Where Responsible AI meets Reality: Practitioner Perspectives on Enablers for Shifting Organizational Practices
Open accessAbstract
Large and ever-evolving technology companies continue to invest more time and resources to incorporate responsible Artificial Intelligence (AI) into production-ready systems to increase algorithmic accountability. This paper examines and seeks to offer a framework for analyzing how organizational culture and structure impact the effectiveness of responsible AI initiatives in practice. We present the results of semi-structured qualitative interviews with practitioners working in industry, investigating common challenges, ethical tensions, and effective enablers for responsible AI initiatives. Focusing on major companies developing or utilizing AI, we have mapped what organizational structures currently support or hinder responsible AI initiatives, what aspirational future processes and structures would best enable effective initiatives, and what key elements comprise the transition from current work practices to the aspirational future.
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
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?
Fairness can manifest in many different forms following the abstract fairness principles.
“fairness can manifest in many different forms following these principles (see [356, 65] for comprehensive surveys, and [228] for AI ethics' checklists).”
From Trustworthy AI: From Principles to Practices (2023) · cites this paper as [65]
Section 2.1 (p. 3) describes broad agreement among abstract ethical guidelines but says how they are translated into practice in each context remains unclear. The study’s table of practitioner workstreams (Section 3, p. 7) includes fairness evaluations and internal checklists, supporting the idea of different practical forms. It does not itself provide the AI ethics checklists for which this citation is offered. (LLM, conf=0.78)
Experimental — accuracy not yet measuredLiterature [74, 10, 65] has pointed a series of reforms to the organization of participants and procedures to build trustworthy AI.
“Literature such as [74, 10, 65] has pointed a series of reforms to the organization of participants and procedures to build trustworthy AI.”
From Trustworthy AI: From Principles to Practices (2023) · cites this paper as [65]
These reforms include responsible product design aligned with trustworthiness principles [6], new human resource strategies [74, 65], and justification of trustworthiness via continuous auditing [74].
“These reforms include responsible product design aligned with trustworthiness principles [6], new human resource strategies [74, 65], and justification of trustworthiness via continuous auditing [74].”
From Trustworthy AI: From Principles to Practices (2023) · cites this paper as [65]
Continuous variables often require binning, quantization, or loss functions to evaluate fairness.
“while continuous variables often further require binning, quantization, or loss functions to evaluate fairness [65].”
From Trustworthy AI: From Principles to Practices (2023) · cites this paper as [65]
Section 4.3 (p. 12) discusses difficulties choosing metrics for responsible AI work, but not how to evaluate fairness for continuous variables. Searches for binning, quantization, and loss functions found no passage addressing the claimed methods. (LLM, conf=0.76)
Experimental — accuracy not yet measuredAn empirical correlation of discrete variables can be evaluated with standard statistical tools such as correlation coefficients and ANOVA.
“An empirical correlation of discrete variables can be evaluated with standard statistical tools, such as correlation coefficients (Pearson/Kendall/Spearman) and ANOVA), while continuous variables often further require binning, quantization, or loss functions to evaluate fairness [65].”
From Trustworthy AI: From Principles to Practices (2023) · cites this paper as [65]
The Study and Methods section (p. 8) describes qualitative interview analysis using affinity diagramming, not statistical evaluation of discrete variables. The discussion of metrics in “How do we measure success?” (p. 12) does not address correlation coefficients or ANOVA. Searches for those methods and for discrete-variable analysis found no supporting passage in our copy. (LLM, conf=0.77)
Experimental — accuracy not yet measured
At the group level, researchers have identified three abstract principles to categorize different types of fairness.
“At the group level, researchers have identified three abstract principles to categorize different types of fairness [65].”
From Trustworthy AI: From Principles to Practices (2023) · cites this paper as [65]
Section 2.1 (p. 3) discusses numerous AI-ethics principles and guidelines, but does not identify three principles for categorizing group-level fairness. Searches for the claimed three principles and for group-fairness categories found no passage addressing that classification. (LLM, conf=0.76)
Experimental — accuracy not yet measuredFairness can be applicable at multiple granularities of system behavior.
“Fairness can be applicable at multiple granularities of system behavior [242, 65, 339].”
From Trustworthy AI: From Principles to Practices (2023) · cites this paper as [65]
Section 2.1 (p. 3) distinguishes technical, statistical disparities from broader questions about just outcomes, but does not say that fairness applies at multiple granularities of system behavior. Searches for granularities, levels of fairness, and individual versus system-level outcomes found no passage establishing that specific point. (LLM, conf=0.68)
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
No citation lists for this paper
No citation lists for this paper yet — checked again daily.
References 36 · Cited by 329
Research works printed but not linked: 6
- [1][1] Kenneth A. Bamberger and Deirdre K. Mulligan. 2015. Driving Corporate Behavior in the United States and Europe . MIT Press. -
- [2][2] Michalle E Mor Barak. 2016. Managing diversity: Toward a globally inclusive workplace . Sage Publications. -
- [6][6] Joy Buolamwini and Timnit Gebru. 2018. Gender shades: Intersectional accuracy disparities in commercial gender classification. In Conference on fairness, accountability and transparency . 77-91. -
- [8][8] Henriette Cramer, Jenn Wortman-Vaughan, Kenneth Holstein, Hanna Wallach, Hal Daumé III, Miroslav Dudík, Sravana Reddy, and Jean Garcia-Gathright. [n.d.]. Industry Translation Tutorial: Algorithmic fairness in practice. In Conference on Fairness, Accountability, and Transparency, FAT* 2019 . -
- [23][23] Debra Meyerson. 2004. The tempered radicals: How employees push their companies-little by little-to be more socially responsible. Stanford Social Innovation Review (2004), 1-23. -
- [34][34] Yi Zeng, Enmeng Lu, and Cunqing Huangfu. 2018. Linking Artificial Intelligence Principles. arXiv:cs.AI/1812.04814
- Authors
- Rakova, B., Yang, J., Cramer, H., & Chowdhury, R.
- Affiliations
- Partnership on AI & Accenture, San Francisco, CA, USA
- Partnership on AI, San Francisco, CA, USA
- Spotify, San Francisco, CA, USA
- Accenture (United States) (US)
- Venue
- Proceedings of the ACM on Human-Computer Interaction
- Published
- 2021-04-13
- Volume
- 5
- Issue
- CSCW1
- Pages
- 1-23
- ISSN
- 2573-0142 (ISSN-L)
- OpenAlex
- W3158479996
- Cited by
- 329
- References
- 36
- Citations by year
- 2026: 72 · 2025: 84 · 2024: 69 · 2023: 43 · 2022: 43 · 2021: 16 · 2020: 2
- FWCI
- 53.51
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- Proceedings of the ACM on Human-Computer Interaction (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University) (cc-by)
- Topics
- Social Sciences › Safety Research › Ethics and Social Impacts of AI
- Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
- Social Sciences › Decision Sciences › Information Systems and Management › Ethics in Business and Education
- Keywords
- responsible AI, organizational culture, organizational structure, algorithmic accountability, AI ethics, ethical conflicts, technology companies, AI governance, organizational change, practitioner perspectives
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
