Anelis Kaiser Trujillo October 2020
In today’s Internet-dominated world, we come across the terms “gender bias” and “race bias” everywhere. It is not news that Google Translate, Amazon’s facial recognition technology and many other digital systems suffer from these biases. Since the training data used for machine learning is biased, it is obvious that the digital devices themselves will be biased too. And these types of bias will remain in the digital devices until society and science, with much scientific and political effort, erase them. But where were sex/gender and “race” bias before they got into training data and became digital? I use “race” here in quotation marks to highlight the social construction of this category, following neuropsychologist Emily Ngubia Kessé, on whose work I base this blog post.
Long before sex/gender and “race” bias, as notions, were understood and defined in the logics of computer science, they existed and were discussed in other scientific fields. Sex/gender bias existed in medicine, meaning that there was research pointing at the exclusion or neglect of issues related to the health and bodies of women. Sex/gender and “race” bias research existed in psychology, and helped better understand psychological aspects of “racial” injustice and discrimination against socially constructed identities. Also in cognitive psychology, there was research on sex/gender biases: for over 30 years now, the male bias in grammar has consistently been shown to influence our mental representations demonstrating that we do actually NOT think of women when we read or use generic male linguistic forms. Studies in cognitive neuroscience have examined what happens in our brains when we perceive human faces under the influence of sex/gender and “race” bias.
And it is exactly this topic of research, sex/gender and “race” biases in face recognition, that I want to address here. Because before sex/gender and “race” biases became digital, they already existed in our minds and brains – where they still are.
Face recognition is a growing field of research in cognitive research. It examines how people detect facial features, recognise faces and – and this is where sex/gender and “race” bias come in – it examines how humans manage and fail to draw conclusions from what they see and perceive and how people link these (wrong) conclusions to prior social understandings. This social knowledge may be based on generalisations or stereotypical social and cultural knowledge. Of specific importance here is how our brains process these biases and, based on that, whether there is anything we can learn from cognitive neuroscience about sex/gender and “race” biases in order to confront and overcome them and thus to contribute to reduction of “racial” and sex/gender based injustice.
Face Recognition: The Experiments on Sex/Gender Bias and “Race” Bias
Cognitive neuroscience aims to explain how the brain and mind work. One very common approach in this field is the use of experimental stimuli to trigger mental processes mirroring the neurobiology of our “thoughts” when they are “in the making”. Stimuli implemented in experiments evoke mental processes, in which (social) knowledge is represented and processed, for instance knowledge related to sex/gender or ethnicity. In the gallery at the end of this article we see pictures presented in a study. This study is a representative example because it demonstrates the way “race” is color-coded and depicted. In this study, sex/gender and racialised typicality and untypicality were examined based on the presentation of human faces. The chosen imagery raises questions: Why exactly are Black women missing in this representation of faces and at the same time why do white men appear twice in the picture (they stand for sex/gender and “race”)? Needless to say, this is an example from science on what Crenshaw described already almost 30 years ago, namely that social identities like “race”, class or sex/gender “mutually constitute, reinforce, and naturalize one another” (Crenshaw, 1991, p. 302). For this example, it means that the categories overlap or add up or are simply something different than one single identity, with the result that Black women fall out of visualisation and thus of examination. Thus, the examination of these categories would in fact require its own framework of analysis – a framework of “intersectionality” that we do not usually find in such neuroscientific experiments.
The “thinking” in experiments on sex/gender and “race” bias occurs by mentally operating with categories, taking decisions based on generalizations, heuristically dividing persons into in-group/out-group memberships, performing exclusionary cuts based on social common knowledge etc. All these selective processes are done in our minds – and they all have a neurobiological correlate, which means that they can be localised in specific regions of the brain.
Against this backdrop, there is no conceptual or theoretical discussion within this scientific field on how sex/gender and “race” should or could be defined in its experimental settings. Rather, by showing faces to represent “race” or sex/gender bias in order to measure their neurobiological correlates, cognitive neuroscience pretends to know in advance what “race” and sex/gender are. Through this powerful neurobiological pretention, neurocognitive science determines categories of difference and makes “race” and sex/gender exist, as shown in the face stimuli above. However, a solid corpus of work from feminist Science and Technology Studies but also research from within science, for instance from neuroscience (Cubelli & della Sala 2018) or genetics (Yudell et al. 2016), has articulated that there has to be an end to using “race” as a variable in scientific research due to its use being “problematic at best and harmful at worst” (Yudell et al. 2016: 564). Others have postulated that there needs to be an engagement with “race” in neuroscience (Shen 2020), while others still have critically highlighted the urge to challenge the white norm of neurofeminism itself (Kuria 2014, Fitsch et al. 2020). Postcolonial and intersectionality scholars have convincingly shown that, similar to sex/gender and intersected to it, “race” has emerged/originated in a colonial discourse and as such cannot be understood in terms of uncontextualised, universal, categorical logics (Segato 2011). Applying this perspective to research based on color-coded face stimuli reveals both the “racialisation” of face perception in the depicted facial pictures and the hegemonic understanding of Black and white as determined by a US-related framing of “race”. One question arising from such a critical account is how to adapt this research frame to contexts such as Switzerland, where, as Noémi Michel has pointed out, a “raceless racism” prevails (Michel 2015).
Yet, despite the problematic approach of cognitive neuroscience to “race” at the level of stimuli and the (im)possibility of this endeavour of not prescribing “racial” and sex/gender identity as well as of not being hegemonically racist and sexist itself, I think it is important for feminist research to get to know and even to conduct similar neuroscientific research dedicated to the topic of “race” and sex/gender, though informed by interdisciplinary perspectives, for two reasons. First, if we want to make it better, i.e. if we aim to renew science and approach intersectionality, we need a starting point from within the laboratories. Second, I claim that while it is neither “race” nor sex/gender that our brains are perceiving in studies like the one on “race” bias whose pictures are depicted above, there is something happening in the brain that could be of interest to us as empirical feminists. Something is happening in our minds when we perform biases while we perceive and categorise, when we wrongly relate new information to previous (stereotypical) knowledge, and it could be worth exploring these neurobiological mechanisms in the name of feminist affirmative neuroscience.
The Neuroscience of Implicit “Race” Bias in Face Recognition Research
The story of “race” bias in brain imaging began around 2000 with the amygdala, a brain region involved in fear conditioning, emotional evaluation and the analysis of possible dangers, which was found to play a significant role in “racial” evaluation in face recognition (Phelps et al. 2000, Hart et al. 2000). Against this backdrop, later studies examined the impact of a region in the brain said to be “the seat of individual personality and social behaviour”, the frontal cortex, when consciously trying to control bias towards Black face pictures as compared to white ones. And so, one after the other, different parts of the frontal cortex, the dorsolateral prefrontal cortex (Cunningham et al. 2004), the prefrontal cortex (Richeson et al. 2003, Xu et al. 2009, Zuo & Schiebinger 2013), and the anterior cingulate cortex became part of the neurobiological network involved in the processing of “race” bias. Most experimental paradigms made participants recognise Black people as different from white people, thus making face categorisation into a component of face processing. Let us criticise here that the experimental design itself, in that case the mere asking participants to distinguish between Black and white, is an instantiation of othering and facial profiling.
In 2014, an article summarised the state of the art and presented a review on the neurobiology of prejudice and stereotyping, understood to be the relevant cognitive processes of “race” biases. Three neural networks were described: 1) the network for prejudice, 2) a network for stereotyping and a 3) network for the regulation of prejudice and stereotyping (Amodio, 2014). The first network implies an automatic and associative component characterised by amygdala activation. The second network modulates previously learned and stored information of stereotype-related content in the frontal and prefrontal cortex. The last circuit shows activation, among others, in the middle part of the frontal cortex. Although these neurobiological processes reflect how prejudice gets embedded very deeply and in very early processing of neurobiology it is certainly not the same as saying prejudice is biologically “pre-programmed” in our brains.
Doing Feminist Neuroscience
As described above, human cognitive neuroscience, and particularly brain imaging, directed attention to the neurobiology of human categorical thinking about implicit mental biases related to racialised or sex/gender categories. We have learned quite a lot about these underlying neurobiological processes since the 2000s. Now, can we solve the problem of sex/gender and “race” bias by knowing how cognitive bias works neurobiologically? Can we solve the aim of inclusion and equality based on empirical neurobiological data? Certainly not as a whole, but if the decision is to stay within science and stick to interdisciplinary research, then neuroscience can help.
A number of feminist empirical interventions are possible (Bryant et al. 2019). It is possible to train our minds and brains not to react according to our biases, and thus to unlearn such biases, and to examine those processes psychologically, cognitively and neurobiologically.
Interestingly, from the beginning of neurocognitive “race” bias research in the 2000s, parts of the research have focused on intervention, i.e. on prejudice intervention. That means that research aimed to examine how to change the attitude towards “racial” beliefs and investigate where these changes take place in the brain. Mattan et al. (2018) summarised existing brain imaging studies and classified several groups of interventional research. The following three examples of examinations scrutinised the role of amygdala activation. One study showed that it is possible to “counterstereotype” by showing that individual experience modifies cultural evaluations of social groups and reflected in an inconsistent pattern of amygdala activation (Phelps et al. 2000). Another experiment demonstrated that less “intergroup contact”, measured by the presentation of novel faces, can lead to larger activation in the left amygdala as compared to familiar faces (Cloutier et al. 2014). Lastly, when inviting participants to code presented faces by means of individuation, i.e. by forcing participants to think of the person presented in the picture as an individual instead of a representative of a social group, again the amgydala showed a less racialised reaction (Wheeler & Fiske 2005). In sum, these outcomes reveal that we can change the context in which we perceive a target person. This change can be scientifically captured as in- and out-group perception in brain and behaviour, which again reflects the adaptability of our brains to moderate prejudice. Based on similar approaches, Kubota et al. (2012) have conducted research on emotion regulation, controlling and changing “race” preferences. In their article entitled “The neuroscience of “race””, they argue “If good people who intend well act in a manner inconsistent with their own standards of egalitarianism because of the “racial” groups to which ‘the other’ belongs, then the question of change takes on new and urgent meaning” (946).
If we want to involve neuroscientific research to get rid of “race” and sex/gender bias, the message here is twofold. Either we stick to research from within neuroscience meaning we have to deal with the way it has already been framed, namely with research presented, as above, based on more or less clear categorical cuts of “race” and sex/gender. Conceptually, this will ever be only a compromise, because it is impossible to take “race” out of the described experiments without re-filling the emerging “race” slot with something else. As long as we have no other category of use, we will always reify “race” by addressing “race” bias. But at least we focus empirically on the formability of our brains and we can concentrate on new experiments of intervention as depicted above, and on the possibility of change by creating experiments that show neurobiological signs of how biases can be erased out of our minds and brains. Alternatively, we engage with “race” but conceptually disengage from “racialising”. We construct a completely new framework from outside neuroscience and include the intersectionality of sex/gender in the experimental design from the get-go and measure, in the brain or elsewhere, what Roy and Subramaniam (2016, 38) call “the political economies” that “are running” through “the flesh” of minorities or discriminated groups”. This approach signifies a reverse view of cause and effect by saying that socially induced exposure is influenced by human behaviour and physiology and eventually it will affect biological factors, such as cells or (neurobiological) circuits.
Both of these ways, the one from within and the one from outside neuroscience, are possible ways of doing feminist science.
FIGURE 1 | Reproduction from a drawing of the brain in "De humani corporis fabrica libri septem" by the anatomist Andreas Vesalius (Basel, 1543) Source: Wikimedia Commons
FIGURE 2 | Pictures of sex/gender and “race” stimuli (top). The stimuli and brain activation (buttom) depicted above show an example of fMRI research belonging to the field of social recognition/categorisation. The brain pictures depict activation in the back part of the cortex, more precisely, the early visual field (ECV), and in a region called fusiform gyrus (rFG, r for “right”). Excerpts of figures from: Stolier & Freeman (2017).
Amodio (2014): The Neuroscience of Prejudice and Stereotyping. Nature reviews Neuroscience, 15 (10), p. 670-682.
Bryant et al. (2019): Feminist Interventions on the Sex/Gender Question in Neuroimaging Research. S&F Online, 15 (2).
Cloutier et al. (2014): The impact of childhood experience on amygdala response to perceptually familiar black and white faces. Journal of Cognitive Neuroscience, 26 (9), p. 1992-2004
Crenshaw (1991): Mapping the margins: Intersectionality, identity politics and violence against women of color. Stanford Law Review, 43, p. 1241-1299.
Cubelli & della Sala (2018): Taking race out of Neuroscience too. Cortex 99, p. A1-A2.
Cunningham et al. (2004): Separable Neural Components in the Processing of Black and White Faces. Psychological Science, 15 (12), p. 806-813.
Fitsch et al. (2020, in press): Coalition-Making, and the Practice of Feminist STS in the time of COVID-19. Catalyst, feminism, theory and technoscience.
Hart et al. (2000): Differential response in the human amygdala to racial outgroup vs ingroup face stimuli. Neuroreport, 11 (11), p. 2351-2355.
Kubota et al. (2012): The neuroscience of race. Nature Neuroscience 15 (7), p. 940-948.
Kuria (2014): Theorizing race(ism) while NeuroGendering. In: Schmitz & Höppner: Gendered Neurocultures: Feminist and Queer Perspectives on Current Brain Discourses, p. 109-123.
Mattan et al. (2018): The social neuroscience of race-based and status-based prejudice. Current Opinion in Psychology, 24, p. 27-34.
Michel (2015): Sheepology: the postcolonial politics of raceless racism in switzerland. Postcolonial Studies,18 (4), p. 410-426.
Phelps et al. (2000): Performance on Indirect Measures of Race Evaluation Predicts Amygdala Activity. Journal of Cognitive Neuroscience, 12 (5), p. 929-738.
Richeson et al. (2003): An fMRI investigation of the impact of interracial contact on executive function. Nature Neuroscience, 6 (12), p.1323-1328.
Roy and Subramanian (2016): Matter in the Shadows. Feminist New Materialism and the Practices of Colonialism. In: Pitts-Taylor, Victoria: Mattering: Feminism, Science, and Materialism, p. 23-42.
Segato (2011): Género y colonialidad: en busca de claves de lectura y de un vocabulario estratégico descolonial. In: Bidaseca & Vazquez laba: Feminismos y Poscolonialidad. Descolonizando el feminismo desde y en América Latin, pp. 17-48.
Shen (2020): Racial Injustice and Neuroethics: Time for Action. AJOB Neuroscience, 11 (3), p. 212-216.
Stolier & Freeman (2017): A Neural Mechanism of Social Categorization. Journal of Neuroscience, 37 (23), p. 5711-5721.
Wheeler & Fiske (2005): Controlling Racial Prejudice. Social-Cognitive Goals Affect Amygdala and Stereotype Activation. Psychological Science, 16 (1), p. 56-63.
Xu et al. (2009): Do You Feel My Pain? Racial Group Membership Modulates Empathic Neural Responses. The Journal of Neuroscience, 29 (26), p. 8525-8529.
Yudell et al. (2016): Taking race out of human genetics. Science, 352 (6273), p. 564-565.
Zuo & Schiebinger (2013): Design AI so that it’s fair. Nature, 559, p. 324-326.
22 October 2020
Anelis Kaiser Trujillo