The Heterogeneity of Implicit Biases

Jules Holroyd, The University of Sheffield

Joseph Sweetman, University of Exeter

Are we responsible for implicit biases? What social environments promulgate implicit bias? What institutional and social contexts can mitigate or eliminate implicit bias? What are the epistemic implications of implicit bias? These are just some of questions about implicit bias that philosophers have taken up. But attempts to answer these questions might go wrong if our thinking about them is premised on false assumptions about the nature of implicit bias. One such assumption concerns whether all implicit biases function in the same way, have the same properties, and stand in the same relation to other mental states and behaviours. On one model of implicit bias, it is a unitary phenomenon about which generalisations can be made, such that we should expect answers to the above questions to remain constant irrespective of the particular bias at issue. On another model, implicit biases are diverse phenomena: different biases may generate different answers to the above questions, and we should be cautious about generalisations across different implicit biases.

Some philosophical engagement with questions about implicit bias seems to tacitly assume the first model. In the contribution to Brownstein & Saul’s volumes, Joseph Sweetman and I have the main aim of motivating the second way of thinking about implicit bias, and to suggest that we should treat implicit biases as a heterogeneous phenomena. This leaves open the possibilities that we may be responsible for some implicit biases, and not others; that the epistemic implications of implicit bias might be varied; that our awareness of implicit biases might vary according to the bias at issue; and that mitigation strategies might differ depending on the particular bias targeted.

The line of argument we developed focused first on obvious differences in content: implicit racial biases may encode different stereotypes. Compare weapons bias (Payne 2006), which encodes stereotypes about black males and weapons; and the biases operative in Dovidio & Gaertner’s 2000 CV studies, which seem to encode different associations to do with race and competence. Obviously, different implicit biases also target different social identities (race, gender, age, sexuality), each of which may encode different bits of stereotypical information. At the time of writing, research was also emerging that suggested functional differences: different implicit biases might predict different behavioural tendencies (Amodio & Devine 2006) and may stand in different relationships to explicit attitudes and beliefs (Devine et al 2002). All these differences are masked by talk of ‘implicit bias’ simpliciter.

Why is it important to acknowledge these differences in content and function? In addition to the different answers that may be generated to philosophical questions, one crucial thought is that these differences may have import for mitigation strategies. If different biases underpin different behavioural tendencies, then bias mitigation strategies had better be targeted at the relevant underlying implicit bias. For example, consider the kinds of differential micro-behaviours that have been found to inflect interracial interactions, with white individuals manifesting more tension and discomfort when engaging with black interlocutors (Dovidio et al 2002). If such behaviours are not predicted by competence biases of the sort that influence evaluations of CVs, then different strategies will be needed in attempts to avoid discrimination in hiring. Since different implicit biases may inflect both evaluations of application materials, and interactions that occur as part of the hiring process (e.g. micro-behaviours may affect interactions in interviews) attempts to address implicit biases will have to target both.

At the time of writing this chapter, questions were also being raised about whether these functional differences amongst implicit biases were due to different underlying processes involved in implicit biases. Amodio & Devine (2006) propose that we should carve up implicit biases into two kinds – affective (to do with evaluative valence and affective response) and semantic (to do with the meanings associated with different social identities) – each of which is underpinned by different psychological processes. Whilst we appreciated this move to consider the dimensions of heterogeneity amongst implicit biases, we argued against this particular way of distinguishing implicit biases. Associations that Amodio & Devine characterised as affective, because of the evaluative valence of the contents, (good, bad, loyal, evil) clearly have semantic content; associations they characterise as semantic (intellectual, athletic) have evaluative valence and affective content. So even if one endorses this distinction in principle (and there may well be reasons to resist this) the distinction is not cleanly applied by the studies in which this distinction is deployed. So we did not find compelling evidence for the idea that this was a dimension of hetereogeneity amongst implicit biases; much less that it is underpinned by different underlying psychological processes. The heterogeneity we believe there is in content and function is not well captured by this distinction.

Since writing this chapter, much more philosophical work has been generated on how we might characterise implicit biases, and the implications of this for ethical and epistemological questions (see e.g. Levy 2014, Mandelbaum 2014, Holroyd 2016). Our hope is that in formulating these models, one of the desiderata for a successful account is that it can accommodate the sorts of heterogeneity in content and function that the evolving empirical findings generate. This seems to us important for the reasons we articulate in the chapter: to avoid misleading generalisations about the ways implicit biases behave; and in order to target mitigating strategies or interventions at the relevant implicit biases.


Amodio, D. M. and Devine, P. G. (2006). “Stereotyping and evaluation in implicit racebias: Evidence for independent constructs and unique effects on behavior.” Journal of Personality and Social Psychology 91: 652–61.

Devine, P. G. et al. (2002). “The regulation of explicit and implicit race bias: The role of motivations to respond without prejudice.” Journal of Personality and Social Psychology 82(5): 835–48.

Dovidio, J. F., Gaertner, S. E., Kawakami, K., & Hodson, G. (2002). “Why can’t we just get along? Interpersonal biases and interracial distrust”. Cultural Diversity and Ethnic Minority Psychology, 8(2), 88.

Dovidio, J. F. and Gaertner, S. L. (2000). “Aversive racism and selection decisions: 1989 and 1999.” Psychological Science 11: 319–23.

Holroyd J. (forthcoming) “What do we want from a model of implicit cognition?”Proceedings of the Aristotelian Society

Levy, N. (2014). “Neither Fish nor Fowl: Implicit Attitudes as Patchy Endorsements” Noûs.

Mandelbaum, E. (2015). “Attitude, inference, association: On the propositional structure of implicit bias”. Noûs.

Payne, B. K. (2006). “Weapon bias split-second decisions and unintended stereotyping.” Current Directions in Psychological Science, 15(6), 287-291.

Cover image: “Nataraja”, by Bridget Riley (1993)

One comment

  1. David

    I think this debate could benefit through cross-fertilization with broader research about decision heuristics, which share a similar structure. It’s becoming clear that we use the same cognitive systems when we predict the actions of people and the events of the world. Much of our predictive reasoning is based on stereotypes, like: Toyotas are reliable; motorcycles are dangerous; Chinese food is greasy; retrievers are loyal.

    Sometimes, basing de re judgments on such stereotypes is epistemically defensible; sometimes not – for example, when the stereotypes don’t hold, or when there are local overriders (a Toyota that was poorly maintained).

    When we base de re judgments about people on stereotypes, surely there are cases when doing so is epistemically defensible. However, this fact would not automatically make the de re judgments ethically defensible. I do wonder, though, if it would be partially exculpatory. I think it’s important to address the interplay between epistemic normativity and ethical normativity, because both are clearly involved in stereotype-based reasoning. Prima facie, the former should have something to do with the latter.

Comments are closed.

Back to Top