Theorizing Moral Cognition: Culture in Action,Situations,and Relationships

Abstract

Dual-process theories of morality are approaches to moral cognition that stress the varying significance of emotion and deliberation in shaping judgments of action. Sociological research that builds on these ideas considers how cross-cultural variation alters judgments, with important consequences for what is and is not considered moral behavior. Yet lacking from these approaches is the notion that, depending on the situation and relationship, the same behavior by the same person can be considered more or less moral. The author reviews recent trends in sociological theorizing about morality and calls attention to the neglect of situational variations and social perceptions as mediating influences on judgment. She then analyzes the moral machine experiment to demonstrate how situations and relationships inform moral cognition. Finally, the author suggests that we can extend contemporary trends in the sociology of morality by connecting culture in thinking about action to culture in thinking about people.

Preface

This paper is about moral judgments in challenging and uncertain times and how social relationships shape our perceptions of right and wrong. My original goal in writing it was to explore a curious gap in the psychological research on morality that we often unwittingly incorporate into our sociological work: how situations and perceptions influence moral cognition and, in turn, social behavior. I never could have imagined how COVID-19 would collapse the distance between theory and practice.

Today, as some 85 percent of Americans live under orders to shelter-in-place, where even the most banal social interaction could prove a link in a chain of contagion and death, these issues feel relevant, urgent, and close. In particular, I begin the paper with a summary of two models of moral cognition: the social intuitionist model and the digital camera model. The latter is premised on what is occasionally cheekily referred to as “Trolleyology,” after a famous series of experiments where participants are confronted with life-and-death decisions. It has often been dismissed by those who claim such circumstances—where one person is tasked with choosing who lives and who dies—are esoteric and unrelatable. Still, as someone who studies decision-making about violence, it seemed worth sorting through the truths inherent in the scenario. I am devastated by how relevant those truths seem today.

As I write, there are places in the world where there aren’t enough hospital beds or respirators to allow all patients to receive adequate medical care. Italian doctors have been forced to weigh one life against another. Very soon, their colleagues in the United States will face such terrible responsibilities as well. There are no good blueprints for such devastating moral decisions. One group of bioethicists proposed that, should equipment or care be rationed, the young and those who are front-line health workers must have priority (Emanuel et al. 2020). State plans in Alabama call for a scheme in which “persons with severe or profound mental retardation, moderate to severe dementia, or catastrophic neurological complications such as persistent vegetative state are unlikely candidates for ventilator support.” Doctors in Pennsylvania created an eight-point scale, since adopted by hospitals around the country, to calculate a patient’s pre-COVID life expectancy (given any preexisting conditions), as well as their likelihood for surviving their current hospitalization as physicians sort who will—and who will not—get life-saving care (White 2020). Many facilities are considering adopting universal do-not-resuscitate orders for coronavirus patients because of the risks such efforts might pose to hospital staff. These heartbreaking preparations, let alone the actual decisions, felt unthinkable not long ago.

So, I publish this article full of emotion. As I walk my neighborhood and see lines of customers outside grocery stores, carefully spaced and wearing whatever masks or bandanas they might have, I think of the sociological research on culture and crisis discussed herein. Situational disruptions upend everything, large and small. We may not personally have to choose who will live and die, but the sociological research on categorization and classification also discussed helps us see how, in this time of “social distancing,” we are actually being called on to broaden our universe of moral obligation (Fein 1993). Everything we do (and don’t do) now shapes how many of the gut-wrenching life-or-death decisions doctors and nurses will have to make. The most vulnerable among us are suddenly more visible: We are deeply connected not only to the elderly and the medically compromised, but also the homeless, the incarcerated, the undocumented. Our responses must be altruistic to save as many as possible, but they often have not been: We have seen people hoard groceries and price-gouge on stockpiles of normally cheap and truly vital supplies. Four of 100 U.S. Senators (Richard Burr [R-N.C.], Feinstein [D-Calif.], James Inhofe [R-Okla.], and Kelly Loeffler [R-GA]), as soon as they were briefed on the spread of the novel coronavirus, but before that information was public, seized the chance to sell hundreds of thousands of dollars in stocks before the market could collapse, as they knew it would.

More than ever, I am convinced we need to think about how situational circumstances interact with social relationships to shape moral judgments. This is not a call to moral relativism, but an urgent acknowledgement that, when it comes to individual judgment, morality is relative. What we believe to be good and bad gets a little fuzzier when we find ourselves in unfamiliar territory, and so we reconsider our relationships, and who and what truly matters. Our theory is incomplete without this understanding.

I welcome all feedback and thank you in advance for taking the time to read and think broadly.

Aliza Luft

April 5, 2020

I would like to especially thank Letta Page, who helped me to compose this preface at the very last minute with her characteristic clarity and thoughtfulness.

Introduction

How do we know when an action is wrong? Research on moral judgment suggests that participating in, observing, or even thinking about moral violations automatically triggers a negative emotional response. As Jonathan Haidt (2007) explained with reference to physical violence, “when we think about sticking a pin into a child’s hand, or we hear a story about a person slapping her father, most of us have an automatic intuitive reaction that includes a flash of negative affect” (p. 998). Haidt’s argument is premised on a wide body of scholarship suggesting that fast and slow cognitive processes underlie human cognition and that has become increasingly influential in sociology (e.g., Chaiken and Trope 1999; Kahneman 2011; Lizardo et al. 2016; Miles, Charron-Chénier, and Schleifer 2019). However, within the cognitive sciences, significant advances have been made in recent years highlighting Haidt’s inattention to situational variation—in particular, whether a dilemma is familiar—while a separate body of work calls attention to the relationship between affect and social perception—how unconsciously held beliefs about others shape how we feel about and subsequently treat them. As a result, in this article I argue that moral judgment is not simply about actions but about situations and relationships. Moreover, sociologists, who have long been attuned to how situations and relationships influence human judgment, are uniquely poised to consider these influences in our analyses. Thus, to complement the growing incorporation of research from cognitive sciences into sociology, in this article I provide an update on advances in the former and how our own disciplinary expertise can both deepen and extend these insights concerning morality.

The article proceeds as follows. First, I review research on culture in thinking about action ¹ with regard to moral judgment, highlighting two prominent models in the cognitive sciences: Jonathan Haidt’s social institutionist model and Joshua Greene’s digital camera model. The former has received significant attention in sociology, but the latter, which is highly influential in moral psychology, has not. I suggest that the latter is especially worth incorporating into our research, as its focus on situational variation mirrors much work in our own discipline that emphasizes how social situations contain their own meanings and expectations that shape people’s judgments and actions. Second, I argue that although both these models have made significant contributions to sociological understandings of the culture-cognition nexus, they are flawed because of their neglect of the crucial mediating role of social perception. This neglect is evident in the experiments that underlie their models, which hinge on vignettes in which people are raced, gendered, and otherwise classified in ways that sociological research has long identified as significant for shaping evaluation. Third, I review research on social perception to explain the cognitive mechanisms of culture in thinking about people and argue that action and perception cannot be divorced. Finally, I apply the argument to evaluate the recent “moral machine experiment” (Awad et al. 2018) and demonstrate how a sociological approach that combines culture in thinking about action with culture in thinking about people can enhance and extend this work. In the conclusion I discuss the implications of this article and propose directions for future research.

Moral Judgment I: Culture in Thinking about Action

In recent years, the idea that emotions guide moral judgment has gained prominence thanks to two groundbreaking models that have emerged in psychology to challenge the decades-old assumption that morality is the result of rational thought: Jonathan Haidt’s social intuitionist model and Joshua Greene’s digital camera model. The former, also known as the elephant and rider model, ² proposes that people make choices of right and wrong on the basis of strongly felt, intuitive reactions to moral dilemmas (type I cognitive processing). Moral reasoning is a post hoc process emerging to justify the initial affective response (type II cognitive processing) (Haidt 2001, 2005, 2007). In the latter, Greene counters Haidt’s conception by calling attention to the influence of situations on moral judgment. According to Greene (Greene 2009; Greene et al. 2001), when things are normal and operating as expected, type I cognition drives evaluation. Like the preset mode of a camera with fixed settings for taking photos in different contexts, type I heuristics offer shortcuts by shaping feelings and guiding judgment in normal times. However, in novel or challenging situations, in which the obvious setting is not clear, “our conscious reasoning and gut feelings dialogue” and more deliberate adjustments become necessary (Greene 2017:67). The result is type II cognition, whereby slow, explicit, and conscious calculations emerge as cognitive manual adjustments. “Moral dilemmas vary in the extent to which they engage emotional processing,” Greene et al. (2001:2105) wrote, and new and complex problems challenge us because our heuristics lack the requisite training to guide us.

Haidt’s social intuitionist model has become deeply influential in contemporary sociological theories of morality largely thanks to a now widely cited article by Vaisey (2009), who introduced Haidt to a sociological audience. According to Vaisey, internalized beliefs and values motivate moral (thinking about) action (see note 1) by providing people with implicit understandings of what is good and worth striving for or bad and requiring avoidance. Like emotions in the social intuitionist model, these internalized beliefs guide our judgments unconsciously; they become explicit as justifications only when the demands of social interaction require. Thus type I cognition is akin to sociological theories of practice whereby embodied dispositions guide action (Bourdieu 1984, 1990a; Giddens 1984). On the other hand, type II cognition resembles Swidler’s (1986, 2001) toolkit model in which cultural meanings justify action, though for Vaisey, these explicit justifications emerge only when people are required to provide them. Vaisey demonstrates his argument by comparing teenagers’ fixed-survey and in-depth interview responses to questions about their values and moral dilemmas and argues that adolescents’ moral intuitions as expressed in surveys better predict their future actions than do their explicit answers to interview questions about prosocial and deviant behaviors. He concludes that if we wish to understand the influence of moral values on action, our research methods must access practical type I as opposed to discursive type II consciousness (see also Martin 2010; Vaisey 2014).

Since Vaisey’s publication, there has been a surge in sociological scholarship examining how acquired implicit values motivate thinking about action as well as action itself from a dual-process perspective (Lizardo et al. 2016). For example, macro-level studies compare the relationship between unconscious moral schemas and self-reported behaviors cross-nationally (Longest, Hitlin, and Vaisey 2013; Miles 2015) and nationally (Miles 2014b) with regard to voting behavior (Johnson et al. 2014) and volunteering (Beyerlein and Vaisey 2013). At the meso level, studies examine how young adults’ changing moral schemas are associated with changes in their personal social networks (Vaisey and Lizardo 2010) and how largely unconscious views of oneself as independent or collaborative influence the likelihood an individual will join a boundary-spanning professional network (Srivastava and Banaji 2011). Finally, micro-level research considers how embodied moral orientations about idealized futures (Frye 2012), the sacredness of the environment (Farrell 2013), or religious-moral values (Winchester 2016) shape individual behaviors. In each of these examples, the main idea is that implicit type I value commitments drive judgments and actions and explicit type II thinking emerges after the fact, and only when prompted, to explain behavioral choices.

Situational Variations

Despite the important contributions of this work, sociological scholarship has tended to prioritize embodied moral values as independently shaping action without attending to how situational variations influence the judgment-action dynamic, especially situations with novel and unfamiliar features (but see Farrell 2014). To be sure, one of the most significant contributions of sociology to the psychological research on morality is attention to contextual-framed-as-cultural variation in what gets classified as moral: when and why, and by whom (Hitlin and Vaisey 2013; Lamont et al. 2017). But often, this cultural variation is conceptualized as static: two or more social groups’ values are compared as if individual group members’ moral judgment processes are unaffected by situational contexts and contingency. In other words, there is a tendency to treat culture as geography by comparing cross-nationally or as sociodemography by comparing across social categories and less attention to culture as features of situations that may lead to intragroup or even intra-individual variation (but see Miles 2014a; Schilke and Rossman 2018; Stets and Carter 2012). ³ Yet, as many social theorists have noted (e.g., Berger and Luckmann 1966; Collins 2004; Garfinkel 1964; Goffman 1959, 1961; Mead 1934), and as a wide body of research has demonstrated since (see below), the features of situations shape judgments and actions as each situation contains its own meanings and expectations, sometimes even built in the process of interaction itself. Even our social identities are situationally situated as different aspects of our selves can be rendered more or less relevant depending on features of the situation (Brubaker and Cooper 2000:14; Rippon et al. 2014:4–5, in Pitts-Taylor 2016:34), shaping implicit judgment in turn (Xiao and Van Bavel 2019).

Subsequently, and building on a pioneering article by DiMaggio (1997), a wide and separate body of research on culture and cognition examines how different features of situations shape judgments and actions. This includes features of the physical environment such as the setting one is in (e.g., a park, a church, on the street; Shepherd 2011:129) and characteristics of the setting including its sounds (Schwarz 2015) and smells (Cerulo 2018). It also includes features of the material environment: how the material qualities of objects shape the meanings people attribute to situations (e.g., Griswold, Mangione, and McDonnell 2013; Klett 2014 ⁴ ; Martin 2011; McDonnell 2010; Mukerji 1994; Zubrzycki 2013). Interactional features of situations matter, too, including both verbal and nonverbal communication: language, voice, accents, intonation, grammar, touch, bodily signals, and gestures, to name a few (Alex 2008; Ignatow 2007; Lembo forthcoming; Martin 2010). Finally, features of the social environment that shape judgment and action include observations of the behaviors of others (Paluck and Shepherd 2012; Shepherd and Paluck 2015), which depend on the network characteristics of one’s relationships (Shepherd 2017), as well as on the social characteristics of the actors involved, a crucial feature of any situation that includes human beings, real or imagined, and to which I return below. Combined, physical, material, interactional, and social features of situations can be classified as forms of “public culture” that provide knowledge about the world in ways that are informational and meaningful, depending, always, on a given receivers’ “personal culture”: internal cognitive associations developed through repeat embodied experience (Lizardo 2017; Patterson 2014; Strauss and Quinn 1997). ⁵ To the extent that individuals’ personal cultures resonate with features of the situations they encounter, they are able to “solve” practical problems and move smoothly through the world (McDonnell, Bail, and Tavory 2017).

What happens, then, when actors lack the requisite personal culture needed to make sense of the situation they are in? In other words, what happens when people find themselves in novel and unfamiliar situations: places with strange sounds and smells, where people speak different languages and use foreign gestures? Or settings with unfamiliar objects, or where the material qualities of objects have suddenly changed? And what about those situations in which networks and patterns of interactions are disarrayed, or in which people are not who you once thought them to be? At the micro level, these are precisely the kinds of situations Swidler (1986, 2001) referred to as “unsettled,” that are characterized by breakdowns in “taken-for-granted externalized cultural scaffolding” (Swidler 1986:278–79), and that Vaisey (2009:1707) suggested, toward the end of his article, “might favor different mixes of schematic and deliberative processing” (p. 1707). Nascent research on culture and cognition demonstrates that in such situations, type I cognition is inadequate.

For example, Stoltz and Taylor (2017) examined newspaper reports of people trying to protest by paying with large amounts of small cash and coins and found that these actions trigger strong reactions because they violate the taken-for-granted meaning of money. By using a familiar object in an unfamiliar way, money suddenly comes to require conscious attention and effortful interpretation, amplifying emotions. Similarly, and also with reference to materiality, Rawlings and Childress (2019) showed how when actors in book clubs are confronted with new cultural objects—in this case, a brand new novel—they actively work to determine its shared meaning by interacting with similar others. Through these interactions, the meaning of the book comes to reflect sociodemographic differences, but it did not start out this way. Given the novelty of the text, book club members’ cultural dispositions were insufficient for guiding their judgments. Explicit deliberation through interaction was required.

Each of these examples is of a situation in which only one feature of the environment was unfamiliar—what Lizardo and Strand (2010:214–20) referred to as small “gaps” in the usual “ontological complicity” between embodied dispositions and the external environment. Elsewhere, Foster (2018:151) described these as “conditions of mild uncertainty” whereby people deliberately rely on “practices available in memory or [other aspects of] the cultural environment” to discern a strategy of action.

In contrast, and in a separate stream of research, the sociological research on crisis considers situational disruptions on a much grander scale: total and unexpected structural collapse. These are situations with “unstable” and “nonexistent” cultural scaffolding (Lizardo and Strand 2010:216, 219–20), in which sudden ruptures cause a break in routine practice and an indeterminacy of outcomes. Such situations force consciousness to the foreground as actors struggle to find a way forward (Ermakoff 2008, 2010, 2013; Kurzman 2004; Reed 2016; Sewell 1996). ⁶ They are also especially emotional as the disruption caused by crisis can sow fear, confusion, and insecurity. Yet precisely because routine practices cannot provide guidance, emotions are insufficient to lead the way (Ermakoff 2017). Even Bourdieu with Wacquant wrote that “rational choice may take over” in “times of crisis” when “the routine adjustment of subjective and objective structures is brutally disrupted”—rational choice, in this sense, referring to nonhabitual and explicit problem-solving (quoted in Ermakoff 2010:541, 2013:89). ⁷ Hence, even the “founder” of practice theory proposed that intentional deliberation is required in situations in which public culture is disarrayed (Ermakoff 2010, 2013).

These examples suggest that it is important for sociologists who use dual-process theories of cognition to attend to how features of situations, settled or unsettled, affect moral judgment and action. As individuals move through the world, they encounter various forms of public culture that structure the meanings of situations as morally good, morally bad, or morally neutral, depending on their preexisting dispositions. Yet in novel, unfamiliar, and crisis situations—precisely the kinds of situations Greene writes about—people cannot rely on personal culture to guide them. They must intentionally make their way. Thus Haidt is perhaps sufficient for explaining routine moral judgments (e.g., it is bad to steal), but the moment an unexpected element creeps into the situation (but I need the drug, which costs money I don’t have, to save my spouse), ⁸ we need Greene. By calling attention to the fact that sometimes, “our conscious reasoning and gut feelings dialogue” (Greene 2017:67), the digital camera model can help explain when and why our automatic intuitions about morality are insufficient to guide our judgments, as well as what we do to overcome such dilemmas. It can also help us theorize the difference between how we feel and how we feel about how we feel (Damasio 1999; Wacquant 2015), as well as the work we do to regulate our feelings (Hochschild 1979), including about moral violations (e.g., Feinberg et al. 2012; Schein and Gray 2018). Finally, the digital camera model can help us theorize mismatch between moral judgments and actions, such as when people behave in ways they feel are wrong but choose to do so anyway (e.g., Luft 2015:162–64). When it comes to moral cognition, sometimes, our “hot” and “fast” instincts are enough. But oftentimes, particularly in challenging situations, type II deliberation is necessary.

Consequently, a fruitful step for sociologists of morality involves reconsidering the dominance of Haidt’s elephant-and-rider model, attending to Vaisey’s suggestion that deliberative processing may emerge not only after the fact but also in “unsettled” times when the “social and cultural ‘scaffolding’” to which their practical consciousness has adapted has changed, and putting Greene’s ideas into conversation with the sociological research on the situational influences of culture on cognition. How does personal culture interact with and respond to novel situations and crises to shape whether humans engage in type I or type II cognitive processing to make moral judgments? Precisely how moral dilemmas get resolved and the relationship between this process and larger social dynamics ought to be a greater focus of our research. It is to this latter aspect of the culture-cognition nexus that I now turn.

Moral Judgment II: Culture in Thinking about People

There is a flaw in Haidt’s and Greene’s work that has been slowly creeping into our discipline and that should give us pause, compelling sociologists to bring even more of our tools to bear on psychological theories of moral judgment: both are insufficiently attentive to the ways social perception mediates moral judgment even though our evaluations of others are also influenced by embodied culture. In other words, the same person in the same situation can feel differently about the exact same behavior depending on the subject(s) involved in the interaction. This is because acquired moral dispositions relate not only to judgments of actions but also to judgments of people (Abend 2014:32). Furthermore, behavioral judgments and evaluations of people often, if not always, intersect. Kunda (1999) was succinct: “the categories applied to people and behaviors can affect our expectations about people and our interpretations of their behavior” (p. 43). Hence, social perception is a necessary feature of the cultural environment that needs greater consideration when examining situational influences on moral judgment.

To understand why the critical influence of social perception has been lacking in research on cultural cognition, it is necessary to review the foundational experiments on which each model of moral judgment is founded. Haidt’s social intuitionist model stems from an analysis of responses to his now well-known vignette: Mark and Julie, college-age siblings, decide to engage in well-protected sex a single time as they travel to France on summer vacation. A majority of respondents to the vignette (80 percent) initially assert that this incestuous act is not okay; after some deliberation, the percentage drops a bit lower (68 percent) (Bjorklund, Haidt, and Murphy 2000; Haidt 2001, 2013). In another experiment, Haidt (2013:71) evaluated respondents’ emotional reactions to a woman who, when cleaning her closet, finds an old American flag. She does not want the flag anymore, so she cuts it into pieces and uses the rags to clean her bathroom. And of course, in the citation that opened this article, Haidt wrote about the cognitive-emotional responses one is likely to associate with violence toward a child or by a woman against her father. What is rarely considered in these studies is how each of the imagined actors in Haidt’s experiments are strangers to the subjects doing the evaluating (Bloom 2011:27)—and raced, gendered, and otherwise categorizable strangers at that. Consequently, Haidt’s experiments might be calling forth moral evaluations on the basis of stereotypes about what membership in these categories mean. More precisely, social perception might be influencing the results of these studies without any explicit consideration of how.

For example, sociologists with expertise in correspondence audit studies would be quick to point out that Mark and Julie are widely perceived as stereotypically white names. Others would note that Mark and Julie’s shared status as college-age students on summer vacation in France connotes ideas about privilege and wealth. These classifications matter, because they can shape the emotional valence triggered in response to the dilemma. Had Mark and Julie been named Lakeisha and Jamal, and had they decided to practice incest while on a work break from their jobs at a call center, respondents with different social identities might react with more or less anger or disgust. ⁹ If Mark and Julie were instead named Ethel and Bernard, presented as an elderly, lonely pair of siblings in an old-age home who decide to engage in intercourse to soothe their sadness, the vignette might elicit a tinge of pity along with repulsion. And what if the siblings were named Ethel and Lucille? Or what if the woman cleaning her bathroom with the American flag were, instead, a black man? It is easy to imagine, given the different interpretations Americans have expressed regarding football player Colin Kaepernick’s refusal to stand for the national anthem to protest police violence against African Americans, that how we categorize subjects in research matters. Sociologists have long been attuned to this fact, and social neuroscience research on implicit bias provides additional validation for the idea that how we perceive others influences our automatic, type I evaluations of them and their actions (e.g., Gawronski and Payne 2010; Greenwald and Banaji 1995; Nosek, Hawkins, and Frazier 2011; see also Shepherd 2011 and Lamont et al. 2017 for useful sociological reviews of this work). Yet this attention to detail has been missing in the foundational models of moral judgment we have been incorporating into our theories.

Similarly, although Greene’s digital camera model presents an advantage over Haidt’s in its attention to contingency, it too neglects to consider how social perception mediates moral judgment. To develop his model, Greene examined a now classic battery of 40 moral dilemmas, including the well-known switch and footbridge renditions of the classic trolley problem (Foot 1967; Thomson 1976, 1985). In the switch case, a person has the option to flip a switch moving a runaway trolley from a path on which it will hit four workmen onto another, where it will hit just one. In the footbridge case, a person has the option to push a large man off a footbridge and onto the path of the runaway trolley, an action that will result in saving five workmen further down the track. How do people decide on the morally correct action? Greene et al. (2001) argued that such unfamiliar and challenging situations compel explicit evaluations of our moral principles: type II cognition.

However, what Greene fails to mention is how the descriptions of subjects as “workmen” and “a stranger who happens to be very large” might be influencing his results. Moreover, in the original version of the trolley problem’s footbridge variation, the man on the footbridge is described as “a fat man, a really fat man” (Thompson 1985:1409). In his book, Greene (2014:114) removed the phrasing of his 2001 study describing the stranger as “very large” and replaced it with a description of a man wearing a large backpack instead. This suggests that Greene knew that describing an imagined subject as a “stranger” versus a “man” and as someone wearing a “large backpack” versus someone who is “very large” has consequences for how respondents might interpret the vignette, but he did not discuss why he made these changes or whether he believed that they affected the results of his experiments.

Of course, as plenty of sociological research attests, fat bodies are perceived as problematic and bad, and fatness is highly stigmatized, leading fat people to be mocked, bullied, and discriminated against in multiple institutional contexts (Brewis 2014; Saguy 2012). Conversely, thinner bodies tend to be perceived as “morally, medically, aesthetically, and sexually desirable” (Saguy and Ward 2011:54). Likewise, gender affects how people are perceived and treated: in a recently released study of gender norms and inequalities, nearly 90 percent of men and women in 75 countries were found to display bias against women in politics, work, and education, and 28 percent considered it acceptable for a man to beat his wife (a proxy for beliefs about women’s physical integrity; Conceição et al. 2019). And what if we consider age? Presumably, the “workmen” in the switch and footbridge experiments are of working age; this might elicit a mental image of an able-bodied man aged 25 to 65. But what if instead of a “really fat” or “very large” “workman” on the footbridge, respondents were told to imagine pushing a little girl in front of the trolley? Or if, instead of five workmen on the tracks, the victims were described as a group of five mixed-sex schoolchildren? Other experiments in Greene’s study ask if it is appropriate, as the head of a household in a developing country, to employ your daughter in child pornography to help feed your family, which also includes two sons, or—more mildly—whether it is acceptable as a receptionist in a dentist’s office to arrange appointments for Mrs. Santiago and Mr. Morris in a way that privileges Mrs. Santiago’s needs. Even as he offered such variants, Greene offered no discussion about how each of these names, genders, or other descriptive modifiers matters.

Social Perception

In contrast, research on social perception explicitly examines the culturally informed cognitive mechanisms involved in recognizing socially significant information. In other words, it is about culture in thinking about people, and it is the first cognitive process involved in understanding the social world. Lizardo (2018) explained,

perception, at its most basic level, is simply identification and identification is specification. And specification is the product of a relation. That is, a world opens up for an organism when the organism is able to specify, and thus make “contact,” with that world in relation to itself.

Consequently, perception is about sense-making, and social perception is about sense-making for ourselves in relation to others. Perception is the primary building block of social categorization and classification, both cultural cognitive practices of evaluation learned and embodied through experience, observation, and practice.

Social categorization refers to the grouping of traits (features of organisms) we believe belong together and in contradistinction to others (Brubaker, Loveman, and Stamatov 2004; Lakoff 1987; Macrae and Bodenhausen 2000; Smith 1990; Zerubavel 1996). This cognitive process of categorization is common to all animals, yet humans are unique in that no human is born with preexisting knowledge of the social categories we use to make sense of our world. This does not mean that we do not recognize difference; as infants, we prefer and recognize the gender of our primary caretakers, be they female or male (Quinn et al. 2002), and we prefer and recognize familiar versus unfamiliar accents (Kinzler, Dupoux, and Spelke 2007) as well as faces that are similar to ours in skin tone versus those that are not (Lee, Quinn, and Pascalis 2017). However, we are unable to associate these traits with the wider categories of woman and man or native and foreigner or even common social category rules about race until later (Hirschfeld 1998; Lee et al. 2017; Liberman, Woodward, and Kinzler 2017; Shutts, Pemberton Roben, and Spelke 2013). For example, white nine-month-olds tend to separate white faces into one category and black and Asian faces into another, suggesting that they can recognize their own skin-color category at this stage and that less familiar skin tones are lumped together into an “all other faces” category (Lee et al. 2017:258). In contrast, white infants who are exposed equally to their own and other-race faces show no separate recognition or categorization of different “kinds” (Bar-Haim et al. 2006; Pauker et al. 2016). Humans are predisposed to recognize some traits as different from others, and they are also predisposed to prefer those most similar to themselves, but how they understand the meanings of these traits and their relevance for social organization differs depending on the cultural environment in which humans are raised (Liberman et al. 2017). ¹⁰

Following this, the second crucial component of culture in thinking about people is classification. Whereas categorization is the “lumping and splitting” of people into different perceptual kinds (Zerubavel 1996), classification is the process of giving categories content to allow us to extract meaning from them, to communicate with them, and to generalize from one experience to another (Kunda 1999). Categorization is descriptive; classification is substantive (Bowker and Star 1999). Classification helps us compare people not simply on the basis of their observable similarities or differences but also on the basis of their presumed intrinsic and essential similarities or differences. Marion Fourcade (2016:176) referred to this as “nominal classification”: classificatory judgments that describe inherent characteristics and relations. Following this, classification allows us to make inferences about people we perceive on the basis of culturally learned ideas about what kinds of characteristics are associated with what kinds of people. This, in turn, helps us determine how to relate to them (Brubaker et al. 2004).

Ostensibly, nominal classification schemes can result in horizontal divisions between social(ly constructed) kinds, but in practice, they are almost always tied to ordinal judgments—classification schemes that attribute value to the kinds so defined (Fourcade 2016:178)—and we frequently use these judgments to vertically rank different categories of people (Fourcade 2016:178; see also Brubaker 2015). Such rankings are far from benign (Bowker and Star 1999; Bourdieu 1990b). As a growing body of research reveals, those we classify as highly valuable are also those to whom we accord high status. We respect and perceive them as good and admirable, and we reward them with symbolic and material resources. In contrast, those we classify as less valuable are accorded lower status. We disrespect and perceive them as bad, unworthy, and even deplorable, and we punish them by inhibiting their access to material and symbolic resources (Lamont 1992, 2000, 2018; Prasad et al. 2009; Ridgeway 2006). Subsequently, classification is related to moral judgment because how we classify categories of people informs how we think and feel we ought to treat them (Lamont 2018).

Finally, although often attended to in our discipline from the perspective of practice—that is, how categorization and classification schemes organize the world socially and politically through time and space—cognitive research on social perception attends to how these practices become acquired and influence judgment and action typically without our explicit awareness (Lamont 2012:205). ¹¹ In fact, we rarely perceive people and then parse through their various descriptive traits before determining to which category they belong, nor do we reflect on various classification schemes before deciding how to react to those people. Rather, social perception is beholden to the same two cognitive processes described earlier. Most of the time, when we see others, our categorization and classification processes automatically activate (Amodio and Mendoza 2010; Fiske and Neuberg 1990). Fast, efficient, type I cognitive processes guide the majority of social perception and help resolve the “blooming, buzzing confusion” of our world (Brubaker 2002:71). Thus as we find ourselves moving from one setting to another, evaluating different people in different positions, cognitive associations as acquired dispositions trigger in response to our social perceptions.

That said, and much like Greene’s model, there are times when type I cognition is insufficient to guide our evaluations of others, and more deliberate, type II cognition is necessary (Ito and Tomelleri 2017). In particular, people most often use fast, automatic, heuristic-based processes to perceive others, but when others don’t “fit” known categories, they use a slow, deliberate, rational approach (Fiske and Taylor 2013:109). As a result, intentional type II deliberate processes can be activated (1) in cases of failed interactions in which there is mismatch between our expectations and reality (Macrae et al. 1999); (2) when category membership is ambiguous, thus prompting a concerted effort to think through how to categorize and classify the perceived (Freeman and Ambady 2011; Freeman et al. 2011); and (3) when we intentionally want to control our perceptions of others, for example, to suppress unwanted stereotypes because of conflicting personal value commitments to equality (Amodio and Devine 2010; Amodio and Swencionis 2018; Blair 2002; Devine 1989; Eberhardt 2019; Macrae and Bodenhausen 2000). Each of these examples is similar to Greene’s (2009) argument about the introduction of type II cognition in novel or challenging situations, thereby helping actors override automatic responses to, in this case, perceptual dilemmas before determining a response. Thus, once again, depending on the features of the situation and the goals of the perceiver, higher order processes can modulate categorization and classification as well as the link between them (Freeman and Ambady 2011). ¹² In turn, scholarly attention to the social features of situations and, in particular, to the social characteristics of people involved in situations, can generate more precise theories of moral judgment.

The following section probes the argument above with the example of the moral machine experiment, a recent iteration of the trolley problem in which the subjects of the actions have been varied. I then conclude with a discussion that reiterates the proposal herein: sociological theories of moral judgment have much to gain by moving beyond the dominance of Haidt’s elephant-and-rider model in our research and considering Greene’s model, which emphasizes the importance of attending to situational variation, in our analyses. Simultaneously, we also have much to gain by considering the crucial mediating influence of social perception in our research on morality, often neglected in psychological theories of moral judgment but central to sociological understandings of how categorization and classification practices shape human thought and action in times of normalcy and in times of crisis. As sociologists, one of the strongest contributions to the study of morality we can make is greater attention to how social features of situations affect moral judgment.

The Moral Machine Experiment

Slightly more than a year ago, a paper was published in Nature reporting the results of a study of self-driving cars (Awad et al. 2018). ¹³ Developed by the MIT Media Lab, the purpose of the study was to crowd-source individuals’ decisions on how cars should prioritize lives in potentially dangerous scenarios. This included different variations of the trolley problem, including whether it was more ethical for a self-driving car to strike and kill men versus women, humans versus pets, children versus the elderly, fit versus unfit, doctors versus criminals, and so on. ¹⁴ The experiment went viral, and millions of people in 233 countries and territories recorded a total of 39.62 million decisions by the time the results were published.

The results of the moral machine experiment revealed significant cross-country variation, not simply in how likely people were to suggest that an autonomous vehicle should kill one or a few to save others (similar to the classic trolley experiment) but also in what kinds of people (or nonhuman animals, in the case of pets) were more worth saving. The results also showed slight but not significant variation in results depending on respondents’ demographics (age, educational attainment, gender, income, political and religious views) and the experimenters were able to generate three geographic clusters of countries with different moral preferences based on their findings as well as other clusters based on countries’ economic characteristics and cultural values.

For example, people worldwide tended to prefer saving more versus fewer people, but the magnitude of this preference was stronger in countries with individualistic cultures such as France, the United States, the United Kingdom, and Canada than in countries with collectivist cultures such as Japan, Taiwan, and China. Participants from individualistic countries were also more likely to prefer sparing youth (depicted as babies, little girls and boys, and pregnant women), whereas those in collectivist countries were more willing to spare the elderly. People from countries with higher gross domestic products and stronger civic institutions, such as Finland and Japan, tended to prefer saving legal road crossers versus jaywalkers compared with those in poorer countries with weaker institutions, such as Nigeria and Pakistan. In countries with greater economic equality (Finland, again), people showed little preference for saving homeless persons versus executives, whereas in countries with more inequality (e.g., Colombia), respondents were more likely to choose killing the person of lower status. Finally, although individual demographic variations did not significantly affect responses below the .01 value, the most notable effects were that men were .06 percent less inclined to save women overall, and those who self-identified as “very religious” were .09 percent more likely to save humans over pets.

The moral machine experiment was designed to evaluate the ethical judgments of decisions made by autonomous vehicles in unavoidable accident scenarios, but in the process, it uncovered significant cross-cultural variations in how people assess the worth of different kinds of life. By asking people all over the world to decide who must be spared and sacrificed, and by varying the characteristics of the subjects in each experiment, the findings of the study not only revealed geographic variations in preferences for moral behaviors (killing vs. saving), it also revealed within-country variations in who or what kinds of people these preferences are for. Hence, the results of the study demonstrate one aspect of the argument in this article: although moral judgments of the same action depend in part on the broader cultural context, a point made by Haidt and many sociologists as well, they also depend on the social characteristics of those involved in the situation. In other words, the same action, when enacted against differently categorized people, can be considered more or less moral by the same individual. Respondents’ moral judgments change depending on the subjects involved in the experiment. This suggests that culture in thinking about people matters, and it is important for scholars to attend to the social features of situations when researching morality.

Simultaneously, there is room for improvement with the moral machine experiment and space for sociologists to contribute even more. For example, the moral machine researchers are unable to show whether their version of the trolley problem compelled implicit or explicit thinking about action or whether, with repeat experience, people moved faster through each situational variation as they adapted to the challenge of choosing whom to save and whom to kill. ¹⁵ Alternatively, as the social characteristics of the people involved in each situation changed, subjects may have taken more time to determine a response. Lengthier deliberations in such cases could be the result of repeatedly having to choose between multiple categories of people with culturally “high” moral value (e.g., three boys and two men vs. five elderly women, followed by one elderly woman, a man, and a boy vs. a man, a woman, and a baby), or the result of being unable, in some instances, to discern to what category an individual or group of individuals belonged. According to the arguments herein, both situations would disrupt implicit judgments of right and wrong and compel lengthier, explicit deliberations.

Additionally, although we have information about the sociodemographic characteristics of test-takers, it is unclear which aspects of their social identities were relevant for their judgments, how they may or may not have intersected depending on the scenario considered, or if an altogether different aspect of their social identity mattered more. For example, a respondent could be a parent, recently unemployed, or someone who identifies with or is frequently identified by others as belonging to a racialized minority. We all have many ways of thinking about ourselves, far beyond the categories examined in this experiment. Which identity drives our moral judgment, and how identity matters for moral judgment, including the moral judgment of other people’s value and worth, depends on the situation we are in. As a result, different social characteristics can become more or less moralized at different moments in time, even for the same individual, shifting the perceived morality or immorality of different actions. Knowing a person’s sociodemographic characteristics can only get us so far. We must also attend to what they mean for different people, and how their meanings can change contingent on situations, altering moral judgment in turn.

A final concern with the moral machine experiment is that we know nothing about the real-world situations in which the experiments were done. Granted, respondents were told to imagine themselves as having “control over choosing what the car should do,” but presumably, even this exercise would have been affected by respondents’ situational environments when they clicked “keep going and hit the pedestrians” or “swerve and hit the pedestrians on the other lane.” Were respondents alone at home in front of their computers when they did the experiment, or were they out with friends, passing time and making choices as a group while reading through the dilemmas on their cell phones? What about respondents who did the experiment as passengers in the back of a cab versus those who enacted judgments absent-mindedly while crossing the street? Given the sensitivity of replicability to context (Van Bavel et al. 2016) and priming (Bargh and Chartrand 2000), it is easy to imagine how such simple and seemingly insignificant variations could in fact be very significant for shaping judgments about the targets of self-driving cars. ¹⁶

The moral machine experiment is no panacea. It cannot tell us about the extent to which each variation was or was not perceived as an unfamiliar or challenging dilemma, compelling implicit versus explicit cognition, or whether some aspects of respondents’ social identities were more or less salient at the time they did the experiment, shaping their evaluations as well. We also do not know how the real-world situations respondents were in when they made their choices might have affected their judgments. What the moral machine experiment can tell us is that the social characteristics of subjects in imaginary moral dilemmas matter, suggesting that in real life, the social features of situations matter too. This is a start worthy of further empirical investigation. Sociologists, who are especially attuned to how situations and social perceptions are culturally informed, can be at the forefront of this exciting research.

Discussion and Conclusion

In this article I argue that moral judgments are determined by actions, situations, and relationships. Dual-process theories of morality emphasize that morality is often experienced as a gut reaction that tells us something is wrong, while recognizing that “our guts learn their intuitions” (Hitlin and Harkness 2018:49; Sapolsky 2017:508). Sociological research that builds on or relates to these ideas considers how cross-cultural variation, with culture often standing in for geography or social groups, alters moral judgments with important consequences for what is and is not considered moral behavior. Yet missing from these approaches is the idea that, depending on the features of the situation and especially the social features of the situation, the same behavior judged by the same person can be considered more or less moral. Moreover, as various situational dilemmas “crop up” in the external environment, our moral judgments can shift from fast and automatic to slow and deliberate as we strive to make sense of others’, or even our own, actions. Many of us behave in some situations in ways we would not consider acceptable in others. This suggests that our moral judgments are much less internally stable than current theories allow. Subsequently, how we evaluate the morality, immorality, or even the moral relevance of an action also depends on features of the situation, including how familiar or novel it is, and on the perceived social categories of the subjects involved in the interaction.

Sociologists, of course, have long been attuned to how different situations call forth different meanings and interpretations and subsequent emotions and actions. We have also long been attuned to how categorization and classification practices affect judgments of worth and value. But in the recent turn toward cultural cognitive understandings of morality, some of these insights have been left behind. This article traces this neglect to the growing reliance in our discipline on Haidt’s social intuitionist model, which no longer dominates in psychology but continues to drive much of the research on dual-process theorizing about morality in sociology. I therefore introduce Greene’s digital camera model as an alternative perspective, especially insofar as situational variations are concerned, and also connect Greene to a wide body of research in sociology on culture, cognition, and crisis, to suggest that theories of morality ought to distinguish between familiar and unfamiliar dilemmas in their research.

Simultaneously, I argue that neither Haidt nor Greene goes far enough in considering what influences moral judgment, and this is because neither considers the crucial mediating influence of social perception. Defined and explained here as the cognitive process that undergirds cultural practices of social categorization and classification, I review how both Haidt and Greene’s models hinge on experiments in which their subjects are raced, classed, gendered, and otherwise described in ways that might alter respondents’ judgments of their actions. Yet neither examines how the perceived category membership of subjects might affect respondents’ moral judgments. Sociologists, who are experts in considering how socially constructed categories influence judgments and actions, have much to contribute here. But first, we must slow down our uncritical incorporation of these theories and consider, using our distinct disciplinary perspective, the possible flaws in the premises of their studies.

Finally, I assess the moral machine experiment conducted by the MIT Media Lab to illustrate different aspects of my argument. The goal of the moral machine experiment is to determine what ethical principles ought to guide machine behavior in the case of self-driving cars. In researchers’ variation of the trolley problem, they confronted respondents with unavoidable accident scenarios in which an autonomous vehicle could either stay on course and kill some people or swerve and kill others. Participants were given 13 accidents to evaluate, and in each, the subjects who could be killed or saved were altered. Hence, much of the situation stayed the same, but in changing the characteristics of those who could be spared or harmed, the experimenters found significant differences in how actors evaluated the rightfulness or wrongfulness of killing. The morality of the action depended on respondents’ evaluations of people involved in the dilemma. This finding validates a central argument in this article: that social perception shapes moral judgment.

Simultaneously, the experiment does not go far enough in its considerations of other arguments herein. In particular, it could be improved with greater attention to timing and thus whether different judgments were made “fast” or “slow” or even changed from fast to slow (or vice versa) depending on the social characteristics of the people involved. It could also be improved by considering that people identify with multiple and intersecting social categories and whether any particular aspect of one’s identity triggers as relevant for shaping judgment depends on features of the situation as well. I have discussed this only briefly in this article, but how people categorize themselves and how such self-categorizations vary is absolutely an important feature of moral judgment worthy of further theorizing. Finally, it would help to know the real-world details of the situations people were in when they participated in the experiment, as recent assessments of psychological research show the sensitivity of replicability to context and priming. The importance of understanding morality in its “natural” context is something many sociologists have written about (e.g., Hitlin and Vaisey 2013) and an obvious challenge for experiments such as the moral machine, which can only get us so far.

Consequently, my analysis of the moral machine experiment demonstrates the argument at the heart of this article and also where, as a discipline, we have much to contribute to research on morality. In particular, sociologists are well placed to examine how anytime an individual makes a moral judgment, his or her sense of right or wrong is determined by three factors: (1) the action itself, (2) the situation in which the action is performed, and (3) the people involved in the interaction. This focus on actions, situations, and relationships joins together culture in thinking about action with culture in thinking about people, and in so doing, it extends contemporary trends in cultural cognitive sociology with recent developments in psychological research on morality as well as timeworn insights from our own discipline: that the cultural features of situations, including and especially the social features of situations, shape our judgments and actions. Hence, future research on morality ought to more fully consider the nexus of actions, situations, and relationships in their analyses. In combining existing research on moral cognition with sociological understandings of how situations and social perceptions affect judgment, we can begin to craft a fuller picture of the axes of human behavior.