oxford) for sharing his wisdom with us. Our discussion about connecting philosophical debates and empirical research was enriched by his critical insights.

 References Benson AJ, Tefft BC, Svancara AM, Horrey WJ. Potential Reductions in Crashes, Injuries, and Deaths from Large-Scale Deployment of Advanced Driver Assistance Systems. Res Brief. 2018 [cited 29 Aug 2022]. Available: https://trid.trb.org/view/1566022 WHO. Global status report on road safety 2018. 2018 [cited 27 Apr 2022]. Available: https://www.who.int/publications-detail-redirect/9789241565684 Verkehrsunfälle—Fachserie 8 Reihe 7–2017. In: Statistisches Bundesamt [Internet]. [cited 27 Apr 2022]. Available: https://www.destatis.de/DE/Themen/Gesellschaft-Umwelt/Verkehrsunfaelle/Publikationen/Downloads-Verkehrsunfaelle/verkehrsunfaelle-jahr-2080700177004.html Department for Transport. Contributory Factors for Reported Road Accidents (RAS50). 2013. Available: https://www.gov.uk/government/statistical-data-sets/ras50-contributory-factors Singh S. Critical Reasons for Crashes Investigated in the National Motor Vehicle Crash Causation Survey (Traffic Safety Facts Crash⋅Stats. Report No. DOT HS 812 115). 2015. Available: https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/812115.pdf Utriainen R. The potential impacts of automated vehicles on pedestrian safety in a four-season country. J Intell Transp Syst. 2021;25: 188–196. doi: 10.1080/15472450.2020.1845671 Combs TS, Sandt LS, Clamann MP, McDonald NC. Automated Vehicles and Pedestrian Safety: Exploring the Promise and Limits of Pedestrian Detection. Am J Prev Med. 2019;56: 1–7. doi: 10.1016/j.amepre.2018.06.024 30337236 LaFrance A. Self-Driving Cars Could Save 300,000 Lives Per Decade in America. In: The Atlantic [Internet]. 29 Sep 2015 [cited 27 Apr 2022]. Available: https://www.theatlantic.com/technology/archive/2015/09/self-driving-cars-could-save-300000-lives-per-decade-in-america/407956/ Gogoll J, Müller JF. Autonomous Cars: In Favor of a Mandatory Ethics Setting. Sci Eng Ethics. 2017;23: 681–700. doi: 10.1007/s11948-016-9806-x 27417644 Mathilde C. Average cost of road casualties and accidents by severity in Great Britain in 2020. 2020. Available: https://www.statista.com/statistics/322862/average-cost-of-road-accidents-and-casualties-in-great-britain-uk/ Healey J, Mays K. Which cars park themselves best? Challenge results. In: USA Today [Internet]. 2012 [cited 27 Apr 2022]. Available: https://www.usatoday.com/story/money/cars/2012/12/06/self-parking-cars-challenge/1743199/ Foot F. The Problem of Abortion and the Doctrine of the Double Effect in Virtues and Vices. Oxf Rev 5. 1967; 5–15. BMVI Ethical Commission. Automated and Connected Driving. 2017. Available: https://www.bmvi.de/SharedDocs/EN/publications/report-ethics-commission-automated-and-connected-driving.pdf?__blob=publicationFile Awad E, Dsouza S, Kim R, et al. The Moral Machine experiment. Nature. 2018 Oct;563: 59–64. Bigman YE, Gray K. Life and death decisions of autonomous vehicles. Nature. 2020;579: E1–E2. doi: 10.1038/s41586-020-1987-4 32132695 Frank D-A, Chrysochou P, Mitkidis P, Ariely D. Human decision-making biases in the moral dilemmas of autonomous vehicles. Sci Rep. 2019;9: 13080. doi: 10.1038/s41598-019-49411-7 31511560 Kallioinen N, Pershina M, Zeiser J, Nosrat Nezami F, Pipa G, Stephan A, et al. Moral Judgements on the Actions of Self-Driving Cars and Human Drivers in Dilemma Situations From Different Perspectives. Front Psychol. 2019;10: 2415. doi: 10.3389/fpsyg.2019.02415 31749736 Mayer MM, Bell R, Buchner A. Self-protective and self-sacrificing preferences of pedestrians and passengers in moral dilemmas involving autonomous vehicles. PLOS ONE. 2021;16: e0261673. doi: 10.1371/journal.pone.0261673 34941936 Bonnefon J-F, Shariff A, Rahwan I. The Social Dilemma of Autonomous Vehicles. Science. 2016;352. doi: 10.1126/science.aaf2654 27339987 Liu P, Liu J. Selfish or Utilitarian Automated Vehicles? Deontological Evaluation and Public Acceptance. Int J Human–Computer Interact. 2021;37: 1231–1242. doi: 10.1080/10447318.2021.1876357 Awad E, Dsouza S, Kim R, Schulz J, Henrich J, Shariff A, et al. Reply to: Life and death decisions of autonomous vehicles. Nature. 2020;579: E3–E5. doi: 10.1038/s41586-020-1988-3 32132696 Bergmann LT, Schlicht L, Meixner C, König P, Pipa G, Boshammer S, et al. Autonomous Vehicles Require Socio-Political Acceptance—An Empirical and Philosophical Perspective on the Problem of Moral Decision Making. Front Behav Neurosci. 2018;12: 31. doi: 10.3389/fnbeh.2018.00031 29541023 Faulhaber AK, Dittmer A, Blind F, Wächter MA, Timm S, Sütfeld LR, et al. Human Decisions in Moral Dilemmas are Largely Described by Utilitarianism: Virtual Car Driving Study Provides Guidelines for Autonomous Driving Vehicles. Sci Eng Ethics. 2019;25: 399–418. doi: 10.1007/s11948-018-0020-x 29357047 Li S, Zhang J, Li P, Wang Y, Wang Q. Influencing Factors of Driving Decision-Making Under the Moral Dilemma. IEEE Access. 2019;7: 104132–104142. doi: 10.1109/ACCESS.2019.2932043 Pugnetti C, Schläpfer R. Customer Preferences and Implicit Tradeoffs in Accident Scenarios for Self-Driving Vehicle Algorithms. Risk Financ Manag. 11(2): 2018. Wintersberger P, Prison A-K, Riener A, Hasirlioglu S. The experience of ethics: Evaluation of self harm risks in automated vehicles. 2017 IEEE Intelligent Vehicles Symposium (IV). 2017. pp. 385–391. doi: 10.1109/IVS.2017.7995749 Chandler J, Shapiro D. Conducting Clinical Research Using Crowdsourced Convenience Samples. Annu Rev Clin Psychol. 2016;12: 53–81. doi: 10.1146/annurev-clinpsy-021815-093623 26772208 Faul F, Erdfelder E, Buchner A, Lang A-G. Statistical power analyses using G*Power 3.1: Tests for correlation and regression analyses. Behav Res Methods. 2009;41: 1149–1160. doi: 10.3758/BRM.41.4.1149 19897823 Majima Y. The Feasibility of a Japanese Crowdsourcing Service for Experimental Research in Psychology. SAGE Open. 2017;7: 2158244017698731. doi: 10.1177/2158244017698731 Yurday E. Average Cost of Cars UK 2021. 2021 [cited 3 Sep 2022]. Available: https://www.nimblefins.co.uk/cheap-car-insurance/average-cost-cars-uk Plummer, M. (2003). JAGS: A Program for Analysis of Bayesian Graphical Models Using Gibbs Sampling. Proceedings of the 3rd International Workshop on Distributed Statistical Computing (DSC 2003), Vienna, 20–22 March 2003, 1–10. van Ravenzwaaij D, Cassey P, Brown SD. A simple introduction to Markov Chain Monte-Carlo sampling. Psychon Bull Rev. 2018;25: 143–154. doi: 10.3758/s13423-016-1015-8 26968853 Inoue T. The Poverty of Rights-Blind Communality: Looking Through the Window of Japan. BYU Law Rev. 1993;1993: 517–551. Shimada T. Elderly driver in fatal Ikebukuro accident gets 5 years in prison | The Asahi Shimbun: Breaking News, Japan News and Analysis. In: The Asahi Shimbun [Internet]. [cited 28 Apr 2022]. Available: https://www.asahi.com/ajw/articles/14431946 Fournier T. Will My Next Car Be a Libertarian or a Utilitarian?: Who Will Decide? IEEE Technol Soc Mag. 2016;35: 40–45. doi: 10.1109/MTS.2016.2554441 Savulescu J, Vergano M, Craxì L, Wilkinson D. An ethical algorithm for rationing life-sustaining treatment during the COVID-19 pandemic. BJA Br J Anaesth. 2020;125: 253–258. doi: 10.1016/j.bja.2020.05.028 32560912 Contissa G, Lagioia F, Sartor G. The Ethical Knob: ethically-customisable automated vehicles and the law. Artif Intell Law. 2017;25: 365–378. doi: 10.1007/s10506-017-9211-z Christman J. Autonomy in Moral and Political Philosophy. Fall 2020. In: Zalta EN, editor. The Stanford Encyclopedia of Philosophy. Fall 2020. Metaphysics Research Lab, Stanford University; 2020. Available: https://plato.stanford.edu/archives/fall2020/entries/autonomy-moral/ Cabinet Secretariat. Social Principles of Human-Centric AI. 2020. Arora C, Savulescu J, Maslen H, Wilkinson D. The Intensive Care Lifeboat: a survey of lay attitudes to rationing dilemmas in neonatal intensive care. BMC Med Ethics. 2016;17. doi: 10.1186/s12910-016-0152-y 27821118 Savulescu J, Gyngell C, Kahane G. Collective Reflective Equilibrium in Practice (CREP) and controversial novel technologies. Bioethics. 2021;35: 652–663. doi: 10.1111/bioe.12869 33945162 10.1371/journal.pone.0275812.r001 Decision Letter 0 Kong Xiaoqiang ‘Jack’ Academic Editor 2022 Xiaoqiang ‘Jack’ Kong This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Submission Version 0

23 Aug 2022

PONE-D-22-20081Personal ethical settings for driverless cars and the utility paradox: an ethical analysis of public attitudes in UK and JapanPLOS ONE

Dear Dr. Wilkinson,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Oct 07 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Xiaoqiang ‘Jack’ Kong

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at 

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and 

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. We note that the grant information you provided in the ‘Funding Information’ and ‘Financial Disclosure’ sections do not match. 

When you resubmit, please ensure that you provide the correct grant numbers for the awards you received for your study in the ‘Funding Information’ section.

3. Thank you for stating the following in the Competing Interests section: 

"I have read the journal's policy and the authors of this manuscript have the following competing interests:Julian Savulescu is a Partner Investigator on an Australian Research Council grant LP190100841 which involves industry partnership from Illumina. He does not personally receive any funds from Illumina. He is an ethics consultant for Avon Cosmetics Ltd (2021-2023) and a Bioethics Committee consultant for Bayer"

Please confirm that this does not alter your adherence to all PLOS ONE policies on sharing data and materials, by including the following statement: "This does not alter our adherence to  PLOS ONE policies on sharing data and materials.” (as detailed online in our guide for authors http://journals.plos.org/plosone/s/competing-interests).  If there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared. 

Please include your updated Competing Interests statement in your cover letter; we will change the online submission form on your behalf.

4. Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: N/A

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: This study conducted two online surveys in UK and Japan to investigate the potential “utility paradox” of fully automated vehicles (AVs). The authors examined the purchase preference under different programming settings, including mandated and optional. Besides, the framing effect and the price sensitivity were also analyzed and discussed. The results of this paper may help the government or automobile manufacturers better understand the public’s attitude to the ethical issues of Avs. This paper is clear, interesting, and meaningful. However, the methods could be written in more detail. There are some suggestions are listed in the following:

1. In the introduction, the authors said that the 25% reduction in incapacitating injuries in the US is due to advanced motor vehicle design. However, the reduction is also related to other factors, such as advanced traffic signal control devices, better street and highway design, or education. More directed data sources may better support the authors’ ideas that AVs could reduce traffic crashes or collisions. Papers about CAV safety and simulation have proved the conclusion.

2. The personal information of the participants, such as race, gender, or age, is not discussed in the paper. If the survey has such information, it should be presented in the paper to help readers better understand the data and prerequisites for results.

3. What kinds of sampling method is applied in the paper, random or stratified sampling? Please illustrate in section 2.1.

4. In section 2.1, the authors estimated that a sample size of 200 people would have sufficient power to detect differences. The authors should illustrate the reasons why the authors make such assumptions. If some references could support your assumption, please cite them.

5. In section 2.5, the authors mention three different prices, but the reason why these three different prices are selected is not well-discussed. The determination of price setting and difference should be present in the paper.

6. In transportation engineering, the most common model for mode choice is the logit model. Please better explain why a normal distribution with a mean of 0 and variance of 1/τ is selected instead of the logit model.

7. In section 2.6, please explain the specific meaning of each population-level parameter, and how to tune the hyperparameters that should be presented in the paper.

8. In section 2.7, what is the meaning of chains and sample here? I think it should explain in more detail. Otherwise, the readers may confuse it with the survey sample (UK: 186, Japan: 346)

9. In section 4.1, the authors mentioned a significant difference between respondents from different countries. The finding of Japanese respondents is well-presented, but the finding of UK respondents is not found.

Overall, this paper is of good quality, I recommended this paper with minor revision.

Reviewer #2: Comment 1：

Page 5, line 4-5.

The introduction could be updated with newer references. What have been the injuries or fatalities trends in recent years?

Comment 2：

Table 1. The table should be structured clearer. Things are not described in the text should not be retained in the table, like “virtual reality” in column Type of study, what does that mean? I can only guess that is the methodology setting used in the study.

Comment 3:

Page 19, line 1-2.

Is there any supplementary material that can support the usage of normal distribution? From my point of view, logistic regression is more used for binary selection and the transformation from utility to probability. You may need to provide or cite some materials to support your model selection.

Comment 4:

Page 19, Equation (4).

What is HDri? Is it a typo?

Comment 5:

Page 21, line 15.

The “same pattern” makes me confused. Do the responses have the same pattern in both country? Since the comparison after this sentence is between UK and Japan. But in UK, “save the most” was the mostly selected, and “save the pedestrians” is the most selected in Japan, which should not be defined as the same. Or it means the two preferences have the same response pattern in each country? You should clarify this.

Comment 6:

Fig 2.

The quality of figures downloaded from the submission is poor and hard to see the words in the figure. The resolution of figures should meet the requirements of the publication, which is at a 300-600 ppi resolution, and not exceeding 10MB in size. Check all your figures other than just Fig 2.

Comment 7:

Page 27, section 4.1.

Is it possible that the participants in Japan survey are mostly pedestrians without a car? Did you consider the vehicle ownership in the survey?

Comment 8:

The survey structure in the methodology, the data summary and model results proposed by the authors are relatively difficult to follow. Flowcharts or tables can be helpful to follow the process.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

**********

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

 10.1371/journal.pone.0275812.r002 Author response to Decision Letter 0 Submission Version 1

9 Sep 2022

(See also attached file)

Response to reviewers’ comments

Key: Highlighted text – changed or added text in the revised manuscript

Page/line numbers – reflect the page and line numbers in the revised manuscript

Editors

E1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

RESPONSE: Thank you ¬– we have checked styles and file naming

E2. We note that the grant information you provided in the ‘Funding Information’ and ‘Financial Disclosure’ sections do not match.

When you resubmit, please ensure that you provide the correct grant numbers for the awards you received for your study in the ‘Funding Information’ section.

RESPONSE: Thank you ¬– we have revised the Financial Disclosure statement (previously it included a grant number that is not related to this project)

E3. Thank you for stating the following in the Competing Interests section:

"I have read the journal's policy and the authors of this manuscript have the following competing interests:Julian Savulescu is a Partner Investigator on an Australian Research Council grant LP190100841 which involves industry partnership from Illumina. He does not personally receive any funds from Illumina. He is an ethics consultant for Avon Cosmetics Ltd (2021-2023) and a Bioethics Committee consultant for Bayer"

Please confirm that this does not alter your adherence to all PLOS ONE policies on sharing data and materials, by including the following statement: "This does not alter our adherence to PLOS ONE policies on sharing data and materials.” (as detailed online in our guide for authors http://journals.plos.org/plosone/s/competing-interests). If there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared.

Please include your updated Competing Interests statement in your cover letter; we will change the online submission form on your behalf.

RESPONSE: We have updated a Competing Interests statement in our cover letter, as below:

““I have read the journal's policy and the authors of this manuscript have the following competing interests: Julian Savulescu is a Partner Investigator on an Australian Research Council grant which involves industry partnership from Illumina (project unrelated to this study). He does not personally receive any funds from Illumina. He is an ethics consultant for Avon Cosmetics Ltd (2021-2023) and a Bioethics Committee consultant for Bayer. This does not alter our adherence to PLOS ONE policies on sharing data and materials."

E4. Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

RESPONSE: Thank you ¬¬– we have checked the references

Reviewer 1

Reviewer #1: This study conducted two online surveys in UK and Japan to investigate the potential “utility paradox” of fully automated vehicles (AVs). The authors examined the purchase preference under different programming settings, including mandated and optional. Besides, the framing effect and the price sensitivity were also analyzed and discussed. The results of this paper may help the government or automobile manufacturers better understand the public’s attitude to the ethical issues of Avs. This paper is clear, interesting, and meaningful. However, the methods could be written in more detail. There are some suggestions are listed in the following:

In the introduction, the authors said that the 25% reduction in incapacitating injuries in the US is due to advanced motor vehicle design. However, the reduction is also related to other factors, such as advanced traffic signal control devices, better street and highway design, or education. More directed data sources may better support the authors’ ideas that AVs could reduce traffic crashes or collisions. Papers about CAV safety and simulation have proved the conclusion.

RESPONSE: Thank you for this suggestion.

We added direct recent data corroborating how advances in motor vehicle design prevented crash fatalities.

“Advances in motor vehicle design have reduced the devastating harm associated with traffic collisions. For example it was estimated that forward-collision waning and autonomous breaking system prevented about 14% of crash fatalities in 2016 in the US (1) .”

“Due to the high rates of accidents caused by human error, driverless cars are believed to have a positive impact on road safety. For example, one study estimated up to 73% reduction in pedestrian crashes in Finland (6), while a US study estimated up to 90% reduction (7). Another survey suggests that if all human-driven cars were replaced by fully automated driverless cars this could in theory prevent 30,000 lives per year in the US (8).”

The personal information of the participants, such as race, gender, or age, is not discussed in the paper. If the survey has such information, it should be presented in the paper to help readers better understand the data and prerequisites for results.

RESPONSE: Thank for this suggestion. We have added demographic data for our sample in the supporting information.

What kinds of sampling method is applied in the paper, random or stratified sampling? Please illustrate in section 2.1.

RESPONSE: Recruitment from crowed sourced services such as Mturk and Prolific is best described as convenience sampling. We added this to Section 2.1.

“The UK Survey was conducted online from April 7 to May 15, 2021. UK residents aged over 18 were recruited via Prolific, using convenience sampling. We performed the same survey with Japanese participants from December 6, 2021, to January 5, 2022, using Crowdworks (crowdworks.jp), again using convenience sampling.”

1.4. In section 2.1, the authors estimated that a sample size of 200 people would have sufficient power to detect differences. The authors should illustrate the reasons why the authors make such assumptions. If some references could support your assumption, please cite them.

RESPONSE: Thanks for bringing this to our attention. We now write:

For the UK survey, using G*Power(27), we calculated that a sample size of 200 participants would have sufficient power (>80%) to detect small to medium differences in price sensitivity between the driverless car models.

1.5. In section 2.5, the authors mention three different prices, but the reason why these three different prices are selected is not well-discussed. The determination of price setting and difference should be present in the paper.

RESPONSE: We chose the numbers based on the average car prices in the UK (ref). It has been estimated that most new cars sell from about £12,000 to £23,000. We used that prize range, starting with £15,000. We added this to section 2.5.

“It has been estimated that most new cars in the UK sell from about £12,000 to £23,000 [29]. We used that price range, starting with £15,000. “

1.6. In transportation engineering, the most common model for mode choice is the logit model. Please better explain why a normal distribution with a mean of 0 and variance of 1/τ is selected instead of the logit model.

RESPONSE: We thank the reviewer for bringing this to our attention. Essentially what we are describing is a probit function linking the (suitably scaled) difference in utility to the response probability. A probit function was favoured over a logit simply because it is easier to implement in a Bayesian modelling framework. We do not expect that the choice of probit vs logit would have an influence on our findings. We have added a note in the text to make this link clear.

“To model the choice probability, we scaled the utility by multiplying by a parameter √τ, and then applied an inverse probit transformation. This allowed us to go from the utility difference (Equation 1) to the probability of choosing A given options A and B

1.7. In section 2.6, please explain the specific meaning of each population-level parameter, and how to tune the hyperparameters that should be presented in the paper.

RESPONSE: We thank the reviewer for noting this. Each individual has their own value of the utility they assign to, eg, the Pedestrian favouring algorithm. These individual utilities are assumed to be drawn from some distribution at the population level with mean H_Ped and variance (1/H_t).

So, H_Ped and H_Occ are the means of the distributions of the utilities of the Pedestrian and Occupant favouring algorithms across the population. H_t controls the heterogeneity of the utilities across the population, so larger values imply a greater degree of agreement between members of the population about the value of these utilities.

Every participant also has a parameter tau that controls their sensitivity to differences in utility between different alternatives. Larger values of tau mean a greater sensitivity. These are assumed to be drawn from some distribution at the population level with mean H_tau and variance (1/H_t2). As above, H_t2 controls the heterogeneity of these sensitivities across the population.

In terms of ‘tuning’ these parameters, the model was fit to the data using Markov Chain Monte Carlo methods, as briefly discussed in Section 2.7. The Supporting Information reports means and 95% HDIs of the posterior estimates for each of these parameters.

“The model was fit to the data via Bayesian methods using JAGS (Plummer et al., 2003), using a form of Markov Chain Monte Carlo (MCMC) [30]. Fits used three MCMC chains and 50000 MCMC samples, with a burn in of 5000 samples.”

1.8. In section 2.7, what is the meaning of chains and sample here? I think it should explain in more detail. Otherwise, the readers may confuse it with the survey sample (UK: 186, Japan: 346)

RESPONSE: These terms describe aspects of the Markov Chain Monte Carlo process that was used to fit the model. ‘Chain’ refers to the results of single run of the algorithm, multiple runs are usually performed to assess convergence. ‘Samples’ refers to the number of steps in the algorithm that have been performed. MCMC algorithms are guaranteed to converge only asymptotically, so the number of samples is typically very large. For more information about MCMC sampling we have provided a reference explaining this technique (see also response 1.7).

1.9. In section 4.1, the authors mentioned a significant difference between respondents from different countries. The finding of Japanese respondents is well-presented, but the finding of UK respondents is not found.

RESPONSE: Thank for highlighting this omission. We have added the following to this paragraph.

“Surveys (largely in European and North American populations) have generally supported driverless algorithms that would save the most lives (Table 1), and this was described as a globally shared preference in the MME(12). We observed a similar pattern in the responses from our UK respondents.”

Overall, this paper is of good quality, I recommended this paper with minor revision.

Reviewer 2

2.1:

Page 5, line 4-5

The introduction could be updated with newer references. What have been the injuries or fatalities trends in recent years?

RESPONSE: Thank you for this suggestion. We have updated the references in this section. See response 1.1

2.2：

Table 1. The table should be structured clearer. Things are not described in the text should not be retained in the table, like “virtual reality” in column Type of study, what does that mean? I can only guess that is the methodology setting used in the study.

RESPONSE: Thank you for this suggestion. We have added a column to the table indicating the methodology of the studies cited. We have also made some revisions for clarification.

2.3:

Page 19, line 1-2.

Is there any supplementary material that can support the usage of normal distribution? From my point of view, logistic regression is more used for binary selection and the transformation from utility to probability. You may need to provide or cite some materials to support your model selection.

RESPONSE: Thank you for this suggestion. We have added details to explain the methodology used. See response to Reviewer 1, point 1.6.

2.4:

Page 19, Equation (4).

What is HDri? Is it a typo?

RESPONSE: Thank you for pointing this out. We have corrected this.

2.5:

Page 21, line 15.

The “same pattern” makes me confused. Do the responses have the same pattern in both country? Since the comparison after this sentence is between UK and Japan. But in UK, “save the most” was the mostly selected, and “save the pedestrians” is the most selected in Japan, which should not be defined as the same. Or it means the two preferences have the same response pattern in each country? You should clarify this.

RESPONSE: Apologies for this lack of clarity. We meant to indicate that UK and Japan had the same pattern in the sense that the algorithm they think should be programmed corresponded with their preferred algorithm to purchase. We have edited this part to hopefully clarify the point.

“While Moral Algorithm Preferences aligned with purchasing preference in each country, there was a statistically significant difference between purchase preferences and Moral Algorithm Preference:”

2.6:

Fig 2.

The quality of figures downloaded from the submission is poor and hard to see the words in the figure. The resolution of figures should meet the requirements of the publication, which is at a 300-600 ppi resolution, and not exceeding 10MB in size. Check all your figures other than just Fig 2.

RESPONSE: Thank you – we have uploaded higher resolution versions of the figures

Comment 7:

Page 27, section 4.1.

Is it possible that the participants in Japan survey are mostly pedestrians without a car? Did you consider the vehicle ownership in the survey?

RESPONSE: Thank you for this excellent question. We did not ask if the participants own a car or not. As you point out, Japan's tendency for wishing to spare pedestrians might relate to lower vehicle ownership (approximately 69% compared to 77% in UK). We have added this possibility in the discussion

“It is also possible that the Japanese preference reflects recent media attention and social concern about accidents involving elderly drivers and pedestrians (27), or lower levels of vehicle ownership in Japan. Further studies are needed to know the reasons for Japanese purchasing behaviours.”

Comment 8:

The survey structure in the methodology, the data summary and model results proposed by the authors are relatively difficult to follow. Flowcharts or tables can be helpful to follow the process.

RESPONSE: Thank you for this suggestion. We have added a flow chart of the survey in the supporting information

Submitted filename: Response to Reviewers.docx

 10.1371/journal.pone.0275812.r003 Decision Letter 1 Kong Xiaoqiang ‘Jack’ Academic Editor 2022 Xiaoqiang ‘Jack’ Kong This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Submission Version 1

26 Sep 2022

Personal ethical settings for driverless cars and the utility paradox: an ethical analysis of public attitudes in UK and Japan

PONE-D-22-20081R1

Dear Dr. Dominic Wilkinson,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Both reviewers has provied few additional comments. Please reflect these comments on your latest version of manuscript! 

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Xiaoqiang ‘Jack’ Kong

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

There are a few comments provided by reviewers! Please address them before you submit the final version.

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: No

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The authors have properly addressed the concern, but there is some feedback in the following.

1. The authors provided the demographic information of the data. It seems there are slight differences between the UK and Japan in terms of gender and age distribution. Due to the convenience sampling, such a difference is tolerable.

2. The choice of logit and probit would nearly not influence the findings, but the former assumes the random variable obeys a logistic distribution, while the latter assumes that the random variable obeys normal distribution.

Overall, I have no further comments for this paper.

Reviewer #2: Comment #1

In the revised manuscript, there are some typos on page 5 between line 4 and 5.

The "forward-collision waning" should be "forward-collision warning" and the "autonomous breaking system" should be "autonomous braking system".

Comment #2

Page 19, line 1 and line 18.

The meaning of "Occ" and "Ped" in both two lines is not clear. Italics are mostly used to indicate variables, and the use of different font styles conveys different meanings, which the author needs to clarify.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

**********

 10.1371/journal.pone.0275812.r004 Acceptance letter Kong Xiaoqiang ‘Jack’ Academic Editor 2022 Xiaoqiang ‘Jack’ Kong This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

20 Oct 2022

PONE-D-22-20081R1

Personal ethical settings for driverless cars and the utility paradox: an ethical analysis of public attitudes in UK and Japan

Dear Dr. Wilkinson:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Xiaoqiang ‘Jack’ Kong

Academic Editor

PLOS ONE