The Intuitive Appeal of Explainable Machines

By Andrew D. Selbst, Solon Barocas. 2018

389 citations recorded by the source corpus.

Subfields: Safety Research, Safety, Robustness & Interpretability, AI Ethics, Fairness & Society

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Safety Research

In Safety, Robustness & Interpretability

In AI Ethics, Fairness & Society

Citations in the corpus

Measurement and Fairness (2019, background): “ate the space of potential harms. Collapsing the distinctions between unobservable theoretical constructs and their operationalizations makes society less fair, less transparent, and less accountable [40, 58]. The language of measurement provides a framework to make our assumptions explicit, and con”
Explainable machine learning in deployment (2019, background): “ide transparency directly targeted to human users, often aiming to increase trustworthiness [44]. The importance of explainability as a concept has been reﬂectedinlegalandethicalguidelinesfordataandML[53].Incases ofautomateddecision-making,Articles13-15oftheEuropeanGeneral Data Protection Regulation”
Explaining Explanations in AI (2018, neutral): “abermas, 1984)). Justification has recently received increasing attention by scholars discussing algorithmic accountability (e.g. Binns (2017); Biran and McKeown (2014); Hildebrandt and Koops (2010); Selbst and Barocas (2018); Weller (2017)). However, justification currently occupies an uneasy posit”
A survey of algorithmic recourse: definitions, formulations, solutions, and prospects (2020, background): “Related to recourse, and of practical concern [21, 161, 165, 171], is a study of the ability of an adversary with access to a recourse API in extracting a model.”
The philosophical basis of algorithmic recourse (2020, background): “For a detailed exposition of the nature of explanations in decision-making, see the work of Selbst and Barocas [30].”
"The human body is a black box": supporting clinical decision-making with deep learning (2019, background): “” AI [18, 19, 20, 21]. Recent work has emphasized that explanations cannot be an end in and of themselves, but must be enacted in the service of specific normative ends within broader social contexts [22, 23]. 2.2 Adoption of healthcare innovations The challenges facing the introduction of machine l”
Ethical Implications and Accountability of Algorithms (2018, neutral): “…as to how algorithms can and should be transparent in order to be governed—including more autonomous algorithms such as machine learning, artificial intelligence, and neural networks (Burrell 2016; Desai and Kroll 2017; Howard 2014; Kroll et al. 2017; Ziewitz 2016; Selbst and Barocas 2018).”
Model Multiplicity: Opportunities, Concerns, and Solutions (2022, background): “KEYWORDS Model multiplicity, predictive multiplicity, procedural multiplicity, fairness, discrimination, recourse, arbitrariness ACMReference Format: Emily Black, Manish Raghavan, and Solon Barocas.”
Legal requirements on explainability in machine learning (2020, background): “Some explainability requirements relate to the model while others relate to the decision (Wachter et al. 2017; Selbst and Barocas 2018).”
How can we fool LIME and SHAP? Adversarial Attacks on Post hoc Explanation Methods (2019, background): “Selbst and Barocas (2018) and Kroll et al. (2016) provide examples showing that even if a model is completely transparent, it is hard to detect and prevent bias due to the existence of correlated variables.”
Explainable Neural Computation via Stack Neural Module Networks (2018, background): “This includes work aimed at both explaining the decision rules implemented by learned models, and the mechanisms by which these rules are derived from data [32,20].”
Explainable AI under contract and tort law: legal incentives and technical challenges (2020, methodology): “…more accurate models [refined via LASSO, for example Tibshi-rani (1996)] may even sometimes be more explainable as they pick up less spurious correlations from the training data and focus on the “relevant” correlations (Selbst and Barocas 2018); these can then be the basis for an explanation.”
Problems with Shapley-value-based explanations as feature importance measures (2020, methodology): “We use the framework set out by Selbst and Barocas [2018], who argue that there are three general motivations behind the call for explanations in AI.”
The fallacy of inscrutability (2018, background): “Much has been said about the ‘black box’ of computer systems, especially those based on data analysis, data-derived models and machine learning [1,2].”