AI generates covertly racist decisions about people based on their dialect.

Hundreds of millions of people now interact with language models, with uses ranging from help with writing1,2 to informing hiring decisions3. However, these language models are known to perpetuate systematic racial prejudices, making their judgements biased in problematic ways about groups such as African Americans4-7. Although previous research has focused on overt racism in language models, social scientists have argued that racism with a more subtle character has developed over time, particularly in the United States after the civil rights movement8,9. It is unknown whether this covert racism manifests in language models. Here, we demonstrate that language models embody covert racism in the form of dialect prejudice, exhibiting raciolinguistic stereotypes about speakers of African American English (AAE) that are more negative than any human stereotypes about African Americans ever experimentally recorded. By contrast, the language models' overt stereotypes about African Americans are more positive. Dialect prejudice has the potential for harmful consequences: language models are more likely to suggest that speakers of AAE be assigned less-prestigious jobs, be convicted of crimes and be sentenced to death. Finally, we show that current practices of alleviating racial bias in language models, such as human preference alignment, exacerbate the discrepancy between covert and overt stereotypes, by superficially obscuring the racism that language models maintain on a deeper level. Our findings have far-reaching implications for the fair and safe use of language technology.

Discriminatory attitudes against unvaccinated people during the pandemic.

During the COVID-19 pandemic, sizeable groups of unvaccinated people persist even in countries with high vaccine access1. As a consequence, vaccination became a controversial subject of debate and even protest2. Here we assess whether people express discriminatory attitudes in the form of negative affectivity, stereotypes and exclusionary attitudes in family and political settings across groups defined by COVID-19 vaccination status. We quantify discriminatory attitudes between vaccinated and unvaccinated citizens in 21 countries, covering a diverse set of cultures across the world. Across three conjoined experimental studies (n = 15,233), we demonstrate that vaccinated people express discriminatory attitudes towards unvaccinated individuals at a level as high as discriminatory attitudes that are commonly aimed at immigrant and minority populations3-5. By contrast, there is an absence of evidence that unvaccinated individuals display discriminatory attitudes towards vaccinated people, except for the presence of negative affectivity in Germany and the USA. We find evidence in support of discriminatory attitudes against unvaccinated individuals in all countries except for Hungary and Romania, and find that discriminatory attitudes are more strongly expressed in cultures with stronger cooperative norms. Previous research on the psychology of cooperation has shown that individuals react negatively against perceived 'free-riders'6,7, including in the domain of vaccinations8,9. Consistent with this, we find that contributors to the public good of epidemic control (that is, vaccinated individuals) react with discriminatory attitudes towards perceived free-riders (that is, unvaccinated individuals). National leaders and vaccinated members of the public appealed to moral obligations to increase COVID-19 vaccine uptake10,11, but our findings suggest that discriminatory attitudes-including support for the removal of fundamental rights-simultaneously emerged.

Brain-phenotype models fail for individuals who defy sample stereotypes.

Individual differences in brain functional organization track a range of traits, symptoms and behaviours1-12. So far, work modelling linear brain-phenotype relationships has assumed that a single such relationship generalizes across all individuals, but models do not work equally well in all participants13,14. A better understanding of in whom models fail and why is crucial to revealing robust, useful and unbiased brain-phenotype relationships. To this end, here we related brain activity to phenotype using predictive models-trained and tested on independent data to ensure generalizability15-and examined model failure. We applied this data-driven approach to a range of neurocognitive measures in a new, clinically and demographically heterogeneous dataset, with the results replicated in two independent, publicly available datasets16,17. Across all three datasets, we find that models reflect not unitary cognitive constructs, but rather neurocognitive scores intertwined with sociodemographic and clinical covariates; that is, models reflect stereotypical profiles, and fail when applied to individuals who defy them. Model failure is reliable, phenotype specific and generalizable across datasets. Together, these results highlight the pitfalls of a one-size-fits-all modelling approach and the effect of biased phenotypic measures18-20 on the interpretation and utility of resulting brain-phenotype models. We present a framework to address these issues so that such models may reveal the neural circuits that underlie specific phenotypes and ultimately identify individualized neural targets for clinical intervention.

Overcoming gender bias in STEM.

Despite prevalent diversity and inclusion programs in STEM, gender biases and stereotypes persist across educational and professional settings. Recognizing this enduring bias is crucial for achieving transformative change on gender equity and can help orient policy toward more effective strategies to address ongoing disparities.

Monitoring hiring discrimination through online recruitment platforms.

Women (compared to men) and individuals from minority ethnic groups (compared to the majority group) face unfavourable labour market outcomes in many economies1,2, but the extent to which discrimination is responsible for these effects, and the channels through which they occur, remain unclear3,4. Although correspondence tests5-in which researchers send fictitious CVs that are identical except for the randomized minority trait to be tested (for example, names that are deemed to sound 'Black' versus those deemed to sound 'white')-are an increasingly popular method to quantify discrimination in hiring practices6,7, they can usually consider only a few applicant characteristics in select occupations at a particular point in time. To overcome these limitations, here we develop an approach to investigate hiring discrimination that combines tracking of the search behaviour of recruiters on employment websites and supervised machine learning to control for all relevant jobseeker characteristics that are visible to recruiters. We apply this methodology to the online recruitment platform of the Swiss public employment service and find that rates of contact by recruiters are 4-19% lower for individuals from immigrant and minority ethnic groups, depending on their country of origin, than for citizens from the majority group. Women experience a penalty of 7% in professions that are dominated by men, and the opposite pattern emerges for men in professions that are dominated by women. We find no evidence that recruiters spend less time evaluating the profiles of individuals from minority ethnic groups. Our methodology provides a widely applicable, non-intrusive and cost-efficient tool that researchers and policy-makers can use to continuously monitor hiring discrimination, to identify some of the drivers of discrimination and to inform approaches to counter it.

Can names shape facial appearance?

Our given name is a social tag associated with us early in life. This study investigates the possibility of a self-fulfilling prophecy effect wherein individuals' facial appearance develops over time to resemble the social stereotypes associated with given names. Leveraging the face-name matching effect, which demonstrates an ability to match adults' names to their faces, we hypothesized that individuals would resemble their social stereotype (name) in adulthood but not in childhood. To test this hypothesis, children and adults were asked to match faces and names of children and adults. Results revealed that both adults and children correctly matched adult faces to their corresponding names, significantly above the chance level. However, when it came to children's faces and names, participants were unable to make accurate associations. Complementing our lab studies, we employed a machine-learning framework to process facial image data and found that facial representations of adults with the same name were more similar to each other than to those of adults with different names. This pattern of similarity was absent among the facial representations of children, thereby strengthening the case for the self-fulfilling prophecy hypothesis. Furthermore, the face-name matching effect was evident for adults but not for children's faces that were artificially aged to resemble adults, supporting the conjectured role of social development in this effect. Together, these findings suggest that even our facial appearance can be influenced by a social factor such as our name, confirming the potent impact of social expectations.

Hiring women into senior leadership positions is associated with a reduction in gender stereotypes in organizational language.

Women continue to be underrepresented in leadership positions. This underrepresentation is at least partly driven by gender stereotypes that associate men, but not women, with achievement-oriented, agentic traits (e.g., assertive and decisive). These stereotypes are expressed and perpetuated in language, with women being described in less agentic terms than men. The present research suggests that appointing women to the top tiers of management can mitigate these deep-rooted stereotypes that are expressed in language. We use natural language processing techniques to analyze over 43,000 documents containing 1.23 billion words, finding that hiring female chief executive officers and board members is associated with changes in organizations' use of language, such that the semantic meaning of being a woman becomes more similar to the semantic meaning of agency. In other words, hiring women into leadership positions helps to associate women with characteristics that are critical for leadership success. Importantly, our findings suggest that changing organizational language through increasing female representation might provide a path for women to break out of the double bind: when female leaders are appointed into positions of power, women are more strongly associated with the positive aspects of agency (e.g., independent and confident) in language but not at the cost of a reduced association with communality (e.g., kind and caring). Taken together, our findings suggest that female representation is not merely an end, but also a means to systemically change insidious gender stereotypes and overcome the trade-off between women being perceived as either competent or likeable.

Historical representations of social groups across 200 years of word embeddings from Google Books.

Using word embeddings from 850 billion words in English-language Google Books, we provide an extensive analysis of historical change and stability in social group representations (stereotypes) across a long timeframe (from 1800 to 1999), for a large number of social group targets (Black, White, Asian, Irish, Hispanic, Native American, Man, Woman, Old, Young, Fat, Thin, Rich, Poor), and their emergent, bottom-up associations with 14,000 words and a subset of 600 traits. The results provide a nuanced picture of change and persistence in stereotypes across 200 y. Change was observed in the top-associated words and traits: Whether analyzing the top 10 or 50 associates, at least 50% of top associates changed across successive decades. Despite this changing content of top-associated words, the average valence (positivity/negativity) of these top stereotypes was generally persistent. Ultimately, through advances in the availability of historical word embeddings, this study offers a comprehensive characterization of both change and persistence in social group representations as revealed through books of the English-speaking world from 1800 to 1999.

Neural representations of others' traits predict social decisions.

To guide social interaction, people often rely on expectations about the traits of other people, based on markers of social group membership (i.e., stereotypes). Although the influence of stereotypes on social behavior is widespread, key questions remain about how traits inferred from social-group membership are instantiated in the brain and incorporated into neural computations that guide social behavior. Here, we show that the human lateral orbitofrontal cortex (OFC) represents the content of stereotypes about members of different social groups in the service of social decision-making. During functional MRI scanning, participants decided how to distribute resources across themselves and members of a variety of social groups in a modified Dictator Game. Behaviorally, we replicated our recent finding that inferences about others' traits, captured by a two-dimensional framework of stereotype content (warmth and competence), had dissociable effects on participants' monetary-allocation choices: recipients' warmth increased participants' aversion to advantageous inequity (i.e., earning more than recipients), and recipients' competence increased participants' aversion to disadvantageous inequity (i.e., earning less than recipients). Neurally, representational similarity analysis revealed that others' traits in the two-dimensional space were represented in the temporoparietal junction and superior temporal sulcus, two regions associated with mentalizing, and in the lateral OFC, known to represent inferred features of a decision context outside the social domain. Critically, only the latter predicted individual choices, suggesting that the effect of stereotypes on behavior is mediated by inference-based decision-making processes in the OFC.

The effect of GLP-1 receptor agonist use on negative evaluations of women with higher and lower body weight.

BACKGROUND: GLP-1 receptor agonists (GLP-1 RAs) have dramatically altered obesity treatment. Media reports suggest that GLP-1 RAs users often report feeling judged for taking a "shortcut" to lose weight, which may be related to negative stereotypes toward people with larger bodies. Media reports also describe negative attitudes about lean people who take GLP-1 RAs to enhance their appearance. The present research used a 2 × 2 experimental design to test the effects of GLP-1 RA use and body size on attitudes and egocentric impressions. SUBJECTS/METHODS: A sample of 357 U.S. adults (Mage = 37.8, SD = 13) were randomly assigned to read about a woman, who either was lean or had obesity, and who lost 15% of her body weight either with diet/exercise or a GLP-1 RA. Participants answered questions measuring endorsement of negative weight-related stereotypes and egocentric attitudes toward the woman, as well as beliefs that she took a shortcut to lose weight and beliefs that biogenetic factors caused her baseline weight. RESULTS: Negative evaluations and egocentric impressions were stronger toward a woman who lost weight with a GLP-1 RA compared to diet/exercise. Losing weight with a GLP-1 RA led to stronger negative evaluations through higher weight loss shortcut beliefs irrespective of body size. Losing weight with a GLP-1 RA also led to higher egocentric impressions through higher shortcut beliefs, and this effect was stronger for a lean woman. Finally, losing weight with a GLP-1 RA led to more negative evaluations through stronger endorsement of biogenetic causal beliefs for a lean woman only. CONCLUSIONS: This timely study provides evidence that people with larger and smaller bodies alike are at-risk for being judged for using GLP-1 RAs due to beliefs that these medications are a shortcut. Findings also demonstrate novel reactions related to egotism when weight loss is achieved with pharmacological interventions. PRE-REGISTRATION AND DATA: osf.io/xme4w.

Reducing Facial Stereotype Bias in Consequential Social Judgments: Intervention Success With White Male Faces.

Initial impressions of others based on facial appearances are often inaccurate yet can lead to dire outcomes. Across four studies, adult participants underwent a counterstereotype training to reduce their reliance on facial appearance in consequential social judgments of White male faces. In Studies 1 and 2, trustworthiness and sentencing judgments among control participants predicted whether real-world inmates were sentenced to death versus life in prison, but these relationships were diminished among trained participants. In Study 3, a sequential priming paradigm demonstrated that the training was able to abolish the relationship between even automatically and implicitly perceived trustworthiness and the inmates' life-or-death sentences. Study 4 extended these results to realistic decision-making, showing that training reduced the impact of facial trustworthiness on sentencing decisions even in the presence of decision-relevant information. Overall, our findings suggest that a counterstereotype intervention can mitigate the potentially harmful effects of relying on facial appearance in consequential social judgments.

Gender-Ambiguous Voices and Social Disfluency.

Recently, gender-ambiguous (nonbinary) voices have been added to voice assistants to combat gender stereotypes and foster inclusion. However, if people react negatively to such voices, these laudable efforts may be counterproductive. In five preregistered studies (N = 3,684 adult participants) we found that people do react negatively, rating products described by narrators with gender-ambiguous voices less favorably than when they are described by clearly male or female narrators. The voices create a feeling of unease, or social disfluency, that affects evaluations of the products being described. These effects are best explained by low familiarity with voices that sound ambiguous. Thus, initial negative reactions can be overcome with more exposure.

Not All Powerful People Are Created Equal: An Examination of Gender and Pathways to Social Hierarchy Through the Lens of Social Cognition.

Across four studies (N = 816 U.S. adults), we uncovered a gender stereotype about dual pathways to social hierarchy: Men were associated with power, and women were associated with status. We detected this pattern both explicitly and implicitly in perceptions of individuals drawn from Forbes magazine's powerful people lists in undergraduate and online samples. We examined social-cognitive implications, including prominent people's degree of recognition by individuals and society, and the formation of men's and women's self-concepts. We found that power (status) ratings predicted greater recognition of men (women) and lesser recognition of women (men). In terms of the self-concept, we found that women internalized the stereotype associating women with status more than power implicitly and explicitly. Although men explicitly reported having less status and more power than women, men implicitly associated the self with status as much as power. No gender differences emerged in the desires for power and status.

The Formation of Negative Attitudes Toward Novel Groups.

People perceive out-groups, minorities, and novel groups more negatively than in-groups, majorities, and familiar groups. Previous research has argued that such intergroup biases may be caused by the order in which people typically encounter social groups. Groups that are relatively novel to perceivers (e.g., out-groups, minorities) are primarily associated with distinct attributes that differentiate them from familiar groups. Because distinct attributes are typically negative, attitudes toward novel groups are negatively biased. Five experiments (N = 2,615 adults) confirmed the generalizability of the novel groups' disadvantage to different aspects of attitude formation (i.e., evaluations, memory, stereotyping), to cases with more than two groups, and to cases in which groups were majority/minority or in-groups/out-groups. Our findings revealed a remarkably robust influence of learning order in the formation of group attitudes, and they imply that people often perceive novel groups more negatively than they actually are.

Stereotype Threat and Gender Bias in Internal Medicine Residency: It is Still Hard to be in Charge.

BACKGROUND: Despite similar numbers of women and men in internal medicine (IM) residency, women face unique challenges. Stereotype threat is hypothesized to contribute to underrepresentation of women in academic leadership, and exploring how it manifests in residency may provide insight into forces that perpetuate gender disparities. OBJECTIVE: To quantify the prevalence of stereotype threat in IM residency and explore experiences contributing to that stereotype threat. DESIGN: We used a mixed methods study design. First, we surveyed IM residents using the Stereotype Vulnerability Scale (SVS) to screen for stereotype threat. Second, we conducted focus groups with women who scored high on the SVS to understand experiences that led to stereotype threat. PARTICIPANTS: The survey was sent to all IM residents at University of California, San Francisco (UCSF), in September-November 2019. Focus groups were conducted at UCSF in Spring 2020. APPROACH: The survey included an adapted version of the SVS. For focus groups, we developed a focus group guide informed by literature on stereotype threat. We used a thematic approach to data analysis. The mixed methods design enabled us to draw metainferences by integrating the two data sources. KEY RESULTS: Survey response rate was 61% (110/181). Women were significantly more likely than men to have a score indicating stereotype threat vulnerability (77% vs 0%, p < 0.001). Four themes from focus groups characterized women's experiences of gender bias and stereotype threat: gender norm tension, microaggressions and sexual harassment, authority questioned, and support and allyship. CONCLUSIONS: Gender-based stereotype threat is highly prevalent among women IM residents. This phenomenon poses a threat to confidence and ability to execute patient care responsibilities, detracting from well-being and professional development. These findings indicate that, despite robust representation of women in IM training, further attention is needed to address gendered experiences and contributors to women's vulnerability to stereotype threat.

Diabetes misconceptions, seriousness, motivation, self-efficacy and stigma: A cross-sectional comparison of eight Australian diabetes communication campaign videos.

AIM: This study examines potential intended (attitudes, motivation and self-efficacy) and unintended (stigmatisation of diabetes) consequences of past Australian National Diabetes Week campaign videos. Further, outcomes are compared by the extent to which participants perceived their allocated video as stigmatising diabetes. METHODS: In this cross-sectional, ten-arm study, participants (adults with or without diabetes; 1:2 ratio) were randomly allocated to view one of eight archival diabetes campaign videos (intervention), or either an active or passive control group. Post-exposure, study-specific scales measured diabetes Misconceptions and Seriousness, General and Diabetes Risk-Reduction Motivation and Self-efficacy, and perceptions of video Stigmatisation of diabetes. Scores were compared by condition (intervention vs. control) and by campaign Stigma (highest vs. lowest tertile score), separately by cohort (with or without diabetes). RESULTS: The sample included n = 1023 without diabetes; and n = 510 with diabetes (79% type 2 diabetes). No significant differences in outcomes were observed between conditions (intervention vs. control), with one exception: a modest effect on General Self-efficacy among those without diabetes only. Those perceiving high campaign Stigma (15%), relative to low Stigma (60%), reported significantly greater diabetes Misconceptions, lower perceived Seriousness and (among those without diabetes only) lower General Motivation but higher Diabetes Risk Reduction Motivation. CONCLUSION: Though limited to a single-exposure, we found little meaningful positive influence of past diabetes campaign videos on diabetes attitudes, behavioural intentions or self-efficacy. Further, campaign videos were perceived as stigmatising by a minority-a potential harmful impact. This novel study has implications for the design, implementation and evaluation of future diabetes campaigns.

Stigma in vitiligo: associated factors and severity strata of the Patient Unique Stigmatization Holistic tool in Dermatology (PUSH-D) score.

BACKGROUND: Vitiligo is the most common cause of skin depigmentation worldwide. Patients with vitiligo may experience stigma and this needs to be addressed. OBJECTIVES: To evaluate stigma in patients with vitiligo, search for associated factors and establish severity strata for the Patient Unique Stigmatization Holistic tool in Dermatology (PUSH-D) for patients with vitiligo. METHODS: We conducted a cross-sectional study in ComPaRe Vitiligo, an e-cohort of adult patients with vitiligo. Stigmatization was assessed using the PUSH-D, a recently validated dermatology-specific stigmatization assessment tool. We conducted univariate and multivariable linear regression to identify patient and disease factors associated with the stigmatization. We used an anchor-based approach to define severity strata for the PUSH-D. RESULTS: In total, 318 patients participated (mean age 49.7 years; 73.9% women). Fitzpatrick skin phototype IV-VI, severe facial involvement (high Self-Assessment Vitiligo Extent Score of the face) and depression (high Patient Health Questionnaire-9 score) were positively -associated with a higher stigmatization score, although this association was weak [r = 0.24 (P < 0.001) and r = 0.30 (P < 0.001), respectively]. PUSH-D cutoff values that best discriminated patients with high and low stigma, as defined by the anchor question, were 13 and 23 (κ = 0.622, 95% confidence interval 0.53-0.71). CONCLUSIONS: Our study is the first to use a skin-specific stigmatization tool to assess stigma in patients with vitiligo. Creating strata helps to better interpret the PUSH-D in daily practice and may facilitate its use in clinical trials.

Vitiligo is an acquired autoimmune condition characterized by well-defined depigmented patches of skin on the body. The condition affects approximately 1% of the world's population and those living with vitiligo have long experienced stigmatization. Despite the fact that previous research has investigated the correlation between stigma and vitiligo using non-specific stigma tools, to our knowledge, no study has specifically assessed stigma in people with vitiligo. This study was carried out among French patients with vitiligo to evaluate both felt and actual stigma using the Patient Unique Stigmatization Holistic tool in Dermatology (PUSH-D), a new skin-specific stigma score. We also looked for correlations between PUSH-D scores and other questionnaires measuring levels of anxiety (GAD-7) and depression (PHQ-9). We found that PHQ-9 scores for depression were significantly positively correlated with PUSH-D scores, although these correlations were weak. When examining which factors were associated with higher stigma, we found that darker skin phototypes and severe facial involvement predicted higher stigma. However, we found that hand involvement did not. Overall, vitiligo is associated with a lot of stigma and it has been shown to be a barrier to employment. Therefore, an objective evaluation of vitiligo is required in order to facilitate access and reimbursement of treatment (including those existing and under development). The findings from this study highlight how further research is needed with more diverse groups of people, to better objectify stigma associated with vitiligo.