An official website of the United States government
Given significant concerns about fairness and bias in the use of artificial intelligence (AI) and machine learning (ML) for psychological assessment, we provide a conceptual framework for investigating and mitigating machine-learning measurement bias (MLMB) from a psychometric perspective. MLMB is defined as differential functioning of the trained ML model between subgroups. MLMB manifests empirically when a trained ML model produces different predicted score levels for different subgroups (e.g., race, gender) despite them having the same ground-truth levels for the underlying construct of interest (e.g., personality) and/or when the model yields differential predictive accuracies across the subgroups. Because the development of ML models involves both data and algorithms, both biased data and algorithm-training bias are potential sources of MLMB. Data bias can occur in the form of nonequivalence between subgroups in the ground truth, platform-based construct, behavioral expression, and/or feature computing. Algorithm-training bias can occur when algorithms are developed with nonequivalence in the relation between extracted features and ground truth (i.e., algorithm features are differentially used, weighted, or transformed between subgroups). We explain how these potential sources of bias may manifest during ML model development and share initial ideas for mitigating them, including recognizing that new statistical and algorithmic procedures need to be developed. We also discuss how this framework clarifies MLMB but does not reduce the complexity of the issue.
more »
« less
Uncovering Latent Biases in Text: Method and Application to Peer Review
Quantifying systematic disparities in numerical quantities such as employment rates and wages between population subgroups provides compelling evidence for the existence of societal biases. However, biases in the text written for members of different subgroups (such as in recommendation letters for male and non-male candidates), though widely reported anecdotally, remain challenging to quantify. In this work, we introduce a novel framework to quantify bias in text caused by the visibility of subgroup membership indicators. We develop a nonparametric estimation and inference procedure to estimate this bias. We then formalize an identification strategy to causally link the estimated bias to the visibility of subgroup membership indicators, provided observations from time periods both before and after an identity-hiding policy change. We identify an application wherein “ground truth” bias can be inferred to evaluate our framework, instead of relying on synthetic or secondary data. Specifically, we apply our framework to quantify biases in the text of peer reviews from a reputed machine learning conference before and after the conference adopted a double-blind reviewing policy. We show evidence of biases in the review ratings that serves as “ground truth”, and show that our proposed framework accurately detects these biases from the review text without having access to the review ratings.
more »
« less
- Award ID(s):
- 1942124
- PAR ID:
- 10310172
- Date Published:
- Journal Name:
- AAAI
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract Neuropsychiatric disorders pose a high societal cost, but their treatment is hindered by lack of objective outcomes and fidelity metrics. AI technologies and specifically Natural Language Processing (NLP) have emerged as tools to study mental health interventions (MHI) at the level of their constituent conversations. However, NLP’s potential to address clinical and research challenges remains unclear. We therefore conducted a pre-registered systematic review of NLP-MHI studies using PRISMA guidelines (osf.io/s52jh) to evaluate their models, clinical applications, and to identify biases and gaps. Candidate studies (n = 19,756), including peer-reviewed AI conference manuscripts, were collected up to January 2023 through PubMed, PsycINFO, Scopus, Google Scholar, and ArXiv. A total of 102 articles were included to investigate their computational characteristics (NLP algorithms, audio features, machine learning pipelines, outcome metrics), clinical characteristics (clinical ground truths, study samples, clinical focus), and limitations. Results indicate a rapid growth of NLP MHI studies since 2019, characterized by increased sample sizes and use of large language models. Digital health platforms were the largest providers of MHI data. Ground truth for supervised learning models was based on clinician ratings (n = 31), patient self-report (n = 29) and annotations by raters (n = 26). Text-based features contributed more to model accuracy than audio markers. Patients’ clinical presentation (n = 34), response to intervention (n = 11), intervention monitoring (n = 20), providers’ characteristics (n = 12), relational dynamics (n = 14), and data preparation (n = 4) were commonly investigated clinical categories. Limitations of reviewed studies included lack of linguistic diversity, limited reproducibility, and population bias. A research framework is developed and validated (NLPxMHI) to assist computational and clinical researchers in addressing the remaining gaps in applying NLP to MHI, with the goal of improving clinical utility, data access, and fairness.more » « less
-
User-generated product reviews are essential for online platforms like Amazon and Yelp. However, the presence of fake reviews misleads customers. GNN is the state-of-the-art method that detects suspicious reviewers by exploiting the topologies of the graph connecting reviewers, reviews, and products. Nevertheless, the discrepancy in the detection accuracy over different groups of reviewers degrades reviewer engagement and customer trust in the review websites. Unlike the previous belief that the difference between the groups causes unfairness, we study the subgroup structures within the groups that can also cause discrepancies in treating different groups. This paper addresses the challenges of defining, approximating, and utilizing a new subgroup structure for fair spam detection. We first identify subgroup structures in the review graph that lead to discrepant accuracy in the groups. The complex dependencies over the review graph create difficulties in teasing out subgroups hidden within larger groups. We design a model that can be trained to jointly infer the hidden subgroup memberships and exploits the membership for calibrating the detection accuracy across groups. Comprehensive comparisons against baselines on three large Yelp review datasets demonstrate that the subgroup membership can be identified and exploited for group fairness.more » « less
-
Bapna, Ravi; Burtch, Gordon (Ed.)Machine learning is commonly used to estimate the heterogeneous treatment effects (HTEs) in randomized experiments. Using large-scale randomized experiments on the Facebook and Criteo platforms, we observe substantial discrepancies between machine learning-based treatment effect estimates and difference-in-means estimates directly from the randomized experiment. This paper provides a two-step framework for practitioners and researchers to diagnose and rectify this discrepancy. We first introduce a diagnostic tool to assess whether bias exists in the model-based estimates from machine learning. If bias exists, we then offer a model-agnostic method to calibrate any HTE estimates to known, unbiased, subgroup difference-in-means estimates, ensuring that the sign and magnitude of the subgroup estimates approximate the model-free benchmarks. This calibration method requires no additional data and can be scaled for large data sets. To highlight potential sources of bias, we theoretically show that this bias can result from regularization and further use synthetic simulation to show biases result from misspecification and high-dimensional features. We demonstrate the efficacy of our calibration method using extensive synthetic simulations and two real-world randomized experiments. We further demonstrate the practical value of this calibration in three typical policy-making settings: a prescriptive, budget-constrained optimization framework; a setting seeking to maximize multiple performance indicators; and a multitreatment uplift modeling setting.more » « less
-
The idealization of a static machine-learned model, trained once and deployed forever, is not practical. As input distributions change over time, the model will not only lose accuracy, any constraints to reduce bias against a protected class may fail to work as intended. Thus, researchers have begun to explore ways to maintain algorithmic fairness over time. One line of work focuses on dynamic learning: retraining after each batch, and the other on robust learning which tries to make algorithms robust against all possible future changes. Dynamic learning seeks to reduce biases soon after they have occurred and robust learning often yields (overly) conservative models. We propose an anticipatory dynamic learning approach for correcting the algorithm to mitigate bias before it occurs. Specifically, we make use of anticipations regarding the relative distributions of population subgroups (e.g., relative ratios of male and female applicants) in the next cycle to identify the right parameters for an importance weighing fairness approach. Results from experiments over multiple real-world datasets suggest that this approach has promise for anticipatory bias correction.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
