Impact Of Such Biases Research Articles

Towards Context-Aware Emotion Recognition Debiasing From a Causal Demystification Perspective via De-Confounded Training.

Understanding emotions from diverse contexts has received widespread attention in computer vision communities. The core philosophy of Context-Aware Emotion Recognition (CAER) is to provide valuable semantic cues for recognizing the emotions of target persons by leveraging rich contextual information. Current approaches invariably focus on designing sophisticated structures to extract perceptually critical representations from contexts. Nevertheless, a long-neglected dilemma is that a severe context bias in existing datasets results in an unbalanced distribution of emotional states among different contexts, causing biased visual representation learning. From a causal demystification perspective, the harmful bias is identified as a confounder that misleads existing models to learn spurious correlations based on likelihood estimation, limiting the models' performance. To address the issue, we embrace causal inference to disentangle the models from the impact of such bias, and formulate the causalities among variables in the CAER task via a customized causal graph. Subsequently, we present a Contextual Causal Intervention Module (CCIM) to de-confound the confounder, which is built upon backdoor adjustment theory to facilitate seeking approximate causal effects during model training. As a plug-and-play component, CCIM can easily integrate with existing approaches and bring significant improvements. Systematic experiments on three datasets demonstrate the effectiveness of our CCIM.

“Hi. I’m Molly, Your Virtual Interviewer!” Exploring the Impact of Race and Gender in AI-Powered Virtual Interview Experiences

The persistent issue of human bias in recruitment processes poses a formidable challenge to achieving equitable hiring practices, particularly when influenced by demographic characteristics such as gender and race of both interviewers and candidates. Asynchronous Video Interviews (AVIs), powered by Artificial Intelligence (AI), have emerged as innovative tools aimed at streamlining the application screening process while potentially mitigating the impact of such biases. These AI-driven platforms present an opportunity to customize the demographic features of virtual interviewers to align with diverse applicant preferences, promising a more objective and fair evaluation. Despite their growing adoption, the implications of virtual interviewer identities on candidate experiences within AVIs remain underexplored. We aim to address this research and empirical gap in this paper. To this end, we carried out a comprehensive between-subjects study involving 218 participants across six distinct experimental conditions, manipulating the gender and skin color of an AI virtual interviewer agent. Our empirical analysis revealed that while the demographic attributes of the agents did not significantly influence the overall experience of interviewees, variations in the interviewees' demographics, significantly altered their perception of the AVI process. Further, we uncovered that the mediating roles of Social Presence and Perception of the virtual interviewer critically affect interviewees' Perceptions of Fairness (+), Privacy (-), and Impression management (+).

Mapping cognitive biases in multidisciplinary team (MDT) decision-making for cancer care in Scotland: a cognitive ethnography study protocol

IntroductionThe efficiency of multidisciplinary teams (MDTs) in cancer care hinges on facilitating clinicians’ cognitive processes as they navigate complex and uncertain judgements during treatment planning. When systems and workflows are...

Assessing the potential of GPT-4 to perpetuate racial and gender biases in health care: a model evaluation study

Large language models (LLMs) such as GPT-4 hold great promise as transformative tools in health care, ranging from automating administrative tasks to augmenting clinical decision making. However, these models also pose a danger of perpetuating biases and delivering incorrect medical diagnoses, which can have a direct, harmful impact on medical care. We aimed to assess whether GPT-4 encodes racial and gender biases that impact its use in health care. Using the Azure OpenAI application interface, this model evaluation study tested whether GPT-4 encodes racial and gender biases and examined the impact of such biases on four potential applications of LLMs in the clinical domain-namely, medical education, diagnostic reasoning, clinical plan generation, and subjective patient assessment. We conducted experiments with prompts designed to resemble typical use of GPT-4 within clinical and medical education applications. We used clinical vignettes from NEJM Healer and from published research on implicit bias in health care. GPT-4 estimates of the demographic distribution of medical conditions were compared with true US prevalence estimates. Differential diagnosis and treatment planning were evaluated across demographic groups using standard statistical tests for significance between groups. We found that GPT-4 did not appropriately model the demographic diversity of medical conditions, consistently producing clinical vignettes that stereotype demographic presentations. The differential diagnoses created by GPT-4 for standardised clinical vignettes were more likely to include diagnoses that stereotype certain races, ethnicities, and genders. Assessment and plans created by the model showed significant association between demographic attributes and recommendations for more expensive procedures as well as differences in patient perception. Our findings highlight the urgent need for comprehensive and transparent bias assessments of LLM tools such as GPT-4 for intended use cases before they are integrated into clinical care. We discuss the potential sources of these biases and potential mitigation strategies before clinical implementation. Priscilla Chan and Mark Zuckerberg.

Impact of Behavioural Biases on Investment Performance: A Comparative Analysis of Investors from India, USA and UK

This study presents evidence towards the existence of behavioural biases while analysing the impact of such biases on the investors’ investment performance in India, the USA, and the UK. Further, the study aims to investigate the moderating role of financial literacy on the relationship between behavioural biases and investment performance. It reveals that behavioural biases like affect heuristic, herding and status quo positively impact the investors’ investment performance amongst Indian investors whereas loss aversion, overconfidence, availability, and representativeness biases positively impact the investors’ investment performance amongst UK investors; and affect heuristic bias amongst USA investors. The findings divulge the negative impact of anchoring bias and mental accounting amongst Indian investors; availability bias, disposition effect, and overconfidence bias amongst USA investors; and affect heuristic and herding amongst UK investors. The study finds that a few of the behavioural biases impact investment performance positively, though less often. Hence investors should be cautious as these biases generally lead to irrational decision-making and might adversely influence the investors’ investment performance in the long run. The study further concludes that financial literacy significantly moderates the relationship between anchoring bias and individuals’ investment performance in India and affect heuristic in the USA. Interestingly, in the UK scenario, no significant moderating impact of financial literacy was found between any of the behavioral biases and investment performance.

Implicit or Unconscious Bias in Diabetes Care.

"Implicit bias," also called "unconscious bias," refers to associations outside of conscious awareness that adversely affect one's perception of a person or group. Awareness of implicit bias has been increasing in the realm of diabetes care. Here, the authors highlight several types of unconscious bias on the part of clinicians and patients, including biases based on race, ethnicity, and obesity. They discuss how these biases can negatively affect patient-centered clinical interactions and diabetes care delivery, and they recommend implementation of evidence-based interventions and other health system policy approaches to reduce the potential impact of such biases in health care settings.

Anxious/Depressive Symptoms Alter the Subjective Perception of Heart Failure Severity in Transthyretin Cardiac Amyloidosis

The subjective perception of cardiac symptom severity is considered a main treatment target in the management of transthyretin-related cardiac amyloidosis (CA), as opposed to objective prognostic markers such as N-terminal pro b-type natriuretic peptide (NT-proBNP), which objectively reflects the severity of heart disease. Nevertheless, anxious and depressive symptoms in patients with CA might affect subjects perceptions of disease, creating a potential gap between objective and subjective parameters. We assess the impact of such bias in consecutive patients with CA. A total of 60 patients aged 62 to 88years with CA were recruited. The level of anxiety and depression was measured by the Hospital Anxiety and Depression Scale and the subjective perception of symptoms severity by the Kansas City Cardiomyopathy Questionnaire (KCCQ). Finally, NT-proBNP plasma levels at rest and glomerular filtration rate were measured. Nearly 1/2 of the patients (48%) reported clinically relevant levels of psychologic symptoms. Higher levels of anxious and depressive symptoms were significantly linked to lower KCCQ scores. Furthermore, the relation between NT-proBNP and KCCQ was significant only when anxious and depressive symptoms were low (β=-0.86, p=0.002; β=-0.86, p=0.002, respectively) and medium (β=-0.49, p=0.004; β=-0.45, p=0.004, respectively) but was otherwise lost. Depression and anxiety in patients with transthyretin-related CA required assessment and management. In conclusion, patients with depression/anxiety have a clear disconnect between their personal assessment and objective measures of cardiac symptoms, with a major influence on the patients' wellbeing and on their subjective response to treatments in clinical trials.

The colonization of Wikipedia: evidence from characteristic editing behaviors of warring camps

PurposeTo add new empirical knowledge to debates about social practices of peer production communities, and to conversations about bias and its implications for democracy. To help identify Wikipedia (WP) articles that are affected by systematic bias and hopefully help alleviate the impact of such bias on the general public, thus helping enhance both traditional (e.g. libraries) and online information services (e.g. Google) in ways that contribute to democracy. This paper aims to discuss the aforementioned objectives.Design/methodology/approachQuantitatively, the authors identify edit-warring camps across many conflict zones of the English language WP, and profile and compare success rates and typologies of camp edits in the corresponding topic areas. Qualitatively, the authors analyze the edit war between two senior WP editors that resulted in imbalanced and biased articles throughout a topic area for such editorial characteristics through a close critical reading.FindingsThrough a large-scale quantitative study, the authors find that winner-take-all camps exhibit biasing editing behaviors to a much larger extent than the camps they successfully edit-war against, confirming findings of prior small-scale qualitative studies. The authors also confirm the employment of these behaviors and identify other behaviors in the successful silencing of traditional medicinal knowledge on WP by a scientism-biased senior WP editor through close reading.Social implicationsWP sadly does, as previously claimed, appear to be a platform that represents the biased viewpoints of its most stridently opinionated Western white male editors, and routinely misrepresents scholarly work and scientific consensus, the authors find. WP is therefore in dire need of scholarly oversight and decolonization.Originality/valueThe authors independently verify findings from prior personal accounts of highly power-imbalanced fights of scholars against senior editors on WP through a third-party close reading of a much more power balanced edit war between senior WP editors. The authors confirm that these findings generalize well to edit wars across WP, through a large scale quantitative analysis of unbalanced edit wars across a wide range of zones of contention on WP.

A Causal Debiasing Framework for Unsupervised Salient Object Detection

Unsupervised Salient Object Detection (USOD) is a promising yet challenging task that aims to learn a salient object detection model without any ground-truth labels. Self-supervised learning based methods have achieved remarkable success recently and have become the dominant approach in USOD. However, we observed that two distribution biases of salient objects limit further performance improvement of the USOD methods, namely, contrast distribution bias and spatial distribution bias. Concretely, contrast distribution bias is essentially a confounder that makes images with similar high-level semantic contrast and/or low-level visual appearance contrast spuriously dependent, thus forming data-rich contrast clusters and leading the training process biased towards the data-rich contrast clusters in the data. Spatial distribution bias means that the position distribution of all salient objects in a dataset is concentrated on the center of the image plane, which could be harmful to off-center objects prediction. This paper proposes a causal based debiasing framework to disentangle the model from the impact of such biases. Specifically, we use causal intervention to perform de-confounded model training to minimize the contrast distribution bias and propose an image-level weighting strategy that softly weights each image's importance according to the spatial distribution bias map. Extensive experiments on 6 benchmark datasets show that our method significantly outperforms previous unsupervised state-of-the-art methods and even surpasses some of the supervised methods, demonstrating our debiasing framework's effectiveness.

Preferential sampling for bivariate spatial data

Preferential sampling provides a formal modeling specification to capture the effect of bias in a set of sampling locations on inference when a geostatistical model is used to explain observed responses at the sampled locations. In particular, it enables modification of spatial prediction adjusted for the bias. Its original presentation in the literature addressed assessment of the presence of such sampling bias while follow on work focused on regression specification to improve spatial interpolation under such bias. All of the work in the literature to date considers the case of a univariate response variable at each location, either continuous or modeled through a latent continuous variable. The contribution here is to extend the notion of preferential sampling to the case of bivariate response at each location. This exposes sampling scenarios where both responses are observed at a given location as well as scenarios where, for some locations, only one of the responses is recorded. That is, there may be different sampling bias for one response than for the other. It leads to assessing the impact of such bias on co-kriging. It also exposes the possibility that preferential sampling can bias inference regarding dependence between responses at a location. We develop the idea of bivariate preferential sampling through various model specifications and illustrate the effect of these specifications on prediction and dependence behavior. We do this both through simulation examples as well as with a forestry dataset that provides mean diameter at breast height (MDBH) and trees per hectare (TPH) as the point-referenced bivariate responses.

Proceed, With Caution: Unconscious Bias in Technical Assessment.

Proceed, With Caution: Unconscious Bias in Technical Assessment.

Health Equity in Artificial Intelligence and Primary Care Research: Protocol for a Scoping Review.

BackgroundThough artificial intelligence (AI) has the potential to augment the patient-physician relationship in primary care, bias in intelligent health care systems has the potential to differentially impact vulnerable patient populations.ObjectiveThe purpose of this scoping review is to summarize the extent to which AI systems in primary care examine the inherent bias toward or against vulnerable populations and appraise how these systems have mitigated the impact of such biases during their development.MethodsWe will conduct a search update from an existing scoping review to identify studies on AI and primary care in the following databases: Medline-OVID, Embase, CINAHL, Cochrane Library, Web of Science, Scopus, IEEE Xplore, ACM Digital Library, MathSciNet, AAAI, and arXiv. Two screeners will independently review all abstracts, titles, and full-text articles. The team will extract data using a structured data extraction form and synthesize the results in accordance with PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews) guidelines.ResultsThis review will provide an assessment of the current state of health care equity within AI for primary care. Specifically, we will identify the degree to which vulnerable patients have been included, assess how bias is interpreted and documented, and understand the extent to which harmful biases are addressed. As of October 2020, the scoping review is in the title- and abstract-screening stage. The results are expected to be submitted for publication in fall 2021.ConclusionsAI applications in primary care are becoming an increasingly common tool in health care delivery and in preventative care efforts for underserved populations. This scoping review would potentially show the extent to which studies on AI in primary care employ a health equity lens and take steps to mitigate bias.International Registered Report Identifier (IRRID)PRR1-10.2196/27799

Engineering Impacts of Anonymous Author Code Review: A Field Experiment

Code review is a powerful technique to ensure high quality software and spread knowledge of best coding practices between engineers. Unfortunately, code reviewers may have biases about authors of the code they are reviewing, which can lead to inequitable experiences and outcomes. In principle, anonymous author code review can reduce the impact of such biases by withholding an author's identity from a reviewer. In this paper, to understand the engineering effects of using author anonymous code review in a practical setting, we applied the technique to 5217 code reviews performed by 300 software engineers at Google. Our results suggest that during anonymous author code review, reviewers can frequently guess authors’ identities; that focus is reduced on reviewer-author power dynamics; and that the practice poses a barrier to offline, high-bandwidth conversations. Based on our findings, we recommend that those who choose to implement anonymous author code review should reveal the time zone of the author by default, have a break-the-glass option for revealing author identity, and reveal author identity directly after the review.

On the detectability of mis-modeled biases in the network-derived positioning corrections and their user impact

High-precision single-receiver positioning requires the provision of reliable network-derived corrections. Care must therefore be exercised to continuously check the quality of the corrections and to detect the possible presence of mis-modeled biases in the network data. In network-RTK or its state-space implementation, PPP-RTK, quality control of the solutions is executed in two separate phases: the network component and the user component. Once confidence in the network-derived solutions is declared, a subset of the solutions is sent as corrections to a single-receiver user, thereby allowing the user to separately check the integrity of his network-aided model. In such a two-step integrity monitoring procedure, an intermediate step is missing, the integrity monitoring of the corrections themselves. It is the goal of this contribution to provide a quality control procedure for GNSS parameter solutions at the correction level, and to measure the impact a missed detection bias has on the (ambiguity-resolved) user position. New detection test statistics are derived with which the single-receiver user can check the overall validity of the corrections even before applying them to his data. A small-scale network of receivers is utilized to provide numerical insights into the detectability of mis-modeled biases using the proposed detectors and to analyze the impact of such biases on the user positioning performance.

Safety of psychotropic medicines: contribution from observational evidence.

The risks associated with psychotropic drugs use should be accurately documented at the population level in view of the growing number of persons exposed to these drugs. The strengths of observational studies regarding the identification of drug-related harms mirror the limitations of randomised controlled trials and vice-versa. Observational studies can be carried out in large samples of unselected participants treated in real-life conditions and who may be followed up over long periods. Serious adverse effects undetected during pre-marketing clinical trials may be observed only in post-marketing use, such as metabolic effects of second-generation antipsychotics. Observational studies play a key role in the identification of teratogenic risks, such as those induced by prenatal exposure to anticonvulsants. These studies are the main source of information to investigate the long-term effects of drugs, such as the possible increased risk of dementia in benzodiazepine users. They may also contribute to the accurate assessment at population level of risks overestimated by studies carried out in non-representative samples, such as the risk of congenital heart diseases in babies prenatally exposed to lithium. Owing to the lack of random allocation of drugs, confounding by indication or by disease severity are the major sources of biases in observational studies exploring drug safety. An adverse outcome may be wrongly imputed to drug exposure while it is a symptom/outcome of the disease motivating the decision to prescribe. Such a bias may occur in studies investigating the link between exposure to antidepressants and suicidality. As several methods have been developed to lessen the impact of such biases, pharmaco-epidemiological studies based upon stringent methodological designs should be regarded as a valid approach for assessing psychotropic drug safety.

Anthropogenic range contractions bias species climate change forecasts

Forecasts of species range shifts under climate change most often rely on ecological niche models, in which characterizations of climate suitability are highly contingent on the species range data used. If ranges are far from equilibrium under current environmental conditions, for instance owing to local extinctions in otherwise suitable areas, modelled environmental suitability can be truncated, leading to biased estimates of the effects of climate change. Here we examine the impact of such biases on estimated risks from climate change by comparing models of the distribution of North American mammals based on current ranges with ranges accounting for historical information on species ranges. We find that estimated future diversity, almost everywhere, except in coastal Alaska, is drastically underestimated unless the full historical distribution of the species is included in the models. Consequently forecasts of climate change impacts on biodiversity for many clades are unlikely to be reliable without acknowledging anthropogenic influences on contemporary ranges. Models of the distribution of North American mammals show that estimated future diversity under climate change is drastically underestimated unless the full historical distribution of species is included in the models.

Gender Bias in the Job Market

For millions of workers, online job listings provide the first point of contact to potential employers. As a result, job listings and their word choices can significantly affect the makeup of the responding applicant pool. Here, we study the effects of potentially gender-biased terminology in job listings, and their impact on job applicants, using a large historical corpus of 17 million listings on LinkedIn spanning 10 years. We develop algorithms to detect and quantify gender bias, validate them using external tools, and use them to quantify job listing bias over time. We then perform a user survey over two user populations (N 1=469 , N 2=273 ) to validate our findings and to quantify the end-to-end impact of such bias on applicant decisions. Our findings show gender-bias has decreased significantly over the last 10 years. More surprisingly, we find that impact of gender bias in listings is dwarfed by our respondents' inherent bias towards specific job types.

Biases and Strategic Behaviour in Performance Evaluation: The Case of the FIFA's best soccer player award

AbstractIn this paper, we study biases in performance evaluation by analysing votes for the FIFA Ballon d'Or award for best soccer player, the most prestigious award in the sport. Our findings suggest that ‘similarity’ biases are substantial, with jury members disproportionately voting for candidates from their own country, own national team, own continent and own league team. Further, we show that the impact of such biases on the total number of votes a candidate receives is fairly limited and hence is likely to affect the outcome of this competition only on rare occasions where the difference in quality between the leading candidates is small. Finally, analysing the incidence of ‘strategic voting’, we find jury members who vote for one leading candidate are more, rather than less, likely to also give points to his main competitor, as compared with neutral jury members. We discuss the implications of our findings for the design of awards, elections and performance evaluation systems in general and for the FIFA Ballon d'Or award in particular.

Assessing the impact of selection bias on test decisions in trials with a time-to-event outcome.

If past treatment assignments are unmasked, selection bias may arise even in randomized controlled trials. The impact of such bias can be measured by considering the type I error probability. In case of a normally distributed outcome, there already exists a model accounting for selection bias that permits calculating the corresponding type I error probabilities. To model selection bias for trials with a time‐to‐event outcome, we introduce a new biasing policy for exponentially distributed data. Using this biasing policy, we derive an exact formula to compute type I error probabilities whenever an F‐test is performed and no observations are censored. Two exemplary settings, with and without random censoring, are considered in order to illustrate how our results can be applied to compare distinct randomization procedures with respect to their performance in the presence of selection bias. © 2017 The Authors. Statistics in Medicine Published by John Wiley & Sons Ltd.

Inverse modeling of GOSAT-retrieved ratios of total column CH<sub>4</sub> and CO<sub>2</sub> for 2009 and 2010

Abstract. This study investigates the constraint provided by greenhouse gas measurements from space on surface fluxes. Imperfect knowledge of the light path through the atmosphere, arising from scattering by clouds and aerosols, can create biases in column measurements retrieved from space. To minimize the impact of such biases, ratios of total column retrieved CH4 and CO2 (Xratio) have been used. We apply the ratio inversion method described in Pandey et al. (2015) to retrievals from the Greenhouse Gases Observing SATellite (GOSAT). The ratio inversion method uses the measured Xratio as a weak constraint on CO2 fluxes. In contrast, the more common approach of inverting proxy CH4 retrievals (Frankenberg et al., 2005) prescribes atmospheric CO2 fields and optimizes only CH4 fluxes. The TM5–4DVAR (Tracer Transport Model version 5–variational data assimilation system) inverse modeling system is used to simultaneously optimize the fluxes of CH4 and CO2 for 2009 and 2010. The results are compared to proxy inversions using model-derived CO2 mixing ratios (XCO2model) from CarbonTracker and the Monitoring Atmospheric Composition and Climate (MACC) Reanalysis CO2 product. The performance of the inverse models is evaluated using measurements from three aircraft measurement projects. Xratio and XCO2model are compared with TCCON retrievals to quantify the relative importance of errors in these components of the proxy XCH4 retrieval (XCH4proxy). We find that the retrieval errors in Xratio (mean = 0.61 %) are generally larger than the errors in XCO2model (mean = 0.24 and 0.01 % for CarbonTracker and MACC, respectively). On the annual timescale, the CH4 fluxes from the different satellite inversions are generally in agreement with each other, suggesting that errors in XCO2model do not limit the overall accuracy of the CH4 flux estimates. On the seasonal timescale, however, larger differences are found due to uncertainties in XCO2model, particularly over Australia and in the tropics. The ratio method stays closer to the a priori CH4 flux in these regions, because it is capable of simultaneously adjusting the CO2 fluxes. Over tropical South America, comparison to independent measurements shows that CO2 fields derived from the ratio method are less realistic than those used in the proxy method. However, the CH4 fluxes are more realistic, because the impact of unaccounted systematic uncertainties is more evenly distributed between CO2 and CH4. The ratio inversion estimates an enhanced CO2 release from tropical South America during the dry season of 2010, which is in accordance with the findings of Gatti et al. (2014) and Van der Laan et al. (2015). The performance of the ratio method is encouraging, because despite the added nonlinearity due to the assimilation of Xratio and the significant increase in the degree of freedom by optimizing CO2 fluxes, still consistent results are obtained with respect to other CH4 inversions.