The DOI matches its Crossref record; 49 of the 50 references it cites were found in a registry or an index, and 1 are not checked yet.
Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead
Open accessAbstract
Black box machine learning models are currently being used for high stakes decision-making throughout society, causing problems throughout healthcare, criminal justice, and in other domains. People have hoped that creating methods for explaining these black box models will alleviate some of these problems, but trying to explain black box models, rather than creating models that are interpretable in the first place, is likely to perpetuate bad practices and can potentially cause catastrophic harm to society. There is a way forward - it is to design models that are inherently interpretable. This manuscript clarifies the chasm between explaining black boxes and using inherently interpretable models, outlines several key reasons why explainable black boxes should be avoided in high-stakes decisions, identifies challenges to interpretable machine learning, and provides several example applications where interpretable models could potentially replace black box models in criminal justice, healthcare, and computer vision.
Overview
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
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 cited source doesn’t support the claim as written.
In the citing paper
“A central challenge in this domain is navigating the well-documented trade-off between model accuracy and inherent interpretability [6].”
From Explainable AI for Clinical Decision Support Systems: Literature Review, Key Gaps, and Research Synthesis (2025) · cites this paper as [6]
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measuredThe cited source doesn’t support the claim as written.
In the citing paper
“However, this remarkable capability comes with a fundamental trade-off: as models become more powerful, they simultaneously become more opaque, creating what researchers term the interpretability–performance paradox [5].”
From Beyond Post hoc Explanations: A Comprehensive Framework for Accountable AI in Medical Imaging Through Transparency, Interpretability, and Explainability (2025) · cites this paper as [5]
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Rudin takes this further, saying that post-hoc explanations “must be wrong”; that they are by definition not completely faithful to the original model and must be less accurate with respect to the primary task.”
From The false hope of current approaches to explainable artificial intelligence in health care (2021) · cites this paper as [35]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“For instance, this work has demonstrated that simple interpretable models (with two and respectively three features) can perform as well as the widely used criminal justice COMPAS algorithm, which instead draws on 137 features.”
From Accountable Artificial Intelligence: Holding Algorithms to Account (2021) · cites this paper as Rudin 2019
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“While Artificial Intelligence (AI) promises significant enhancements for Clinical Decision Support Systems (CDSSs), the opacity of many AI models remains a major barrier to clinical adoption, primarily due to interpretability and trust challenges.”
From Explainable AI for Clinical Decision Support Systems: Literature Review, Key Gaps, and Research Synthesis (2025) · cites this paper as [6]
Experimental — accuracy not yet measured
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
“However, as with cognitive offloading, there is a potential risk that over-reliance on AI for decision-making could diminish human capacity for critical thinking (Rudin, 2019).”
From The impact of digital technology, social media, and artificial intelligence on cognitive functions: a review (2023) · cites this paper as Rudin, 2019
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Many research groups have tackled early prediction of AKI using electronic health records (EHR) data [8–11], but no model so far explains the rationale behind specific predictions despite a clear need for explainable and actionable predictions [6, 12].”
From Machine learning model for early prediction of acute kidney injury (AKI) in pediatric critical care (2021) · cites this paper as [12]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“This goes against the move towards transparency and open science, and makes it hard to uncover the origin of, or gaps in, chatbots' knowledge 10 .”
From ChatGPT: five priorities for research (2023) · cites this paper as [10]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“As users of these systems have grown beyond ML experts and the research community, the need to better interpret and understand them has grown as well, particularly in the context of high-stakes decisions that affect individuals ' health or wellbeing (Lakkaraju et al., 2016; Lipton, 2018; Rudin, 2019).”
From Explainable machine learning for public policy: Use cases, gaps, and research directions (2023) · cites this paper as Rudin, 2019
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Against this background, research into explainability/interpretability 1 of ML models has experienced rapid expansion and innovation in recent years, with a focus on method development.”
From Explainable machine learning for public policy: Use cases, gaps, and research directions (2023) · cites this paper as Rudin, 2019
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“XAI often is a technique used to support methods, and keywords are often not mentioned in the title or body of papers (Rudin (2019)).”
From Explainable artificial intelligence (XAI) in deep learning-based medical image analysis (2022) · cites this paper as Rudin (2019)
The cited source says what the claim attributes to it.
In the citing paper
“Black-box models, despite their high predictive performance, often lack the necessary transparency [76].”
From Deepfake Media Forensics: Status and Future Challenges (2025) · cites this paper as [76]
The cited source says what the claim attributes to it.
In the citing paper
“While “black‐box” algorithms raise complex opaqueness and interpretability challenges, simpler interpretable ML algorithms do not, but have generally also been regarded as less powerful (Guidotti et al. 2019; see however, Rudin 2019).”
From Accountable Artificial Intelligence: Holding Algorithms to Account (2021) · cites this paper as Rudin 2019
The cited source says what the claim attributes to it.
In the citing paper
“Rudin [6], for example, argues that opaque models should be avoided in high-stakes settings such as medicine, where interpretability and trustworthiness are at least as important as predictive accuracy.”
From Explainable AI for Clinical Decision Support Systems: Literature Review, Key Gaps, and Research Synthesis (2025) · cites this paper as [6]
The cited source says what the claim attributes to it.
In the citing paper
“a substantial computer science literature now advocates either post hoc explanation techniques or the outright substitution of interpretable models in high-stakes settings (Rudin, 2019).”
From Auditing the Black Box: Rethinking Audit Assurance and Professional Liability in the Age of Artificial Intelligence (2026) · cites this paper as Rudin, 2019
The cited source says what the claim attributes to it.
In the citing paper
“Interpretability is therefore not a stylistic preference in forensic work but a condition of testimonial usefulness, which is an applied vindication of Rudin's (2019) argument for interpretable models in high-stakes settings.”
From Auditing the Black Box: Rethinking Audit Assurance and Professional Liability in the Age of Artificial Intelligence (2026) · cites this paper as Rudin's (2019)
The cited source says what the claim attributes to it.
In the citing paper
“As perfectly stated in [347], it is not necessarily true that models that are more complex are inherently more accurate.”
From Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI (2020) · cites this paper as [347]
The cited source says what the claim attributes to it.
In the citing paper
“What is more, even when applicable, the legal requirements are ambiguous as to what is required (i.e., “meaningful information about the logic involved”) and risk compromising the value that can be derived from this information: “Because the definition of what constitutes a viable explanation is unclear, even strong regulations such as “right to explanation” can be undermined with less‐than‐satisf”
From Accountable Artificial Intelligence: Holding Algorithms to Account (2021) · cites this paper as Rudin 2019
The cited source says what the claim attributes to it.
In the citing paper
“Inherently therefore, explanation models are not fully faithful representations of the original model, but are rather “approximations” thereof, which necessarily simultaneously reduces their explanatory potential (Rudin 2019).”
From Accountable Artificial Intelligence: Holding Algorithms to Account (2021) · cites this paper as Rudin 2019
The cited source says what the claim attributes to it.
In the citing paper
“First, modern AI systems are often black-box models for which humans cannot see how or why the model made a prediction [54].”
From Improving Human-AI Collaboration With Descriptions of AI Behavior (2023) · cites this paper as [54]
The cited source says what the claim attributes to it.
In the citing paper
“However, post-hoc explanations are approximations to the underlying model rather than exact representations [65].”
From Toward generalizable and interpretable machine learning models in healthcare: Insights from ICU outcome predictions (2026) · cites this paper as [65]
The cited source says what the claim attributes to it.
In the citing paper
“The challenge remains to provide higher-level explainability that aligns with human cognitive processes [Rudin, 2019].”
From Authorship Attribution in the Era of LLMs: Problems, Methodologies, and Challenges (2025) · cites this paper as Rudin, 2019
The cited source says what the claim attributes to it.
In the citing paper
“Inherently interpretable ML models are designed such that an end user could understand its decisionmaking process (Lakkaraju et al., 2016; Rudin, 2019).”
From Explainable machine learning for public policy: Use cases, gaps, and research directions (2023) · cites this paper as Rudin, 2019
The cited source says what the claim attributes to it.
In the citing paper
“Interpretable models often rely on carefully curated representations of data with meaningful input features (Rudin, 2019), often through discretization or binary encoding (Caruana et al., 2015; Lakkaraju et al., 2016; Ustun and Rudin, 2019).”
From Explainable machine learning for public policy: Use cases, gaps, and research directions (2023) · cites this paper as Rudin, 2019
The cited source says what the claim attributes to it.
In the citing paper
“as black-box ML models are often too complex to be explained entirely, post hoc methods typically derive an approximate explanation (Gilpin et al., 2018; Rudin, 2019), which makes ensuring the fidelity of the explanations to the model a key challenge in this work.”
From Explainable machine learning for public policy: Use cases, gaps, and research directions (2023) · cites this paper as Rudin, 2019
The check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“While largerscale evaluations are desirable, prior work (Rudin, 2019; Wiegreffe et al., 2021) has shown that even small samples, when selected for conceptual diversity and difficulty, yield robust insights into model behavior.”
From GIER: Gap-Driven Self-Refinement for Large Language Models (2025) · cites this paper as Rudin, 2019
Experimental — accuracy not yet measuredIn the citing paper
“With regards to methodology, explainability can either be an inherent characteristic of an algorithm or can be approximated by other methods [2].”
From Explainability for artificial intelligence in healthcare: a multidisciplinary perspective (2020) · cites this paper as [2]
In the citing paper
“An important point to note here is that explainability has many facets and, unfortunately, the terminology of explainability is not well defined. Other terms such as interpretability and/or transparency are often used synonymously [2, 3].”
From Explainability for artificial intelligence in healthcare: a multidisciplinary perspective (2020) · cites this paper as [2]
In the citing paper
“Importantly, however, inherent explainability will, in general, be more accurate than methods that only approximate explainability [2].”
From Explainability for artificial intelligence in healthcare: a multidisciplinary perspective (2020) · cites this paper as [2]
In the citing paper
“While the work of Rudin et al. is important to raise attention to the shortcomings of approximating explainability methods, it is likely that some approximating methods, in contrast to the notion of [2], have value given the complex nature of explaining machine learning models.”
From Explainability for artificial intelligence in healthcare: a multidisciplinary perspective (2020) · cites this paper as [2]
In the citing paper
“However, caution would be recommended since ML explainability can be inaccurate or non-meaningful due to the underlying mathematical assumptions13.”
From Mitigating bias in machine learning for medicine (2021) · cites this paper as [13]
In the citing paper
“However, it is important to note that this trade-off does not always occur, in which case a more interpretable model is generally preferred (i.e. explainable modelling) [76].”
In the citing paper
“The prototype networks are comparatively difficult to train than regular deep neural networks because of the scarcity of available techniques for their training [110].”
From Transparency of deep neural networks for medical image analysis: A review of interpretability methods (2022) · cites this paper as [110]
In the citing paper
“As the field is still in its infancy, there is no consensus on the definition of the terms interpretability and explainability in this context, and recently various definitions have been proposed for these terms [9,88,110].”
From Transparency of deep neural networks for medical image analysis: A review of interpretability methods (2022) · cites this paper as [110]
In the citing paper
“This can instill a false sense of confidence in the model which can be detrimental for troubleshooting and model deployment in clinical practice [110].”
From Transparency of deep neural networks for medical image analysis: A review of interpretability methods (2022) · cites this paper as [110]
In the citing paper
“Post-hoc interpretability methods are only approximations and do not accurately depict the correct model behavior and therefore compromise trust in the explanations [5,110].”
From Transparency of deep neural networks for medical image analysis: A review of interpretability methods (2022) · cites this paper as [110]
In the citing paper
“It depends on the researcher's ability to discover the patterns in an interpretable way while at the same time allowing for flexibility to fit the data accurately [110].”
From Transparency of deep neural networks for medical image analysis: A review of interpretability methods (2022) · cites this paper as [110]
In the citing paper
“For tabular data, SHAP values can rank the importance of different clinical factors and show whether a prognosis was driven primarily by tumor size, patient age, or biomarker status [162,163,164,165].”
From Artificial Intelligence in Clinical Oncology: From Productivity Enhancement to Creative Discovery (2025) · cites this paper as [162]
In the citing paper
“To this end, explainability is an often-cited challenge for”
From Artificial Intelligence in Oncology: Current Landscape, Challenges, and Future Directions (2024) · cites this paper as [203]
In the citing paper
“Interpretable models use a small set of features with a low complexity of interaction between them.”
From Explainable AI is Dead, Long Live Explainable AI!: Hypothesis-driven Decision Support using Evaluative AI (2023) · cites this paper as [33]
In the citing paper
“Rudin [33] argues that for high-stakes decisions, people should avoid using black-box models altogether, and instead should use interpretable models.”
From Explainable AI is Dead, Long Live Explainable AI!: Hypothesis-driven Decision Support using Evaluative AI (2023) · cites this paper as [33]
In the citing paper
“Counterfactual explanations are normally considered contrastive by nature and therefore present a source of valuable complementary information to a given automatic prediction [52].”
From A Survey of Contrastive and Counterfactual Explanation Generation Methods for Explainable Artificial Intelligence (2021) · cites this paper as [52]
In the citing paper
“In this data-rich environment, tax administrations enjoy conditions that are very favourable to the use of AI technologies: 31 (i) the volume of data is so great that its direct examination is not feasible through human labour; 32 and (ii) the large amounts of data mean that any”
From Towards eXplainable Artificial Intelligence (XAI) in Tax Law: The Need for a Minimum Legal Standard (2022) · cites this paper as [31]
In the citing paper
“Lastly, the opacity of most current AI models is one of the most popular topics of contemporary AI research (Rudin 2019).”
From Trustworthy artificial intelligence (2021) · cites this paper as Rudin 2019
In the citing paper
“The literature supports this finding, emphasizing the importance of explainable AI in healthcare to ensure transparency and trustworthiness [30].”
From Leading with AI in critical care nursing: challenges, opportunities, and the human factor (2024) · cites this paper as [30]
In the citing paper
“The absence of any link to a factual knowledge source that can help LLMs reason on their generation introduce what is known as the “black box” effect (Rudin, 2019).”
From A review of the explainability and safety of conversational agents for mental health to identify avenues for improvement (2023) · cites this paper as Rudin, 2019
In the citing paper
“Such neglect of detail is necessary to provide explanations that are understandable for human agents, and, so the worry goes, there is a real risk that the omitted details are in fact crucial to the true causal story of how the decision was in fact produced”
From AI, Opacity, and Personal Autonomy (2022) · cites this paper as Rudin, 2019
In the citing paper
“Even with a multiscale model, the translation of genotype to phenotype can be complex, and the use of standard “black-box” machine learning algorithms prevents a straightforward biological interpretation (Rudin, 2019).”
From Mapping the multiscale structure of biological systems (2021) · cites this paper as Rudin, 2019
In the citing paper
“AI technology is widely known to be difficult to explain and understand; AI has long been criticized for inexplicable “black box algorithms,” and calls to demystify these algorithms have created an entire category of explainable AI [37,38].”
From Reimagining Core Entrustable Professional Activities for Undergraduate Medical Education in the Era of Artificial Intelligence (2023) · cites this paper as [38]
In the citing paper
“The reviewed studies suggest an inherent trade-off between model complexity, input representation, and explainability in clinical audio-based AI systems [32].”
From Explainable and Interpretable AI for Voice and Speech Analysis in Clinical Care: Systematic Review (2026) · cites this paper as [32]
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
Classification and Regression Trees.
Ryzin, J. V., Breiman, L., Friedman, J. H., Olshen, R. A., & Stone, C. J. (1986) · Journal of the American Statistical Association · doi:10.2307/2288003 · Cited by 21,030
The magical number seven, plus or minus two: Some limits on our capacity for processing information.
Miller, G. A. (1956) · Psychological Review · doi:10.1037/h0043158 · Cited by 17,812
The magical number seven, plus or minus two: Some limits on our capacity for processing information.
Miller, G. A. (1994) · Psychological Review · doi:10.1037/0033-295x.101.2.343 · Cited by 17,559
Domain-Adversarial Training of Neural Networks
Ganin, Y., Ustinova, E., Ajakan, H., Germain, P., Larochelle, H., Laviolette, F., Marchand, M., & Lempitsky, V. (2017) · Advances in Computer Vision and Pattern Recognition · doi:10.1007/978-3-319-58347-1_10 · Cited by 8,318
(1996) · Cited by 4,738
Definitions, methods, and applications in interpretable machine learning
Murdoch, W. J., Singh, C., Kumbier, K., Abbasi-Asl, R., & Yu, B. (2019) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1900654116 · Cited by 2,167
European Union Regulations on Algorithmic Decision Making and a “Right to Explanation”
Goodman, B., & Flaxman, S. (2017) · AI Magazine · doi:10.1609/aimag.v38i3.2741 · Cited by 2,131
Very Simple Classification Rules Perform Well on Most Commonly Used Datasets
Holte, R. C. (1993) · Machine Learning · doi:10.1023/a:1022631118932 · Cited by 1,800
Zech, J. R., Badgeley, M. A., Liu, M., Costa, A. B., Titano, J. J., & Oermann, E. K. (2018) · PLOS Medicine · doi:10.1371/journal.pmed.1002683 · Cited by 1,719 · cited 1×
“[34], who noticed that their neural network was picking up on the word “portable” within an x-ray image, representing the type of x-ray equipment rather than the medical content of the image.”
BackgroundThe Magical Mystery Four: How Is Working Memory Capacity Limited, and Why?
Cowan, N. (2010) · Current Directions in Psychological Science · doi:10.1177/0963721409359277 · Cited by 1,255
Flagged first, then by citations
Research works printed but not linked: 4
- Kodratoff, Y. The comprehensibility manifesto. KDD Nugget Newsletter https://www.kdnuggets.com/news/94/n9.txt (1994).
- Quinlan, J. R. C4. 5: Programs for Machine Learning Vol. 1 (Morgan Kaufmann, 1993).
- Angelino, E., Larus-Stone, N., Alabi, D., Seltzer, M. & Rudin, C. Certifiably optimal rule lists for categorical data. J. Mach. Learn. Res. 19, 1–79 (2018).
- Hu, X. (S.), Rudin, C. & Seltzer, M. Optimal sparse decision trees. Preprint at https://arxiv.org/abs/1904.12847 (2019). · arXiv
Cited by
Sorted by citations
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
Machine learning and deep learning
Janiesch, C., Zschech, P., & Heinrich, K. (2021) · Electronic Markets · doi:10.1007/s12525-021-00475-2 · Cited by 2,959
On the Opportunities and Risks of Foundation Models
Bommasani, R., Hudson, D. A., Adeli, E., Altman, R., Arora, S., von Arx, S., Bernstein, M. S., Bohg, J., Bosselut, A., Brunskill, E., et al. (2021) · arXiv · doi:10.48550/arxiv.2108.07258 · Cited by 2,676
Explainability for artificial intelligence in healthcare: a multidisciplinary perspective
the Precise4Q consortium, Amann, J., Blasimme, A., Vayena, E., Frey, D., & Madai, V. I. (2020) · BMC Medical Informatics and Decision Making · doi:10.1186/s12911-020-01332-6 · Cited by 2,203
Time-series forecasting with deep learning: a survey
Lim, B. Y., & Zohren, S. (2021) · Philosophical Transactions of the Royal Society A Mathematical Physical and Engineering Sciences · doi:10.1098/rsta.2020.0209 · Cited by 2,242
Metadata
- Authors
- Rudin, C.
- Affiliations
- Duke University (US)
- Corresponding author
- Rudin, Cynthia
- ORCID
- Rudin, Cynthia
- Venue
- Nature Machine Intelligence
- Published
- 2019-05-13
- Volume
- 1
- Issue
- 5
- Pages
- 206-215
- ISSN
- 2522-5839 (ISSN-L)
- PMC
- PMC9122117
- OpenAlex
- W2945976633
- Cited by
- 10,561
- References
- 50
- Citations by year
- 2026: 2997 · 2025: 2162 · 2024: 1555 · 2023: 1293 · 2022: 1094 · 2021: 914 · 2020: 461 · 2019: 72
- FWCI
- 289.88
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- Locations
- Nature Machine Intelligence (publishedVersion)
- arXiv (Cornell University) (publishedVersion)
- PubMed (publishedVersion)
- PubMed Central (acceptedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- National Science Foundation
- Laura and John Arnold Foundation
- National Institutes of Health
- Defense Advanced Research Projects Agency
- National Institute of Biomedical Imaging and Bioengineering (R01EB025021)
- 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 › Engineering › Safety, Risk, Reliability and Quality › Safety Systems Engineering in Autonomy
- Keywords
- black-box models, explainable AI, model interpretability, high-stakes decision making, criminal justice, health care, algorithmic harms, computer vision
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
