Auditing black-box models for indirect influence
Open accessThe DOI matches its Crossref record; 37 of the 39 references it cites were found in a registry or an index.
Abstract
Data-trained predictive models see widespread use, but for the most part they are used as black boxes which output a prediction or score. It is therefore hard to acquire a deeper understanding of model behavior and in particular how different features influence the model prediction. This is important when interpreting the behavior of complex models or asserting that certain problematic attributes (such as race or gender) are not unduly influencing decisions. In this paper, we present a technique for auditing black-box models, which lets us study the extent to which existing models take advantage of particular features in the data set, without knowing how the models work. Our work focuses on the problem of indirect influence : how some features might indirectly influence outcomes via other, related features. As a result, we can find attribute influences even in cases where, upon further direct examination of the model, the attribute is not referred to by the model at all. Our approach does not require the black-box model to be retrained. This is important if, for example, the model is only accessible via an API, and contrasts our work with other methods that investigate feature influence such as feature selection. We present experimental evidence for the effectiveness of our procedure using a variety of publicly available data sets and models. We also validate our procedure using techniques from interpretable learning and feature selection, as well as against other black-box auditing procedures. To further demonstrate the effectiveness of this technique, we use it to audit a black-box recidivism prediction algorithm.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“These methods can also treat previously trained ML models as a black-box [2,70].”
From Fairness in machine learning: A survey (2024) · cites this paper as [2]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“The AI systems auditing literature is also multidisciplinary in that it harbours contributions from many academic disciplines, including computer science (Adler et al., 2018;Kearns et al., 2018), systems engineering (Dennis et al., 2016;Leveson, 2011), law (Laux et al., 2021;Selbst, 2021), media and communication studies (Bandy & Diakopoulos, 2019;Sandvig et al., 2014), social science (Metaxa et”
From Auditing of AI: Legal, Ethical and Technical Approaches (2023) · cites this paper as Adler et al., 2018
The cited source says what the claim attributes to it.
In the citing paper
“Speciically, such approaches explore how partially blinding features (sensitive or otherwise) afect model performance.”
From Fairness in machine learning: A survey (2024) · cites this paper as [2]
The cited source says what the claim attributes to it.
In the citing paper
“Blinding (or partial blinding) has also been used as a fairness audit mechanism [2,82,143].”
From Fairness in machine learning: A survey (2024) · cites this paper as [2]
The check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“While there are some readily available sets of declared sensitive variables, there are relatively few works that actively seek to determine whether other variables or rare (i.e., minority) combinations should be protected or not.”
From Fairness in machine learning: A survey (2024) · cites this paper as [2]
Experimental — accuracy not yet measured
2 more claims about this paper
Create a free account to see every claim other papers make about this one.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Breiman, L. (2001) · Machine Learning · doi:10.1023/a:1010933404324 · Cited by 133,538 · cited 4×
“In addition to early work by Breiman [4] and the recent works by Henelius et al. [13] and Datta et al. [8], a number of other works have looked at black-box auditing, primarily for direct influence [9], [27], [25], [24].”
BackgroundResultInfluential(1992) · Cited by 23,431
"Why Should I Trust You?": Explaining the Predictions of Any Classifier
Ribeiro, M. T., Singh, S., & Guestrin, C. (2016) · Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/2939672.2939778 · Cited by 17,555
Visualizing and understanding convolutional networks
Zeiler, M. D., & Fergus, R. (2014) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-10590-1_53 · Cited by 15,839 · cited 1×
“visualizing the behavior of neurons and inputs have also been studied [28], [14], [16].”
American Apartheid: Segregation and the Making of the Underclass.
Fernandez, R. M., Massey, D. S., & Denton, N. A. (1993) · Contemporary Sociology · doi:10.2307/2074502 · Cited by 6,193
A survey on feature selection methods
Chandrashekar, G., & Sahin, F. (2014) · Computers & Electrical Engineering · doi:10.1016/j.compeleceng.2013.11.024 · Cited by 5,189 · cited 1×
“From a technical standpoint (see [7]), the wrapper approach to feature selection (in which features are evaluated based on the quality of the resulting prediction) is most related to our work.”
BackgroundMotwani, R., & Raghavan, P. (1995) · Cambridge University Press eBooks · doi:10.1017/cbo9780511814075 · Cited by 5,123
Bias in random forest variable importance measures: Illustrations, sources and a solution
Strobl, C., Boulesteix, A.-L., Zeileis, A., & Hothorn, T. (2007) · BMC Bioinformatics · doi:10.1186/1471-2105-8-25 · Cited by 3,688
Conditional variable importance for random forests
Strobl, C., Boulesteix, A.-L., Kneib, T., Augustin, T., & Zeileis, A. (2008) · BMC Bioinformatics · doi:10.1186/1471-2105-9-307 · Cited by 3,254
Privacy-preserving data mining
Agrawal, R., & Srikant, R. (2000) · ACM SIGMOD Record · doi:10.1145/335191.335438 · Cited by 2,961
Fair prediction with disparate impact: a study of bias in recidivism prediction instruments
Chouldechova, A. (2017) · Big Data · doi:10.1089/big.2016.0047 · Cited by 2,159
Buciluǎ, C., Caruana, R., & Niculescu-Mizil, A. (2006) · Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining · doi:10.1145/1150402.1150464 · Cited by 2,068 · cited 1×
“(This idea is similar to model compression, but without the need to find new test data [5].”
BackgroundCertifying and Removing Disparate Impact
Feldman, M., Friedler, S. A., Moeller, J., Scheidegger, C., & Venkatasubramanian, S. (2015) · Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/2783258.2783311 · Cited by 1,827 · cited 7×
“The work of [13] assumes that X \Xi is used to train a function f : X \Xi → Ŷ so that Ŷ can be guaranteed to be -obscured with respect to Xi .”
BackgroundMethodResultInfluentialSelf-citationMachine-learning-assisted materials discovery using failed experiments
Raccuglia, P., Elbert, K. C., Adler, P. D. F., Falk, C., Wenny, M. B., Mollo, A., Zeller, M., Friedler, S. A., Schrier, J., & Norquist, A. J. (2016) · Nature · doi:10.1038/nature17439 · Cited by 1,727 · cited 4×
“The top ranked decision tree features additionally include the averagemolecular polarizability for the organic components, whichwas previously hypothesized as important to synthesis of templated vanadium selenites explored via this data set [24].”
BackgroundResultInfluentialSelf-citationA metric for distributions with applications to image databases
Rubner, Y., Tomasi, C., & Guibas, L. J. (2002) · Sixth International Conference on Computer Vision (IEEE Cat. No.98CH36271) · doi:10.1109/iccv.1998.710701 · Cited by 1,726 · cited 1×
“distribution given by A, then O is maximally obscured, but W also minimally changes, in that A also minimizes the function ∑ x∈O d(Wx, A) where d(·, ·) was the earthmover distance [23] between the distributions using `2 as the base metric.”
MethodPrivacy-preserving data mining
Agrawal, R., & Srikant, R. (2000) · Proceedings of the 2000 ACM SIGMOD international conference on Management of data · doi:10.1145/342009.335438 · Cited by 1,719
Building high-level features using large scale unsupervised learning
Le, Q. V. (2013) · 2013 IEEE International Conference on Acoustics, Speech and Signal Processing · doi:10.1109/icassp.2013.6639343 · Cited by 1,695 · cited 1×
“For neural networks, various approaches based on visualizing the behavior of neurons and inputs have also been studied [28], [14], [16].”
MethodOn the histogram as a density estimator:L 2 theory
Freedman, D., & Diaconis, P. (1981) · Zeitschrift f�r Wahrscheinlichkeitstheorie und Verwandte Gebiete · doi:10.1007/bf01025868 · Cited by 1,601 · cited 1×
“Bins are chosen using the Freedman-Diaconis rule for choosing histogram bin sizes [11].”
MethodClassification by pairwise coupling
Hastie, T., & Tibshirani, R. (1998) · The Annals of Statistics · doi:10.1214/aos/1028144844 · Cited by 1,299 · cited 1×
“We built two models that are notoriously opaque to simple examination; SVMs6 [12] and feedforward neural networks (FNNs).”
MethodCorrelation-based Feature Subset Selection for Machine Learning
Hall, M. A. (1998) · Cited by 1,065
Inherent Trade-Offs in the Fair Determination of Risk Scores
Kleinberg, J., Mullainathan, S., & Raghavan, M. (2016) · arXiv · doi:10.48550/arxiv.1609.05807 · Cited by 941 · cited 1×
“This work has spurred an interesting discussion about definitions of fairness and the impossibility of achieving all types of fairness when the base rates of arrest (how recidivism is measured) are different for different demographic groups [9,19].”
BackgroundLeakage in data mining: Formulation, detection, and avoidance
Kaufman, S., Rosset, S., Perlich, C., & Stitelman, O. (2012) · ACM Transactions on Knowledge Discovery from Data · doi:10.1145/2382577.2382579 · Cited by 791
Datta, A., Sen, S., & Zick, Y. (2016) · 2016 IEEE Symposium on Security and Privacy (SP) · doi:10.1109/sp.2016.42 · Cited by 728 · cited 3×
“Datta et al. [8] provide a generalization of this idea, linking it to game-theoretic notions of influence and showing that different choices of probability spaces and random variables yield a number of different interesting auditing measures.”
BackgroundResultState of the Art: Automated Black-Box Web Application Vulnerability Testing
Bau, J., Bursztein, E., Gupta, D., & Mitchell, J. (2010) · 2010 IEEE Symposium on Security and Privacy · doi:10.1109/sp.2010.27 · Cited by 320
A multidisciplinary survey on discrimination analysis
Romei, A., & Ruggieri, S. (2014) · The Knowledge Engineering Review · doi:10.1017/s0269888913000039 · Cited by 288 · cited 1×
“Finally, we note that a compelling application for blackbox audits of indirect influence includes considerations of algorithmic fairness [22].”
BackgroundAuditing Black-Box Models for Indirect Influence
Adler, P., Falk, C., Friedler, S. A., Rybeck, G., Scheidegger, C., Smith, B., & Venkatasubramanian, S. (2016) · 2016 IEEE 16th International Conference on Data Mining (ICDM) · doi:10.1109/icdm.2016.0011 · Cited by 280
Self-citationAlgorithmic Accountability Reporting: On the Investigation of Black Boxes
Diakopoulos, N. A. (2014) · Columbia Academic Commons (Columbia University) · doi:10.7916/d8zk5tw2 · Cited by 221
Interpretable classification models for recidivism prediction
Zeng, J., Ustun, B., & Rudin, C. (2017) · Journal of the Royal Statistical Society Series A: Statistics in Society · doi:10.1111/rssa.12227 · Cited by 220 · cited 1×
“For this work, we processed the data additionally following the description of the processing performed in a previous study on interpretable models [29].”
MethodBuilding high-level features using large scale unsupervised learning
Le, Q. V., Ranzato, M., Monga, R., Devin, M., Chen, K., Corrado, G. S., Dean, J., & Ng, A. Y. (2011) · arXiv (Cornell University) · doi:10.48550/arxiv.1112.6209 · Cited by 163 · cited 1×
“For neural networks, various approaches based on visualizing the behavior of neurons and inputs have also been studied [28], [14], [16].”
MethodInherent Trade-Offs in Algorithmic Fairness
Kleinberg, J. (2018) · Abstracts of the 2018 ACM International Conference on Measurement and Modeling of Computer Systems · doi:10.1145/3219617.3219634 · Cited by 125
A peek into the black box: exploring classifiers by randomization
Henelius, A., Puolamäki, K., Boström, H., Asker, L., & Papapetrou, P. (2014) · Data Mining and Knowledge Discovery · doi:10.1007/s10618-014-0368-8 · Cited by 98 · cited 3×
“Henelius et al. [13] propose looking at variable sets and their influence by studying the consistency of a black-box predictor when compared to a random permutation of a set.”
BackgroundMethodResultLearning-based Rule-Extraction from Support Vector Machines
Barakat, N., & Diederich, J. (2004) · Cited by 60
Understanding Where Your Classifier Does (Not) Work -- The SCaPE Model Class for EMM
Duivesteijn, W., & Thaele, J. (2014) · Proceedings · doi:10.1109/icdm.2014.10 · Cited by 32
Supersparse Linear Integer Models for Interpretable Classification
Ustun, B., Tracà, S., & Rudin, C. (2013) · arXiv · doi:10.48550/arxiv.1306.6677 · Cited by 26 · cited 2×
“The classic example of such models are decision trees [19], while recently supersparse linear integer models (SLIMs) have been developed as an alternative interpretable model for classification [26].”
BackgroundUnderstanding classifier errors by examining influential neighbors
Kabra, M., Robie, A., & Branson, K. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7299017 · Cited by 22 · cited 1×
“visualizing the behavior of neurons and inputs have also been studied [28], [14], [16].”
Zacarias, O. P., & Boström, H. (2013) · International Conference on Advances in ICT for Emerging Regions · doi:10.1109/icter.2013.6761181 · Cited by 16 · cited 1×
“[8], a number of other works have looked at black-box auditing, primarily for direct influence [9], [27], [25], [24].”
Background
Research works printed but not linked: 2
- G. Casella and R. L. Berger. Statistical Inference. Cengage Learning, 2nd edition, 2001.
- Clark P, Niblett T (1989) The cn2 induction algorithm. Mach Learn 3(4):261–283
Cited by
Arrieta, A. B., Díaz-Rodríguez, N., Del Ser, J., Bennetot, A., Tabik, S., Barbado, A., García, S., Gil-López, S., Molina, D., Benjamins, R., et al. (2019) · arXiv · doi:10.48550/arxiv.1910.10045 · Cited by 11,321
Barredo Arrieta, A., Díaz-Rodríguez, N., Del Ser, J., Bennetot, A., Tabik, S., Barbado, A., Garcia, S., Gil-Lopez, S., Molina, D., Benjamins, R., et al. (2020) · Information Fusion · doi:10.1016/j.inffus.2019.12.012 · Cited by 10,423
The ethics of algorithms: Mapping the debate
Mittelstadt, B. D., Allo, P., Taddeo, M., Wachter, S., & Floridi, L. (2016) · Big Data & Society · doi:10.1177/2053951716679679 · Cited by 4,567
Ali, S., Abuhmed, T., El-Sappagh, S., Muhammad, K., Alonso-Moral, J. M., Confalonieri, R., Guidotti, R., Del Ser, J., Díaz-Rodríguez, N., & Herrera, F. (2023) · Information Fusion · doi:10.1016/j.inffus.2023.101805 · Cited by 1,762
Understanding Black-box Predictions via Influence Functions
Koh, P. W., & Liang, P. (2017) · arXiv · doi:10.48550/arxiv.1703.04730 · Cited by 1,514
“, 2016) or by perturbing the test point to see how the prediction changes (Simonyan et al., 2013; Li et al., 2016b; Datta et al., 2016; Adler et al., 2016).”
Background
Metadata
- Authors
- Adler, P., Falk, C., Friedler, S. A., Nix, T., Rybeck, G., Scheidegger, C., Smith, B., & Venkatasubramanian, S.
- Affiliations
- Haverford College (US)
- University of Arizona (US)
- University of Utah (US)
- Corresponding author
- Friedler, Sorelle A.
- Venue
- Knowledge and Information Systems
- Published
- 2017-10-25
- Volume
- 54
- Issue
- 1
- Pages
- 95-122
- ISSN
- 0219-1377 (ISSN-L)
- 0219-3116
- OpenAlex
- W2567533607
- Cited by
- 280
- References
- 39
- Citations by year
- 2026: 12 · 2025: 19 · 2024: 23 · 2023: 28 · 2022: 41 · 2021: 43 · 2020: 34 · 2019: 42 · 2018: 23 · 2017: 10
- FWCI
- 21.70
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- www.springer.com/tdm (tdm)
- www.springer.com/tdm (vor)
- Locations
- Knowledge and Information Systems (publishedVersion)
- Proceedings (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- National Science Foundation (IIS-1251049, CNS-1302688, IIS-1513651, DMR-1307801, IIS-1633724, IIS-1633387)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data
- Keywords
- indirect influence, feature influence, explainable AI, algorithmic fairness, algorithmic discrimination, interpretable learning, feature selection, recidivism prediction, model interpretability
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.