Falsifying the Dehumanization Hypothesis

Discrimination is a pressing, global problem (Smith, 2011, 2016). Like the commentators on my original article (Giner-Sorolla et al., 2021; Goldenberg et al., 2021; Smith, 2021; Vaes et al., 2021), I believe that understanding the psychological causes underlying discrimination will bring us one small step closer to reducing its prevalence and impact. A particularly influential idea within the social sciences is that a psychological process of dehumanization is one important cause of discrimination (Harris & Fiske, 2011; Haslam, 2006; Smith, 2016). According to the dehumanization hypothesis, victims of intergroup harm are perceived as similar to nonhuman entities. As a result, they are rendered more vulnerable to harm.

In my original article, I questioned the dehumanization hypothesis and argued that the construct of dehumanization may have less explanatory power than it initially appears (Over, 2021). I posed seven challenges for proponents of the dehumanization hypothesis to answer. I was glad to receive four commentaries on these challenges. For the most part, the commentators defended the dehumanization hypothesis. Thus, while the commentators and I are united on the need to understand and reduce discrimination, we remain divided on how best to characterize psychological processes that contribute to it.

I begin this response by contrasting different variants of the dehumanization hypothesis with my alternative viewpoint. I then turn my attention to suggestions for future research. I focus on future research because I believe that, in cases of continuing disagreement, the most constructive route forward is empirical work that distinguishes among the various alternatives.

I am grateful to the commentators for their important work on these topics and for engaging with my challenges. I hope the continued debate will be productive for the field and, ultimately, for the crucial task of reducing discrimination.

The Dehumanization Hypothesis and the Alternatives

In my original article, I summarized the dehumanization hypothesis as two interrelated claims. First, victims of intergroup harm are perceived as being similar to nonhuman entities. Second, as a result, they are rendered more vulnerable to harm. The commentators each echo a point I tried to articulate in my original article but perhaps did not convey clearly enough. The dehumanization hypothesis is not a single theory but rather a family of theories. Sometimes, these theories disagree. The philosopher Smith (2021) argued that to dehumanize a group is to conceive of them as subhuman creatures. Vaes et al., on the other hand, stated that “no psychological account presents dehumanization as a categorical or qualitative denial of humanity to humans” (p. 28).

I maintain that, despite these differences, the various conceptualizations of the dehumanization hypothesis share similar flaws. First, it is not at all clear that the construct of dehumanization, in any of its current formulations, accurately characterizes how people think about out-groups. Second, it is not clear why conceiving of a group as less than human (Smith, 2011, 2016), or somewhat less human (Haslam, 2006; Haslam & Loughnan, 2014; Leyens et al., 2007; Vaes et al., 2012), should increase the risk of harm against them.

I presented an alternative view. Considering examples of extreme intergroup harm, I argued that it is premature to attribute dehumanizing beliefs to perpetrators. When propagandists describe a target group using such words as “rats,” “lice,” and “parasites,” it is undoubtedly dangerous and deeply offensive. However, it might not reflect a belief that these groups are less human; rather, it may be one strategy among many to present them as low status, threatening, and homogeneous in character (Bloom, 2017; Manne, 2016). I also argued that apparent evidence for subtle forms of dehumanization from lab-based studies may, in fact, collapse to in-group preference and stereotyping. Rather than attributing traits that are “more human” to our in-groups (Haslam, 2006; Haslam & Loughnan, 2014), I suggest that we attribute characteristics that are “more positive” to them. Rather than perceiving our in-groups to experience human-like emotions more strongly (Leyens et al., 2007), I suggest we perceive our in-groups to experience prosocial emotions more strongly.

The commentators responded to my critique of the dehumanization hypothesis by outlining a wide range of data that they believe support their various formulations. There are data suggesting that we sometimes dehumanize racial groups, national groups, individuals who attend different universities, cyclists, artists, businessmen, asexuals, individuals with mental-health problems, doctors, and people with particularly wide faces (Delbosc et al., 2019; Deska et al., 2018; Goff et al., 2008; Haslam, 2006; Leyens et al., 2007; Loughnan & Haslam, 2007; MacInnis & Hodson, 2012; Schroeder & Fishbach, 2015).

The commentaries by Vaes et al. (2021), Goldenberg et al. (2021), and Giner-Sorolla et al. (2021) all voice a second defense. That is, many moderating variables are at play in these complex social situations. Thus, only some out-groups are dehumanized, and these out-groups are dehumanized only some of the time. For example, right-wing people (but not left-wing people) tend to dehumanize Muslim terrorists and do so more strongly when their own mortality has been made salient to them (Sanchez & Garcia, 2016).

However, there is little value in demonstrating that many groups appear to be dehumanized or in showing that additional variables appear to moderate these effects if the construct of dehumanization has been operationalized in a problematic way. I argue that because all of these previous studies have been based on similar underlying formulations of dehumanization that fail to distinguish the phenomenon from stereotyping and prejudice, they share similar flaws.

Looking to the future, I issue one further challenge to proponents of the dehumanization hypothesis. That is, to outline the conditions that would falsify their various theories, I use the remainder of this commentary to take the first steps down this path of falsification. To avoid any confusion, some of the studies I suggest apply more to some formulations of the dehumanization hypothesis than to others. I begin by discussing the research of Smith (2011, 2014, 2016). I then move on to discuss how to falsify infrahumanization theory (Leyens et al., 2007; Vaes et al., 2012) and Haslam’s dual-route model (Haslam, 2006, 2019; Haslam & Loughnan, 2014).

Testing Smith’s Theory

Smith (2011, 2014) argues that when we dehumanize others, we conceive of them as “less than human.” As evidence for this theory, Smith draws on historical data. For example, propagandists in Nazi Germany sometimes referred to their Jewish victims as “rats,” “lice,” and “parasites.” In the American South, African American people who were enslaved were referred to as “subhuman” and “ape-like.” Smith (2011, 2014, 2016) has collated a huge number of such horrifying examples this from diverse geographical locations and time periods. In some of these examples, the writers appear to intend their words to be taken literally (Smith, 2016). In his commentary, Smith argues that we should take these examples seriously, and I agree. Indeed, many psychologists seem to agree, given that these examples are regularly cited as cases of extreme dehumanization in the psychological literature (Harris & Fiske, 2011; Haslam, 2019; Haslam & Loughnan, 2014).

However, taking a hypothesis seriously is not the same as endorsing it. It is possible that these examples, numerous though they may be, are the product of a confirmation bias (Nickerson, 1998). Proponents of the dehumanization hypothesis have meticulously searched the historical record for examples of propaganda and writing in which out-groups are compared with nonhuman entities. Are there counterexamples? If so, how common are they? In an earnest attempt to find evidence in favor of the theory, counterexamples may have been overlooked. One class of counterexamples consists of those in which in-groups refer to themselves as similar to nonhuman entities. For example, referring to themselves as “lions,” “rats,” or “rattlesnakes” in pro–in-group propaganda. These examples are problematic for Smith’s (2011) dehumanization hypothesis because they suggest that there might not be a unique relation, or even a probabilistic relation, between being compared with a nonhuman entity and being discriminated against. In another class of counterexamples, out-group members are referred to in terms that make sense only when applied to humans (e.g., “enemies,” “criminals,” and “traitors”). These cases are equally problematic for Smith’s (2011) dehumanization hypothesis because they suggest that the targets of propaganda might, at least implicitly, have been represented as human (Bloom, 2017; Manne, 2016). In their commentary, Giner-Sorolla et al. (2021) speculate that confirmatory examples might be more common than the counterexamples I highlight. At present, however, it is impossible to know because the literature has not yet been systematically searched for both types of example.

Needless to say, claims based on historical data cannot be tested in the same way as hypotheses from lab-based research. The methods of experimental psychology are ill-equipped to understand extreme intergroup harm of the type Smith studies. Smith’s commentary rightly emphasizes the scale of the challenge involved in understanding the historical data. However, if scientists are going to use such data to inform their psychological theories, then it is important to start to test them with the best tools available. One way to start to test the claim that victims of historical atrocities are subject to extreme dehumanization would be to preregister content analyses of historical documents in which confirmatory and disconfirmatory examples are both coded. For example, Nazi propaganda from the 1930s and 1940s and Rwandan radio broadcasts from before and during the 1994 genocide could be coded. Such content analyses would be informative because they would uncover the relative frequency of “dehumanizing” language. The claim that the victims of this propaganda were dehumanized would gain support from a relatively high frequency of comparisons between out-groups and nonhuman entities.

As Goldenberg et al. point out in their commentary, it is possible that victims of propaganda might sometimes be dehumanized and sometimes be marginalized in other ways that do not involve dehumanization. This is a reasonable hypothesis and one it is possible to start to test. If this is the case, then we might expect to see a relatively high frequency of comparisons to nonhuman entities in some texts and a paucity in others. If, however, comparisons between out-groups and nonhuman entities are merely one form of slur among many, then we might expect these comparisons to appear alongside other types of slur. For example, references to the out-group as “enemies” and “traitors” would appear in the same text as references to these groups as “rats” and “lice.”

Smith has revised his theory to account for some of these concerns. In more recent work, Smith (2016) argued that victims of dehumanization are viewed as simultaneously human and subhuman and, as a result of this simultaneous membership in two categories, appear horrifying to us. This new theory can explain the existence of the counterexamples that I and others (Bloom, 2017; Lang, 2010, 2020; Manne, 2016) have highlighted. However, other counterexamples appear to present themselves. Anthropomorphized characters, such as those in cartoons, could be considered simultaneously human and nonhuman and yet we do not typically regard them as horrifying. Nor do we seem to find in-group members horrifying when they are referred to as “lions” or “bears.” Regardless, the same need for falsification applies to this new theory from Smith too. What type of evidence would falsify the claim that dehumanized individuals are viewed as both human and subhuman? I hope the field will address this difficult but important question in future work.

Testing Infrahumanization Theory

Leyens and colleagues (2007; Vaes et al., 2012) claim that dehumanization can take place in subtle ways and that these subtle forms of dehumanization are prevalent within Western society. They use the term infrahumanization for this hypothesized subtle form of dehumanization. According to this account, when out-groups are infrahumanized, they are thought to experience uniquely human or “secondary” emotions, such as pride and nostalgia, to a lesser extent than do the in-group. Leyens and colleagues (2007) explicitly argue that infrahumanization is separable from in-group preference because out-groups are thought to experience both positive and negative secondary emotions to a lesser extent than do in-groups. For example, out-groups are thought to experience the negative emotion of guilt to a lesser extent than do the in-group (Demoulin et al., 2009; Leyens et al., 2003, 2007).

This, however, is a conceptual misunderstanding. Guilt is a negative emotion in the sense that it is negative to experience. However, it is prosocial in character. Guilt is thought to foster prosocial, reparative responses. As a result, individuals who experience guilt are thought to be nicer than individuals who do not experience guilt (Stearns & Parrott, 2012). Thus what appears to be evidence for a psychological process of infrahumanization may, in fact, be evidence that individuals tend to attribute more prosocial traits to their in-group.

One way to distinguish these two alternatives would be to compare participants’ judgments of uniquely human secondary emotions that are prosocial (e.g., compassion) and antisocial (e.g., hostility) in character, rather than positive and negative to experience, and how they apply to different groups. If participants believe that in-group members typically experience secondary emotions more strongly, then this should hold true regardless of whether the emotions are prosocial or antisocial. If what appears to be evidence for infrahumanization is, in fact, evidence for more basic processes of in-group preference and stereotyping, then participants should report that in-group members experience prosocial emotions to a greater extent and out-group members experience antisocial emotions to a greater extent.

Testing Haslam’s Dual-Route Theory

Haslam (2006) sought to characterize dehumanization by first establishing the content of the lay category “human.” Haslam (2006; Haslam & Loughnan, 2014) argues that there are two forms of humanness. Qualities such as civility, refinement, moral sensibility, rationality, and maturity are thought to be specific to humans (so-called human-unique traits). Qualities such as emotional responsiveness, interpersonal warmth, openness, individuality, and depth are thought to be characteristic of humans (so-called human-nature traits). When a group is dehumanized, they are hypothesized to be attributed some or all of these traits to a lesser extent.

The trouble is that Haslam’s account appears not to accurately characterize the content of the category “human.” I suggest that, because of errors in how the content of the category was measured, Haslam’s characterization of human traits weighs positive and prosocial characteristics too heavily. That is, the particular questions Haslam asked of participants may inadvertently have put the focus on more positive attributes of humanity at the expense of more negative ones. I am not suggesting that negative and antisocial traits are completely absent from Haslam’s characterization of human traits but rather that they are underrepresented.

I do not doubt that humans are sometimes viewed as deep, rational, and civilized, but what of more negative human characteristics? Surely there are situations in which humans are viewed as spiteful, arrogant, and disloyal. In their commentary, Vaes et al. (2021) defend Haslam’s model against this critique by arguing that whereas rationality, civility, and depth are human traits, spite and disloyalty are ways of evaluating other humans. The subtlety of this distinction may be lost on many. Fortunately, however, Vaes et al. suggest an empirical test for this claim. To be considered human traits, attributes such as spite, arrogance, and jealousy would have to be considered more human than other character traits. Such a test is easily devised.

An undue reliance on positive traits in characterizing the lay category of human is problematic because it means that apparent evidence for dehumanization is confounded with in-group favoritism. The tendency to attribute positive traits such as warmth, depth, and civility more strongly to in-group members could represent a tendency to subtly dehumanize the out-group, but it could also represent a tendency to make more positive attributions toward the in-group.

These two explanations for the observed pattern of results in previous dehumanization studies can be disentangled by incorporating more negative and antisocial human characteristics into the stimulus set. If out-groups are subtly dehumanized, then participants will report that out-groups possess both positive human attributes (e.g., warmth and depth) and negative human attributes (e.g., spite and jealousy) but to a lesser extent than do in-groups. If apparent evidence for dehumanization can be explained by in-group favoritism, then participants will attribute positive human traits more strongly to the in-group and negative human traits more strongly to the out-group.

Testing the Hypothesized Connection Between Dehumanization and Harm

Although there are many reasons to be interested in the construct of dehumanization, much of the interest in the field stems from the idea that dehumanization plays a causal role in intergroup harm. For example, Smith (2016) eloquently describes dehumanization as “a psychological lubricant for the machinery of violence” (p. 46). Likewise, Haslam and Loughnan (2014) argue that “dehumanization is important as a psychological phenomenon because it can be so common and yet so dire in its consequences” (p. 401). The hypothesized causal relation between dehumanization and harm is further underlined by recent articles entitled “The Many Roles of Dehumanization in Genocide” (Haslam, 2019) and “How Dehumanization Promotes Harm” (Haslam & Loughnan, 2016).

In my original article, I argued that the hypothesized causal connection between dehumanization and harm is suspect for at least two reasons. First, viewing a group as less than human is not necessary for harm—out-groups are often harmed because of their uniquely human characteristics such as their alleged disloyalty, cunning, or spite (Lang, 2010, 2020; Rai et al., 2017). Second, viewing a group as less than human does not seem sufficient for harm—family pets are considered less than human and yet they are cherished.

The commentaries by Goldenberg et al. (2021) and Giner-Sorolla et al. (2021) actually suggest further reasons to doubt the hypothesized causal connections to harm. According to work they reference, individuals sometimes dehumanize out-group members and sometimes dehumanize in-group members. Sometimes they even dehumanize themselves (Morris et al., 2014). If we sometimes dehumanize individuals we like and protect, then why should we assume that dehumanization ever has a causal relationship to harm?

Lab-based research has attempted to test the hypothesized relationship between dehumanization and harm by investigating variations in prosocial behavior. For example, work stemming from infrahumanization theory has claimed that when a group is infrahumanized, it is less likely to be the recipient of help. In one highly cited study on this topic, Vaes et al. (2021) described someone in humanized terms or not and measured the influence on participants’ prosocial intentions toward this person. To do this, the experimenters sent participants an e-mail that began with either a secondary emotion (e.g., expressing the sender’s disappointment) or a primary emotion (e.g., expressing the sender’s anger). Participants reported that they would be more likely to reply to the message if it started with a secondary emotion than with a primary emotion. Vaes et al. (2002) interpret these results as evidence that people are more helpful when they are interacting with individuals who highlight their humanity through the use of secondary emotion terms. An alternative interpretation, however, is that individuals who start e-mails by expressing their disappointment are viewed as somewhat nicer than individuals who start e-mails by expressing their anger. It is straightforward to envisage an experimental paradigm that could distinguish between these two hypotheses. In a future study, senders could express themselves in terms of uniquely human secondary emotions that are antisocial in character (e.g., referencing feelings of scorn or hostility). It seems unlikely that using such terms would increase prosocial behaviors, uniquely human though they may be.

Given the many problems with the hypothesized causal connection between dehumanization and harm, it may be that several proponents of the dehumanization hypothesis now believe that dehumanization is particularly interesting as a post hoc explanation offered by perpetrators of harm (see the commentary by Vaes et al). It seems plausible to me that perpetrators might sometimes seek to excuse their behavior by claiming that their victims were less than human. However, if dehumanization is merely an excuse, it would tell us little about the causes of intergroup bias.

Concluding Thoughts

To be valuable contributions to scientific debate, theories of dehumanization must be falsifiable (Popper, 1959). It may be that the dehumanization hypothesis, or some variants of it, can withstand the challenges I outlined in my original article and in this response. It may be that we do not need to reject the dehumanization hypothesis, but rather to refine it (as Smith and Giner-Sorolla et al. suggest in their commentaries). Whatever the case, the field will benefit from further theoretical clarification and empirical data as we continue to address our shared goal of reducing the terrible, real-world consequences of intergroup bias.

Transparency

Action Editor: Laura A. King

Editor: Laura A. King