CoChat
Sign Up
Cited by 1,863Psychology of Moral and Emotional JudgmentEthics and Social Impacts of AIEthics in Business and Education

The Moral Machine experiment

Open access
Awad, Edmond, Dsouza, Sohan, Kim, Richard, Schulz, Jonathan, Henrich, Joseph, Shariff, Azim, Bonnefon, Jean-François, Rahwan, Iyad·2018·Nature
DOI 10.1038/s41586-018-0637-6
Open this paper in CoChat — free accountOpen PDF

Abstract

With the rapid development of artificial intelligence have come concerns about how machines will make moral decisions, and the major challenge of quantifying societal expectations about the ethical principles that should guide machine behaviour. To address this challenge, we deployed the Moral Machine, an online experimental platform designed to explore the moral dilemmas faced by autonomous vehicles. This platform gathered 40 million decisions in ten languages from millions of people in 233 countries and territories. Here we describe the results of this experiment. First, we summarize global moral preferences. Second, we document individual variations in preferences, based on respondents' demographics. Third, we report cross-cultural ethical variation, and uncover three major clusters of countries. Fourth, we show that these differences correlate with modern institutions and deep cultural traits. We discuss how these preferences can contribute to developing global, socially acceptable principles for machine ethics. All data used in this article are publicly available.

AI snapshot

Written from the paper’s full text.

TL;DR

The Moral Machine used a multilingual online game to gather millions of judgments about unavoidable autonomous-vehicle crash dilemmas, revealing strong global preferences for sparing humans, more lives, and younger lives. It also found demographic variation that did not reverse preferences, three broad cultural clusters, and country-level correlations with cultural and economic indicators.

Key findings

  • The platform collected 39.61 million decisions from 233 countries and territories; the abstract describes roughly 40 million decisions from over 2.3 million people in ten languages.
  • Worldwide, the strongest preferences were for sparing humans over animals, sparing more characters, and sparing younger characters; weaker preferences included sparing women and sparing pedestrians.
  • Among 492,921 respondents who completed the optional demographic survey, individual characteristics such as age, gender, income, education, politics, and religion did not change the direction of any preference, with effects all below 0.1.
  • Country-level clustering identified three groups—Western, Eastern, and Southern—and preferences correlated with individualism/collectivism, GDP per capita, rule of law, economic inequality, and gender gaps in health and survival.

Methods

  • The Moral Machine was an online “serious game” launched in June 2016 and later translated into nine additional languages; users chose between two outcomes in unavoidable accident scenarios.
  • In Judge mode, sessions contained 13 dilemmas generated around nine factors: species, number of lives, age, gender, physical fitness, social status, pedestrian legality, passenger versus pedestrian status, and whether the car swerves.
  • Analyses used conjoint analysis to estimate average marginal component effects, hierarchical clustering of 130 countries with at least 100 respondents, and country-level correlations with socio-economic and cultural indicators.

Limitations

  • The paper states that the sample was self-selected and not guaranteed to be representative, so policymakers should not treat the results as the final word on societal preferences.
  • The paper states that scenarios assumed certainty about character classification and life-and-death outcomes, did not manipulate respondent-character relationships, and captured preferences before autonomous vehicles are deployed.
  • Evident from the text: the findings are based on hypothetical choices from an internet-connected, tech-interested population, and the cross-sectional design cannot establish causal mechanisms behind country-level correlations.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41586-018-0637-6

Cites

33 of 33 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

3 of 21 cited sources checked · 1 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    Our previous work did not find a strong impact of the relation between respondents and characters on moral preferences.

    “Our previous work did not find a strong impact of this variable on moral preferences 12”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [12] Bonnefon, J.-F., Shariff, A. & Rahwan, I. The social dilemma of autonomous vehicles. Science 352, 1573–1576 (2016). source
  • Premisecitation

    These codes seldom recognize that humans experience inner conflict, interpersonal disagreements, and cultural dissimilarities in the moral domain.

    “These codes seldom recognize, though, that humans experience inner conflict, interpersonal disagreements, and cultural dissimilarities in the moral domain 31 , 32 , 33 .”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites [31] Gastil, J., Braman, D., Kahan, D. & Slovic, P. The cultural orientation of mass political opinion. PS Polit. Sci. Polit. 44, 711–714 (2011). source
    • cites [32] Nishi, A., Christakis, N. A. & Rand, D. G. Cooperation, decision time, and culture: online experiments with American and Indian participants. PLoS ONE 12, e0171252 (2017). source

      The Introduction describes conflicted choices between cooperation and defection, and the Discussion reports cross-cultural differences in cooperation between American and Indian participants. The paper does not address interpersonal disagreements or whether the codes in the citing sentence recognize these experiences; its findings concern cooperation in economic games rather than the moral domain generally. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
    • cites [33]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Findingcitation

    The three clusters are broadly consistent with geographical and cultural proximity according to the Inglehart-Welzel Cultural Map 2010-2014.

    “are broadly consistent with both geographical and cultural proximity according to the Inglehart-Welzel Cultural Map 2010-2014 21 .”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [21] Inglehart, R. & Welzel, C. Modernization, Cultural Change, and Democracy: The Human Development Sequence (Cambridge Univ. Press, Cambridge, 2005).
    • Rests on an assumption: The 130 countries with at least 100 respondents are representative of their countries' moral preferences.
  • Findingown data

    Spain appears in the Western cluster, adjacent to Portugal and other European countries, rather than in the Spanish-speaking Latin American sub-cluster.

    “Spain appears in the Western cluster, adjacent to Portugal and other European countries, rather than in the Spanish-speaking Latin American sub-cluster.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The 130 countries with at least 100 respondents are representative of their countries' moral preferences.
  • Findingown data

    Clusters largely differ in the weight they give to some preferences.

    “clusters largely differ in the weight they give to some preferences.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The 130 countries with at least 100 respondents are representative of their countries' moral preferences.
  • May combine two claims. 28
  • The quote only partly states this claim. 1
  • Classified as probably not a substantial claim. 16

  • May combine two claims.

    Interpretationreasoning

    Participants from countries which are poorer and suffer from weaker institutions are more tolerant of pedestrians who cross illegally, presumably because of their experience of lower rule compliance and weaker punishment of rule deviation.

    “participants from countries which are poorer and suffer from weaker institutions are more tolerant of pedestrians who cross illegally, presumably because of their experience of lower rule compliance and weaker punishment of rule deviation 26 .”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [26] Gächter, S. & Schulz, J. F. Intrinsic honesty and the prevalence of rule violations across societies. Nature 531, 496–499 (2016). source

      The paper reports in its main text (labelled Abstract in our copy; page unavailable) that participants from countries with more prevalent rule violations showed less intrinsic honesty, and it suggests that unpunished cheating may make dishonesty seem justifiable. It does not report that participants from poorer countries were more tolerant of pedestrians crossing illegally, or establish weaker punishment as the reason for such tolerance; its experiment measured honesty using a die-rolling task. (LLM, conf=0.83)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Findingown data

    Global moral preferences include strong preferences for sparing more lives and younger individuals, and weak preferences for sparing women and pedestrians.

    “some of them strong (e.g., sparing more lives or younger individuals), and some of them weak (e.g., sparing women, sparing pedestrians)”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The preference to spare younger characters rather than older characters is much less pronounced for countries in the Eastern cluster and much higher for countries in the Southern cluster.

    “the preference to spare younger characters rather than older characters is much less pronounced for countries in the Eastern cluster, and much higher for countries in the Southern cluster.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The 130 countries with at least 100 respondents are representative of their countries' moral preferences.

  • May combine two claims.

    Findingown data

    The Southern cluster consists of Latin American countries of Central and South America, plus some countries characterized in part by French influence.

    “The third cluster (a broadly Southern cluster) consists of the Latin American countries of Central and South America, in addition to some countries that are characterized in part by French influence”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The 130 countries with at least 100 respondents are representative of their countries' moral preferences.

  • May combine two claims.

    Findingown data

    Only the weak preference for sparing pedestrians over passengers and the moderate preference for sparing the lawful over the unlawful appear to be shared to the same extent in all clusters.

    “Only the (weak) preference for sparing pedestrians over passengers and the (moderate) preference for sparing the lawful over the unlawful appear to be shared to the same extent in all clusters.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The 130 countries with at least 100 respondents are representative of their countries' moral preferences.

  • May combine two claims.

    Findingown data

    Both male and female respondents indicated a positive preference for sparing females, but female respondents showed stronger preference.

    “both male and female respondents indicated Δ preference for Sparing Females, but the latter group showed stronger preference”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Male respondents are 0.06 percentage point less inclined to spare females and 0.09 percentage point more inclined to spare humans compared to female respondents.

    “male respondents are 0.06 percentage point less inclined to spare females and 0.09 percentage point more inclined to spare humans compared to female respondents”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The Eastern cluster contains many far eastern countries such as Japan and Taiwan, belonging to the Confucianist cultural group, and Islamic countries such as Indonesia, Pakistan, and Saudi Arabia.

    “The second cluster (which we call the Eastern cluster) contains many far eastern countries such as Japan and Taiwan, belonging to the Confucianist cultural group, and Islamic countries such as Indonesia, Pakistan and Saudi Arabia.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The 130 countries with at least 100 respondents are representative of their countries' moral preferences.

  • May combine two claims.

    Findingown data

    Cross-country differences in moral preferences correlate with modern institutions and deep cultural traits.

    “these differences correlate with modern institutions, but also with deep cultural traits”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The 130 countries with at least 100 respondents are representative of their countries' moral preferences.

  • May combine two claims.

    Findingown data

    The four most spared characters are the baby, the little girl, the little boy, and the pregnant woman.

    “the four most spared characters are the baby, the little girl, the little boy, and the pregnant woman”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    We see weaker but clear effects for sparing women over men, overweight persons over athletes, and executives over homeless persons.

    “or for that matter, women over men, overweight persons over athletes, or executives over homeless persons--all of which we see weaker but clear effects for.”
    Quote in the paperNot a substantial claim

  • May combine two claims. The quote only partly states this claim.

    Findingown data

    The Moral Machine collected about 40 million decisions (39.61 million) in ten languages from over 2.3 million people in 233 countries, dependencies, or territories.

    “This platform enabled us to gather 40 million decisions in ten languages from over 2.3 million people in 233 countries and territories”
    • Also stated in: § Abstract With the rapid development of Artificial Intelligence technology come widespread concerns about how machines will behave in morally charged situations.Addressing these concerns raises the major challenge of quantifying societal expectations about the ethical principles that should guide machine behavior.To address this challenge, we deployed the Moral Machine, an Internet-based experimental platform that is designed to explore the multi-dimensional moral dilemmas faced by autonomous vehicles.This platform enabled us to gather 40 million decisions in ten languages from over 2.3 million people in 233 countries and territories, and thus to assess the paths and obstacles to machine ethics.First, we summarize global moral preferences, some of them strong (e.g., sparing more lives or younger individuals), and some of them weak (e.g., sparing women, sparing pedestrians).Second, we document individual variations of these preferences, based on respondents' demographics (e.g., age, gender, income), and find that demographics do not change the direction of any preference.Third, we report cross-cultural ethical variation, and uncover three major clusters of countries exhibiting substantial differences along key moral preferences about autonomous vehicles.Fourth, we show that these differences correlate with modern institutions, but also with deep cultural traits.We discuss how these three layers of preferences can contribute to developing global, harmonious, and socially acceptable principles for machine ethics.All data used in this article can be accessed and downloaded at https://goo.gl/JXRrBP .We are entering an age in which machines are not only tasked to promote well-being and minimize harm, but also to distribute the well-being they create, and the harm they cannot eliminate.Distributing well-being and harm inevitably creates tradeoffs, whose resolution falls in the moral domain 1 , 2 , 3 .Think of an autonomous vehicle (AV) that is about to crash, and cannot find a trajectory that would save everyone.Should it swerve onto one jaywalking teenager to spare its three elderly passengers?Should it swerve away from a group of children and kill a single adult passenger instead?Even in the more common instances in which harm is not inevitable, but just possible, AVs will need to decide how to divvy up the risk of harm between the different stakeholders on the road.Car manufacturers and policymakers are currently struggling with these moral dilemmas, in large part because they cannot be solved by any simple normative ethical principles like Asimov's laws of robotics 4 . Asimov's laws were not designed to solve the problem of universal machine ethics, and they were not even designed to let machines distribute harm between humans.They were a narrative device whose goal was to generate good stories, by showcasing how challenging it is to create moral machines with a dozen lines of code.And yet, we do not have the luxury to give up on creating moral machines 5 , 6 , 7 , 8 .AVs will cruise our roads soon, necessitating agreement on the principles that should apply when, inevitably, life-threatening dilemmas emerge.The frequency at which these dilemma will emerge is extremely hard to estimate, just as it is extremely hard to estimate the rate at which human drivers find themselves in comparable situations.Human drivers who die in crashes cannot report whether they were faced with a dilemma; and human drivers who survive a crash may not have realized that they were in a dilemma situation.Note though that ethical guidelines for AV choices in dilemma situations do not depend on the frequency of these situations.Whether these cases are rare, very rare, or extremely rare, we need to agree beforehand on how they should be solved. The keyword here is "we".As emphasized by former U.S. president Barack Obama 9 , consensus in this matter is going to be important.Decisions about the ethical principles that will guide AVs cannot be left to solely to either the engineers or the ethicists.For consumers to switch from traditional human-driven cars to AVs, and for the wider public to accept the proliferation of AI-driven vehicles on their roads, both groups will need to understand the origins of the ethical principles programmed into these vehicles 10 .In other words, even if ethicists were to agree on how AVs should solve moral dilemmas, their work would be useless if citizens were to disagree with their solution, and thus opt out of the future that AVs promise in lieu of the status quo.Any attempt to devise AI ethics must be at least cognizant of public morality. Accordingly, we need to gauge social expectations about the way AVs should solve moral dilemmas.This enterprise, however, is not without challenges 11 .The first challenge comes from the high-dimensionality of the problem.In a typical survey, one may test whether people prefer to spare many lives rather than few 9,12 , 13 ; or whether people prefer to spare the young rather than the elderly 14 , 15 ; or whether people prefer to spare pedestrians who cross legally, rather than pedestrians who jaywalk; or yet some other preference, or a simple combination of two or three of these preferences.But combining a dozen of such preferences leads to millions of possible scenarios, requiring a sample size that defies any conventional method of data collection. · plain
      “allowed us to collect 39.61 million decisions in 233 countries, dependencies, or territories”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The strongest global preferences are sparing humans over animals, sparing more lives, and sparing young lives.

    “the strongest preferences are observed for sparing humans over animals, sparing more lives, and sparing young lives.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The Southern cluster has a strong preference for sparing women and a strong preference for sparing fit characters.

    “we observe some striking peculiarities, like the strong preference for sparing women and the strong preference for sparing fit characters in the Southern cluster.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The 130 countries with at least 100 respondents are representative of their countries' moral preferences.

  • May combine two claims.

    Findingown data

    One standard deviation increase in religiosity is associated with 0.09 more inclination to spare humans, 0.08 less to spare the lawful, 0.06 more to spare pedestrians, and 0.06 less to spare the fit.

    “one increase in standard deviation of religiosity of respondent is associated with 0.09 more inclination to spare humans, 0.08 less inclination to spare the lawful, 0.06 more inclination to spare pedestrians, and 0.06 less inclination to spare the fit.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    There is an association between the preference for sparing the lawful and each of rule of law and log of GDP per capita.

    “Association between the preference for sparing the lawful and each of rule of law and log of GDP per capita (inset).”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The country from which users access the website, geo-localized through IP address, is a valid proxy for the country-level moral preferences of that country.

  • May combine two claims.

    Findingown data

    There is an association between individualism and each of the preference for sparing more characters, and the preference for sparing the young.

    “Association between individualism and each of the preference for sparing more characters, and the preference for sparing the young (inset).”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The country from which users access the website, geo-localized through IP address, is a valid proxy for the country-level moral preferences of that country.

  • May combine two claims.

    Findingnone given

    The Moral Machine was deployed and millions of people from around the world participated.

    “The Moral Machine was deployed to initiate such a conversation, and millions of people weighed in from around the world .”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisereasoning

    The frequency of AV dilemma situations is extremely hard to estimate.

    “The frequency at which these dilemma will emerge is extremely hard to estimate”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    German Ethical Rule #9 does not take a clear stance on whether and when AVs should sacrifice the few to spare the many, but leaves this possibility open.

    “German Ethical Rule #9 does not take a clear stance on whether and when AVs should be programmed to sacrifice the few to spare the many, but leaves this possibility open”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [19] Luetge, C. The German Ethics Code for automated and connected driving. Philos. Technol. 30, 547–558 (2017). source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    German Ethical Rule #7 states that in dilemma situations, the protection of human life should enjoy top priority over the protection of other animal life.

    “German Ethical Rule #7 unambiguously states that in dilemma situations, the protection of human life should enjoy top priority over the protection of other animal life.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [19] Luetge, C. The German Ethics Code for automated and connected driving. Philos. Technol. 30, 547–558 (2017). source

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationown data

    These three preferences may be considered essential building blocks for machine ethics, or at least essential topics to be considered by policymakers.

    “Accordingly, these three preferences may be considered essential building blocks for machine ethics, or at least essential topics to be considered by policymakers.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

    Interpretationnone given

    AVs will cruise roads soon, necessitating agreement on principles for life-threatening dilemmas.

    “AVs will cruise our roads soon, necessitating agreement on the principles that should apply when, inevitably, life-threatening dilemmas emerge.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductown data

    The Western cluster's internal structure exhibits notable face validity, with a sub-cluster of Protestant/Scandinavian countries and a sub-cluster of Commonwealth/English-speaking countries.

    “The internal structure within this cluster also exhibits notable face validity, with a sub-cluster containing Protestant / Scandinavian countries, and a sub-cluster containing Commonwealth / English-speaking countries.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The 130 countries with at least 100 respondents are representative of their countries' moral preferences.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Premisecitation

    Consensus on ethical principles for AVs is important.

    “consensus in this matter is going to be important.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [9] Dadich, S. Barack Obama, neural nets, self-driving cars, and the future of the world. Wired https://www.wired.com/2016/10/president-obama-mit-joi-ito-interview/ (2016).

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisereasoning

    The high-dimensionality of AV moral preferences leads to millions of possible scenarios, requiring a sample size that defies conventional data collection.

    “But combining a dozen of such preferences leads to millions of possible scenarios, requiring a sample size that defies any conventional method of data collection.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    German Ethical Rule #9 states that any distinction based on personal features, such as age, should be prohibited.

    “German Ethical Rule #9 also states that any distinction based on personal features, such as age, should be prohibited.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [19] Luetge, C. The German Ethics Code for automated and connected driving. Philos. Technol. 30, 547–558 (2017). source

  • Classified as probably not a substantial claim. May combine two claims.

    Premisereasoning

    Asimov's laws were not designed to solve universal machine ethics or to let machines distribute harm between humans.

    “Asimov's laws were not designed to solve the problem of universal machine ethics, and they were not even designed to let machines distribute harm between humans.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Premisecitation

    Distributing well-being and harm inevitably creates tradeoffs whose resolution falls in the moral domain.

    “Distributing well-being and harm inevitably creates tradeoffs, whose resolution falls in the moral domain 1 , 2 , 3 .”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [1] Greene, J. Moral Tribes: Emotion, Reason and the Gap Between Us and Them (Atlantic Books, London, 2013).

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [2] Tomasello, M. A Natural History of Human Thinking (Harvard Univ. Press, Cambridge, 2014). source
    • cites [3] Cushman, F. & Young, L. The psychology of dilemmas and the philosophy of morality. Ethical Theory Moral Pract. 12, 9–24 (2009). source

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    Ethical guidelines for AV choices in dilemma situations do not depend on the frequency of these situations.

    “Note though that ethical guidelines for AV choices in dilemma situations do not depend on the frequency of these situations.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Even if ethicists agree on AV moral solutions, their work would be useless if citizens disagree and opt out.

    “even if ethicists were to agree on how AVs should solve moral dilemmas, their work would be useless if citizens were to disagree with their solution, and thus opt out of the future that AVs promise in lieu of the status quo.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    Individual variations are theoretically important but not essential information for policymakers.

    “In sum, the individual variations we observe are theoretically important, but not essential information for policymakers.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Attempts at establishing broad ethical codes for intelligent machines, like the Asilomar AI Principles, often recommend that machine ethics should be aligned with human values.

    “Attempts at establishing broad ethical codes for intelligent machines, like the Asilomar AI Principle s 30 , often recommend that machine ethics should be aligned with human values.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [30] Haidt, J. The Righteous Mind: Why Good People Are Divided by Politics and Religion (Knopf Doubleday, New York, 2012).

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The fact that broad regions of the world displayed relative agreement suggests that our journey to consensual machine ethics is not doomed from the start.

    “But the fact that broad regions of the world displayed relative agreement suggests that our journey to consensual machine ethics is not doomed from the start.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductreasoning

    We did not introduce uncertainty about the fates of the characters, and we did not introduce any uncertainty about the classification of these characters.

    “we did not introduce uncertainty about the fates of the characters, and we did not introduce any uncertainty about the classification of these characters.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Our project builds on a long tradition of investigating cultural variations in ethical judgments.

    “our project builds on a long tradition of investigating cultural variations in ethical judgments 29 .”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [29] Future of Life Institute. Asilomar AI Principles https://futureoflife.org/ai-principles/ (2017).

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationnone given

    What the Moral Machine captured is the state of moral preferences before AVs are deployed on the road, and before they start facing moral dilemmas.

    “what the Moral Machine captured is the state of moral preferences before AVs are deployed on the road, and before they start facing moral dilemmas.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductreasoning

    Our sample is self-selected and not guaranteed to be representative or to exactly match the socio-demographics of each country.

    “its downside is that our sample is self-selected, and not guaranteed to exactly match the socio-demographics of each country.”
    • Also stated in: § DISCUSSION · plain
      “But the fact that our samples are not guaranteed to be representative”
    Quote in the paperSubstantial claim
38 of 38

  • The paper relies on this without support.

    The 130 countries with at least 100 respondents are representative of their countries' moral preferences.

    “We selected the 130 countries with at least 100 respondents (N range = [101 -448,125])”

  • The paper relies on this without support.

    No research ever attempted to measure moral preferences using a 9-dimensional experimental design, in more than 200 countries.

    “No research ever attempted to measure moral preferences using a 9-dimensional experimental design, in more than 200 countries.”

  • The authors state this as an assumption.

    In the Moral Machine scenarios, characters were recognized as adults, children, etc. with 100% certainty, and life-and-death outcomes were predicted with 100% certainty.

    “In our scenarios, characters were recognized as adults, children, etc. with 100% certainty, and life-and-death outcomes were predicted with 100% certainty.”
    “These assumptions are technologically unrealistic, but they were necessary to keep the project tractable.”

  • The paper relies on this without support.

    The country from which users access the website, geo-localized through IP address, is a valid proxy for the country-level moral preferences of that country.

    “The country from which users access the website is geo-localized through the IP address of their computer or mobile device.This information is used to compute a vector of moral preferences for each country.”
4 of 4

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Moral Machine results clash with the IEEE code and German rules that prohibit distinctions based on personal features such as age, yet people strongly preferred sparing the young.

    “These results by contrast clash with many rules such as the IEEE code of ethics and some of the German ethical rules that clearly prohibit any distinction based on personal features such as age. However, results of the moral machine show that people strongly preferred sparing the young (such as children).”

    From Public acceptance and perception of autonomous vehicles: a comprehensive review (2021) · cites this paper as [153]

    The Global Preferences section (page unavailable) says German Ethical Rule #9 prohibits distinctions based on personal features such as age and explicitly contrasts that rule with the Moral Machine’s strong preference for sparing children. Our copy does not establish that the IEEE code contains a comparable prohibition, so that part of the attributed claim is unsupported. (LLM, conf=0.89)

    Experimental — accuracy not yet measured

  • At the level of abstraction of AI Ethics principles, meaningful guidance may be impossible.

    “At this level of abstraction meaningful guidance may be impossible.”

    From Principles alone cannot guarantee ethical AI (2019) · cites this paper as [44]

    The introductory discussion says autonomous-vehicle dilemmas cannot be solved by simple normative principles such as Asimov’s laws, supporting the concern that abstract rules may offer inadequate guidance. The paper makes that point about vehicle decisions, not AI Ethics principles generally; its Discussion also proposes preferences as building blocks for machine ethics. No page number is available in our copy. (LLM, conf=0.84)

    Experimental — accuracy not yet measured

  • MIT's Moral Machine platform collected opinions from millions of people in 233 countries, with 40 million decisions in ten languages.

    “The platform gathered the opinion of millions of people from 233 countries with 40 million decisions in ten languages.”

    From Public acceptance and perception of autonomous vehicles: a comprehensive review (2021) · cites this paper as [153]

  • Moral Machine results show broad differences in relative preferences among countries, but nearby countries have closer moral preferences, with three dominant clusters in the West, East, and South.

    “However, results show broad differences in relative preferences among different countries [153]. However, results show countries within close proximity to one another showed closer moral preferences, with three dominant clusters in the West, East, and South [154].”

    From Public acceptance and perception of autonomous vehicles: a comprehensive review (2021) · cites this paper as [153]

  • Moral Machine results show people prefer sparing humans over animals, sparing more lives, and sparing young lives.

    “Generally, results of the moral machine show that people prefer sparing humans over animals, sparing more lives, and sparing young lives [153].”

    From Public acceptance and perception of autonomous vehicles: a comprehensive review (2021) · cites this paper as [153]

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Related

Authors
Awad, E., Dsouza, S., Kim, R., Schulz, J., Henrich, J., Shariff, A., Bonnefon, J.-F., & Rahwan, I.
Affiliations
  • MIT Media Lab (US)
  • Harvard University (US)
  • University of British Columbia (CA)
  • Centre National de la Recherche Scientifique (FR)
Corresponding author
  • Shariff, Azim
  • Bonnefon, Jean-François
  • Rahwan, Iyad
ORCID
Venue
Nature
Publisher
Springer Science and Business Media LLC
Published
2018-10-24
Volume
563
Issue
7729
Pages
59-64
ISSN
  • 0028-0836 (ISSN-L)
  • 1476-4687
DOI
10.1038/s41586-018-0637-6
OpenAlex
W2896252141
Cited by
1,859
References
31
Citations by year
2026: 214 · 2025: 263 · 2024: 222 · 2023: 270 · 2022: 273 · 2021: 263 · 2020: 223 · 2019: 120 · 2018: 10
FWCI
91.60
Type
journal article
Language
en
Open access
Licence
www.springer.com/tdm (tdm)
Locations
  • Nature (publishedVersion)
  • PubMed (publishedVersion)
  • Open Research Exeter (University of Exeter) (acceptedVersion)
  • DSpace@MIT (Massachusetts Institute of Technology) (submittedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • Figshare (submittedVersion)
  • MPG.PuRe (Max Planck Society) (submittedVersion)
Funders
Agence Nationale de la Recherche (ANR-17-EURE-0010, ANR-10-LABX-0029)
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Psychology of Moral and Emotional Judgment
  • Social Sciences › Safety Research › Ethics and Social Impacts of AI
  • Social Sciences › Decision Sciences › Information Systems and Management › Ethics in Business and Education
Keywords
autonomous vehicles, machine ethics, ethical dilemmas, ethical preferences, cross-cultural variation, individual differences, demographic variation, modern institutions, cultural characteristics
MeSH terms
Accidents, Traffic, Pedestrians, Artificial Intelligence, Data Collection, Decision Making, Female, Humans, Male, Morals, Public Opinion, Robotics, Translating, Motor Vehicles, Internet, Internationality, Harm Reduction

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-10.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-10.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review