Recognition Bias Research Articles

Automated speech recognition bias in personnel selection: The case of automatically scored job interviews.

Organizations, researchers, and software increasingly use automatic speech recognition (ASR) to transcribe speech to text. However, ASR can be less accurate for (i.e., biased against) certain demographic subgroups. This is concerning, given that the machine-learning (ML) models used to automatically score video interviews use ASR transcriptions of interviewee responses as inputs. To address these concerns, we investigate the extent of ASR bias and its effects in automatically scored interviews. Specifically, we compare the accuracy of ASR transcription for English as a second language (ESL) versus non-ESL interviewees, people of color (and Black interviewees separately) versus White interviewees, and male versus female interviewees. Then, we test whether ASR bias causes bias in ML model scores-both in terms of differential convergent correlations (i.e., subgroup differences in correlations between observed and ML scores) and differential means (i.e., shifts in subgroup differences from observed to ML scores). To do so, we apply one human and four ASR transcription methods to two samples of mock video interviews (Ns = 1,014 and 414), and then we train and test models using these different transcripts to score multiple constructs. We observed significant bias in the commercial ASR services across nearly all comparisons, with the magnitude of bias differing across the ASR services. However, the transcription bias did not translate into meaningful measurement bias for the ML interview scores-whether in terms of differential convergent correlations or means. We discuss what these results mean for the nature of bias, fairness, and validity of ML models for scoring verbal open-ended responses. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Supplemental Material for Automated Speech Recognition Bias in Personnel Selection: The Case of Automatically Scored Job Interviews

Supplemental Material for Automated Speech Recognition Bias in Personnel Selection: The Case of Automatically Scored Job Interviews

More women than ever are entering MD-PhD programs. What lies ahead for them?

The earliest MD/PhD programs were small and enrolled mostly men. Here we show that since 2014 there has been a steady increase in the number of women in MD/PhD programs, reaching parity with men in 2023. This change was due to an increase in female applicants, a decrease in male applicants, and an increase in the acceptance rate for women, which had previously been lower than for men. Data from the National MD/PhD Program Outcomes Study show that training duration has been similar for men and women, as have most choices of medical specialties and workplaces. However, women were less likely to have full-time faculty appointments, fewer had NIH grants, and those in the most recent graduation cohort at the time of the survey reported spending less time on research than men. Previously-cited reasons for these differences include disproportionate childcare responsibilities, a paucity of role models, insufficient recognition, and gender bias. Institutions can and should address these obstacles, but training programs can help by preparing their graduates to succeed despite the systemic obstacles. The alternative is a persistent gender gap in the physician-scientist workforce, lost opportunities to benefit from diverse perspectives, and a diminished impact of valuable training resources.

Investigating What Factors Influence Users’ Rating of Harmful Algorithmic Bias and Discrimination

There has been growing recognition of the crucial role users, especially those from marginalized groups, play in uncovering harmful algorithmic biases. However, it remains unclear how users’ identities and experiences might impact their rating of harmful biases. We present an online experiment (N=2,197) examining these factors: demographics, discrimination experiences, and social and technical knowledge. Participants were shown examples of image search results, including ones that previous literature has identified as biased against marginalized racial, gender, or sexual orientation groups. We found participants from marginalized gender or sexual orientation groups were more likely to rate the examples as more severely harmful. Belonging to marginalized races did not have a similar pattern. Additional factors affecting users’ ratings included discrimination experiences, and having friends or family belonging to marginalized demographics. A qualitative analysis offers insights into users' bias recognition, and why they see biases the way they do. We provide guidance for designing future methods to support effective user-driven auditing.

Less can be more: representational vs. stereotypical gender bias in facial expression recognition

AbstractMachine learning models can inherit biases from their training data, leading to discriminatory or inaccurate predictions. This is particularly concerning with the increasing use of large, unsupervised datasets for training foundational models. Traditionally, demographic biases within these datasets have not been well-understood, limiting our ability to understand how they propagate to the models themselves. To address this issue, this paper investigates the propagation of demographic biases from datasets into machine learning models. We focus on the gender demographic component, analyzing two types of bias: representational and stereotypical. For our analysis, we consider the domain of facial expression recognition (FER), a field known to exhibit biases in most popular datasets. We use Affectnet, one of the largest FER datasets, as our baseline for carefully designing and generating subsets that incorporate varying strengths of both representational and stereotypical bias. Subsequently, we train several models on these biased subsets, evaluating their performance on a common test set to assess the propagation of bias into the models’ predictions. Our results show that representational bias has a weaker impact than expected. Models exhibit a good generalization ability even in the absence of one gender in the training dataset. Conversely, stereotypical bias has a significantly stronger impact, primarily concentrated on the biased class, although it can also influence predictions for unbiased classes. These results highlight the need for a bias analysis that differentiates between types of bias, which is crucial for the development of effective bias mitigation strategies.

Emotion-specific recognition biases and how they relate to emotion-specific recognition accuracy, family and child demographic factors, and social behaviour

ABSTRACT The errors young children make when recognising others’ emotions may be systematic over-identification biases and may partially explain the challenges some have socially. These biases and associations may be differential by emotion. In a sample of 871 ethnically and racially diverse preschool-aged children (i.e. 33–68 months; 49% Hispanic/Latine, 52% Children of Colour), emotion recognition was assessed, and scores for accuracy and bias were calculated by emotion (i.e. anger, sad, happy, calm, and fear). Child and family characteristics and teacher-reported social behaviour were also collected. Multilevel structural equation modelling revealed emotion-specific recognition accuracies varied between 36 and 65% whereas biases varied between 4 and 13%. Anger was the strongest bias followed by sad, happy, fear, and calm, in contrast to the pattern for accuracy – happy, sad, angry, fear, and calm. More variance was explained in emotion-specific recognition accuracies by child and family characteristics – 7–38% – than biases – 3–7%. Negatively-valanced emotion recognition biases associated with positively-valanced accuracies, and positively-valued emotion recognition biases associated with negatively-valued accuracies. Biases did not have meaningful associations with social behaviour. This study highlights that children’s emotion recognition errors may partially be systematic, but future studies are needed to understand the underlying cognitive mechanisms.

Implications of Pension and Other Post-Retirement Benefits (OPRB) Recognition Biases on Leverage Ratio Components

Implications of Pension and Other Post-Retirement Benefits (OPRB) Recognition Biases on Leverage Ratio Components

Cultivating Multicultural Dispositions: Assessing the Impact of a Multicultural Education Course on Preservice Teachers at an HBCU

This research examined the impact of a multicultural education course on the cultivation of critical dispositions and cultural competence among future educators at a Historically Black College and University (HBCU). The purpose of this study was to understand how this course affected students' self-awareness, recognition of personal biases, and broadening of worldviews. Using a mixed-methods approach, post-only quantitative data from the Multicultural Teacher Dispositions Scale (MTDS) and qualitative reflections from 23 preservice teachers selected through convenience sampling were combined in the study. The MTDS evaluates three dispositions of multiculturalism in preservice teachers: meekness, social awareness, and advocacy. The results revealed from moderate to strong endorsement of the meekness disposition, which means openness to diverse perspectives and self-reflection. The social awareness was moderately endorsed, with students recognising systemic inequities and valuing diverse viewpoints in education. The advocacy disposition showed mixed results, highlighting areas for growth in challenging institutional biases and promoting inclusivity. The qualitative analysis identified three key themes in students' reflections: embracing discomfort, self-awareness and bias recognition, effective communication as well as cultural sensitivity. Participants reported broadened worldviews, greater appreciation for cultural diversity, and increased confidence in engaging with diverse backgrounds. The study provided empirical evidence for the effectiveness of structured multicultural education courses in enhancing critical dispositions and cultural competence among preservice teachers. It emphasises the importance of incorporating opportunities for self-reflection, exploration of diverse perspectives, and addressing personal biases and discomfort within teacher preparation programs. These findings are crucial for teacher education programs at HBCUs preparing culturally competent educators for increasingly diverse classrooms. The study also suggests a need for on-going research on strategies to cultivate multicultural dispositions and to study the long-term impact of multicultural education courses on teaching practices.

Beyond stigma: Clinician bias in contraceptive counseling to sexual and gender minority youth

ObjectivesSexual and gender minority (SGM) youth experience disparities in sexual and reproductive health; however, little is known about how clinicians engage in contraceptive counseling with this patient population. This study describes pediatric clinician patterns and biases in contraceptive counseling with SGM youth. Study DesignWe conducted 16 in-depth interviews with a convenience sample of clinicians who counsel adolescents on contraception. Participants were recruited and interviewed in-person at the American Academy of Pediatrics National Conference in October 2022. We used codebook thematic analysis. ResultsWhen discussing contraceptive counseling among SGM youth, three major themes emerged: (1) participants’ acceptance of SGM youth identities varied from support to suspicion and rejection; (2) participants’ conceptualizations of their SGM youth patients’ identities circumscribed the scope of the contraceptive care they provided; and (3) participants described using a universal approach to contraceptive counseling that disregarded the relevance of SGM youth identities. For transgender patients, many clinicians focused on menstrual regulation and overlooked potential pregnancy risk. When discussing sexual minority patients, clinicians overemphasized pregnancy prevention and encouraged the use of highly effective contraceptive methods rather than taking a shared decision-making approach to contraceptive care. ConclusionMany clinicians demonstrated bias in approaches to contraceptive care provision to SGM youth patients by holding patients accountable to normative assumptions in transgender medicine and family planning. Training and support for adolescent-facing clinicians in bias recognition and comprehensive contraceptive care are necessary to provide person-centered reproductive health care to SGM youth.

A New Overview of Sex Bias in Fungal Infections.

Fungal infections often disproportionately affect males over females. Since the NIH mandated in 2016 that researchers test their hypotheses in both biological sexes, numerous other fungal infections/colonizations have been found to exhibit sex-specific patterns. These patterns have been observed in various species, including mice, drosophila, cats, and bats, suggesting significant implications for understanding these diseases and developing treatments. Despite the recognition of this sex bias, primary research explaining its underlying causes or mechanisms remains limited. Current evidence suggests that potential causes might be linked to sex hormones, genetic expression, and evolutionary behaviors. This review consolidates recent data on sex bias in fungal infections or colonizations among different species and proposes future research directions to address existing gaps. Thus, this review advances the comprehension of the intricate relationships between biological sex, fungal infections, and broader health implications.

Using JEDI Modules & Interinstitutional Dialogue to Support Implicit Bias Management & Fieldwork Readiness: A Pilot Program

Abstract Date Presented 03/21/24 A study piloted an implicit bias recognition and management curriculum across threeOT degree programs to better understand the instructional methods student perceive should be included when addressing implicit bias in preparation for fieldwork and practice. Primary Author and Speaker: Katherine L. C. Manalang Additional Authors and Speakers: Efekona Nuwere Contributing Authors: Pamela Yvette Lewis-Kipkulei, Kimberly Krebs

The Impact of Canada’s Multiculturalism Policy on the Employment Rates of High-Skilled Immigrants

This study examines the impact of Canada’s multiculturalism policy on the employment rates of high-skilled immigrants in STEM fields from 2000 to 2020. Utilizing a quantitative research design, the study integrates cross-sectional and longitudinal data from Statistics Canada’s Labour Force Survey (LFS) and the Longitudinal Immigration Database (IMDB) to provide a comprehensive analysis. The findings indicate that multiculturalism policies have had a generally positive effect on the employment outcomes of high-skilled immigrants, particularly through initiatives supporting credential recognition, language training, and inclusive hiring practices. However, significant challenges remain, including difficulties in foreign credential recognition, language barriers, and systemic biases in hiring practices. By comparing employment trends and analyzing factors influencing employment outcomes, the study highlights the areas where policy and support measures need to be strengthened to enhance the integration of high-skilled immigrants into Canada’s STEM workforce. The study concludes with recommendations for policymakers and employers to further improve the effectiveness of multiculturalism policies and support the successful employment of high-skilled immigrants.

Surveying Racial Bias in Facial Recognition: Balancing Datasets and Algorithmic Enhancements

Surveying Racial Bias in Facial Recognition: Balancing Datasets and Algorithmic Enhancements

Justice, Equity, Diversity, Inclusion, and Belonging (JEDI-B) in Transition to Practice Programs.

Justice, equity, diversity, inclusion, and belonging (JEDI-B) are essential for creating safe and productive professional environments. Clear definitions of JEDI-B terms are essential for collective understanding and organizational buy-in. Institutional and transition to practice program efforts should focus on the development of inclusive cultures, diverse curriculum, and data-driven evaluation of JEDI-B policies, processes, and practices. Engagement in bias recognition and the unlearning of those biases is also vital. Despite current legislative challenges, the integration of JEDI-B principles remains a vital component in enhancing health care experiences for patients and fostering inclusive workplaces. [J Contin Educ Nurs. 2024;55(6):273-275.].

Phase-dependent word perception emerges from region-specific sensitivity to the statistics of language

Neural oscillations reflect fluctuations in excitability, which biases the percept of ambiguous sensory input. Why this bias occurs is still not fully understood. We hypothesized that neural populations representing likely events are more sensitive, and thereby become active on earlier oscillatory phases, when the ensemble itself is less excitable. Perception of ambiguous input presented during less-excitable phases should therefore be biased toward frequent or predictable stimuli that have lower activation thresholds. Here, we show such a frequency bias in spoken word recognition using psychophysics, magnetoencephalography (MEG), and computational modelling. With MEG, we found a double dissociation, where the phase of oscillations in the superior temporal gyrus and medial temporal gyrus biased word-identification behavior based on phoneme and lexical frequencies, respectively. This finding was reproduced in a computational model. These results demonstrate that oscillations provide a temporal ordering of neural activity based on the sensitivity of separable neural populations.

Understanding the Influence of Single Bias Reduction Strategies on Personal and Systemic Bias Outcomes

Social psychological research has used strategies to increase recognition of and motivation to combat personal or systemic bias, but with little attention to whether single strategies might influence both personal and systemic outcomes. We investigated whether single strategies are effective in both bias domains and potential underlying mechanisms. Across two experiments, non-Black participants were exposed to information concerning (a) their personal racial biases, (b) multiple Black individuals’ discrimination experiences across institutional contexts, or (c) race-unrelated information (control condition). Discrimination experiences exposure (vs. control) increased recognition of systemic bias and motivation to combat both systemic and personal bias (Studies 1 and 2), and we found statistical support for empathy as a mediator (Study 2). In contrast, strategies for highlighting personal bias had weaker effects on personal bias outcomes and no effects on systemic bias outcomes. We discuss theoretical and practical implications of discrimination experiences exposure for combatting systemic and personal bias.

Cultural mobilities and cultural heritage: concepts for an Asia-centric approach

This article sets out an original and exploratory framework for examining emerging concepts of cultural mobilities and heritage with a key focus on the infrastructure and spatialities of cultural mobilities in, of and through Asia—specifically China. To date, the scholarly analysis of mobilities has been dominated by the social sciences in relation to central themes of migration, national borders, crisis and transnational flows of objects and people. This paper seeks to expand the focus in mobilities discourses to humanities, not only to research how infrastructures and spatialities are shaped by culture and heritage but also to analyse cultural mobility through infrastructure and spatialities. We set out epistemic considerations for approaching cultural mobilities through an interdisciplinary lens that seeks to address heritage studies of diverse kinds, from environmental to religious to architectural. By centring Asian epistemes, the paper also challenges recognition and interpretation biases in the humanities and social sciences that continue to privilege Eurocentric hegemonies. Together, the co-authors examine: how Asian cultures operate through material and non-material infrastructures that defy singular location in geopolitical crossings and networks; how culture is mobilised in different ways through infrastructure; the entanglement of cultural heritage with political infrastructure and living practice as well as embedded values. The article discusses China’s Belt and Road Initiative (BRI), yoga as a cultural infrastructure between India and China, and the infrastructure and multiple spatialities of Chinatown as a evolving practice of cultural heritage.

Machine Discriminating: Automated Speech Recognition Biases in Refugee Interviews

This study scrutinizes Automated Speech Recognition (ASR) software, a powerful instrument to expedite the translation process, and their unintended bias against Arabic speakers, particularly refugees. We propose that pre-existing biases in ASR training data reflect societal prejudices, leading to orientalist and Islamophobic misrepresentations. We used four ASR tools to transcribe interviews with Arabic-speaking refugee women, employing ideological textual analysis to detect biases. Our findings indicate that ASR algorithms may inadvertently associate Arabic speakers with conflict, war, religion, and narrow down Arab identities to Islam-centric representations. Acknowledging these biases is essential for fostering a more equitable and culturally sensitive technological environment.

EXplainable Artificial Intelligence (XAI) for facilitating recognition of algorithmic bias: An experiment from imposed users’ perspectives

This study explored the potential of eXplainable Artificial Intelligence (XAI) in raising user awareness of algorithmic bias. This study examined the popular “explanation by example” XAI approach, where users receive explanatory examples resembling their input. As this XAI approach allows users to gauge the congruence between these examples and their circumstances, perceived incongruence then evokes perceptions of unfairness and exclusion, prompting users not to put blind trust in the system and raising awareness of algorithmic bias stemming from non-inclusive datasets. The results further highlight the moderating role of users’ prior experience with discrimination.

Sexual Harassment, Abuse, and Discrimination in Obstetrics and Gynecology

Unlike other surgical specialties, obstetrics and gynecology (OB-GYN) has been predominantly female for the last decade. The association of this with gender bias and sexual harassment is not known. To systematically review the prevalence of sexual harassment, bullying, abuse, and discrimination among OB-GYN clinicians and trainees and interventions aimed at reducing harassment in OB-GYN and other surgical specialties. A systematic search of PubMed, Embase, and ClinicalTrials.gov was conducted to identify studies published from inception through June 13, 2023.: For the prevalence of harassment, OB-GYN clinicians and trainees on OB-GYN rotations in all subspecialties in the US or Canada were included. Personal experiences of harassment (sexual harassment, bullying, abuse, and discrimination) by other health care personnel, event reporting, burnout and exit from medicine, fear of retaliation, and related outcomes were included. Interventions across all surgical specialties in any country to decrease incidence of harassment were also evaluated. Abstracts and potentially relevant full-text articles were double screened.: Eligible studies were extracted into standard forms. Risk of bias and certainty of evidence of included research were assessed. A meta-analysis was not performed owing to heterogeneity of outcomes. A total of 10 eligible studies among 5852 participants addressed prevalence and 12 eligible studies among 2906 participants addressed interventions. The prevalence of sexual harassment (range, 250 of 907 physicians [27.6%] to 181 of 255 female gynecologic oncologists [70.9%]), workplace discrimination (range, 142 of 249 gynecologic oncologists [57.0%] to 354 of 527 gynecologic oncologists [67.2%] among women; 138 of 358 gynecologic oncologists among males [38.5%]), and bullying (131 of 248 female gynecologic oncologists [52.8%]) was frequent among OB-GYN respondents. OB-GYN trainees commonly experienced sexual harassment (253 of 366 respondents [69.1%]), which included gender harassment, unwanted sexual attention, and sexual coercion. The proportion of OB-GYN clinicians who reported their sexual harassment to anyone ranged from 21 of 250 AAGL (formerly, the American Association of Gynecologic Laparoscopists) members (8.4%) to 32 of 256 gynecologic oncologists (12.5%) compared with 32.6% of OB-GYN trainees. Mistreatment during their OB-GYN rotation was indicated by 168 of 668 medical students surveyed (25.1%). Perpetrators of harassment included physicians (30.1%), other trainees (13.1%), and operating room staff (7.7%). Various interventions were used and studied, which were associated with improved recognition of bias and reporting (eg, implementation of a video- and discussion-based mistreatment program during a surgery clerkship was associated with a decrease in medical student mistreatment reports from 14 reports in previous year to 9 reports in the first year and 4 in the second year after implementation). However, no significant decrease in the frequency of sexual harassment was found with any intervention. This study found high rates of harassment behaviors within OB-GYN. Interventions to limit these behaviors were not adequately studied, were limited mostly to medical students, and typically did not specifically address sexual or other forms of harassment.