Notes and Sources

Notes and Sources

Endnotes

1. Ben Armstrong and Julie Shah, Humans in the Loop: The Evolution of Work in Early Experiments with Generative AI (MIT Industrial Performance Center, 8 April 2026), official publication page. The report describes early experiments across more than 20 companies in four industry groups. It does not establish a general outcome for all firms or prescribe a task-by-task redesign. ↩ · ↩ 2

1. Michael Polanyi, The Tacit Dimension, with a foreword by Amartya Sen (University of Chicago Press, 2009), publisher record. Polanyi’s work supports the bounded proposition that tacit knowing exceeds what a person can fully articulate. It does not establish what AI can or cannot learn, or prove an organizational-performance result. ↩ · ↩ 2

1. OECD, The Knowledge-Based Economy, OCDE/GD(96)102 (Paris: OECD, 1996), official PDF. The report distinguishes know-what, know-why, know-how, and know-who. It is a 1996 policy classification, not a current standard or evidence about present-day AI capability. ↩ · ↩ 2

1. Ikujiro Nonaka and Hirotaka Takeuchi, The Knowledge-Creating Company: How Japanese Companies Create the Dynamics of Innovation (Oxford University Press, 1995), https://doi.org/10.1093/oso/9780195092691.001.0001. Their account gives a central role to converting tacit and explicit knowledge. It is not evidence that an AI workflow captures tacit knowledge or improves performance. ↩ · ↩ 2

1. Madeleine Clare Elish, “Moral Crumple Zones: Cautionary Tales in Human-Robot Interaction,” Engaging Science, Technology, and Society 5 (2019): 40–60, https://doi.org/10.17351/ests2019.260. The concept describes responsibility being misattributed to a human with limited control over an automated or autonomous system. It is not a settled regulatory rule, and it does not make every human review point a moral crumple zone. ↩ · ↩ 2

1. Filippo Santoni de Sio and Jeroen van den Hoven, “Meaningful Human Control over Autonomous Systems: A Philosophical Account,” Frontiers in Robotics and AI 5, article 15 (2018), https://doi.org/10.3389/frobt.2018.00015. The paper argues that mere human presence is insufficient for meaningful human control. Its original context includes autonomous systems; it is neither a legal rule nor a universal enterprise-control design. ↩ · ↩ 2

1. Jacob Beck, Stephanie Eckman, Christoph Kern, and Frauke Kreuter, “Bias in the Loop: How Humans Evaluate AI-Generated Suggestions,” Harvard Data Science Review 8, no. 2 (Spring 2026), https://doi.org/10.1162/99608f92.0e98898d. In a randomized experiment with 2,784 participants reviewing AI-extracted greenhouse-gas values from corporate reports, correction burden and attitudes toward AI affected acceptance and correction behavior. It is not evidence that all human review is unreliable or a study of enterprise accountability approvals in general. ↩ · ↩ 2

1. John Lunney and Sue Lueder, “Postmortem Culture: Learning from Failure,” in Site Reliability Engineering: How Google Runs Production Systems, ed. Betsy Beyer, Chris Jones, Jennifer Petoff, and Niall Richard Murphy, chapter 15 (O’Reilly Media, 2016), official chapter. The chapter describes recording incident impact, mitigation, causes, and follow-up actions intended to prevent recurrence. Its blameless framing is an engineering-practice precedent, not validation of an AI-governance, employment, disciplinary, safety, or legal process. ↩ · ↩ 2

Selected Sources and Intellectual Context

The following works offer intellectual context for the questions developed in this book: how organizations retain knowledge, how authority should relate to consequential work, how human judgment operates around automated systems, and how learning and measurement shape operating design. They are included for the limited propositions described here, not as a single doctrine or as a substitute for a firm’s own decision.

1. Michael Polanyi, The Tacit Dimension, with a foreword by Amartya Sen (University of Chicago Press, 2009). Polanyi’s opening formulation that people can know more than they can tell helps frame the distinction between documented material and judgment that has to become usable in later work. It does not establish what AI can or cannot learn, or prove a company outcome.

1. Ikujiro Nonaka and Hirotaka Takeuchi, The Knowledge-Creating Company: How Japanese Companies Create the Dynamics of Innovation (Oxford University Press, 1995), https://doi.org/10.1093/oso/9780195092691.001.0001. Their account treats tacit–explicit knowledge conversion as central to organizational knowledge creation. It is not evidence that an AI workflow captures tacit knowledge or improves performance.

1. OECD, The Knowledge-Based Economy, OCDE/GD(96)102 (1996). The report’s distinction among know-what, know-why, know-how, and know-who is a useful lens for asking what kind of context a workflow needs. It is a 1996 policy classification, not a current standard or a claim about AI capability.

1. Madeleine Clare Elish, “Moral Crumple Zones: Cautionary Tales in Human-Robot Interaction,” Engaging Science, Technology, and Society 5 (2019): 40–60, https://doi.org/10.17351/ests2019.260. Elish develops the concept of responsibility being misattributed to a human with limited control over automated-system behavior. The concept is not a regulatory rule, and it does not make every human review point a moral crumple zone.

1. Filippo Santoni de Sio and Jeroen van den Hoven, “Meaningful Human Control over Autonomous Systems: A Philosophical Account,” Frontiers in Robotics and AI 5, art. 15 (2018), https://doi.org/10.3389/frobt.2018.00015. This work offers a philosophical account of preserving human control and responsibility over autonomous-system decisions. Its original context includes autonomous systems; it is neither a legal rule nor a universal enterprise-control design.

1. John Lunney and Sue Lueder, “Postmortem Culture: Learning from Failure,” in Site Reliability Engineering: How Google Runs Production Systems, ed. Betsy Beyer et al. (O’Reilly Media, 2016), chapter 15. The chapter describes a postmortem practice that records impact, mitigation, causes, and follow-up actions intended to prevent recurrence. Its blameless framing informs this book’s emphasis on changing a method after failure; it does not validate an employment, disciplinary, or legal process.

1. Ben Armstrong and Julie Shah, Humans in the Loop: The Evolution of Work in Early Experiments with Generative AI (MIT Industrial Performance Center, 8 April 2026). The report examines early experiments in which professional and technical work shifted toward supervision, analysis, explanation, and troubleshooting. It is an early-experiment report, not a general outcome claim or a prescribed task-by-task redesign.

1. Jacob Beck, Stephanie Eckman, Christoph Kern, and Frauke Kreuter, “Bias in the Loop: How Humans Evaluate AI-Generated Suggestions,” Harvard Data Science Review 8, no. 2 (Spring 2026), https://doi.org/10.1162/99608f92.0e98898d. The study examines how correction-work design and participants’ attitudes affected acceptance and correction of AI-generated suggestions in a defined experimental setting. It is not evidence that all human review is unreliable, nor a study of organizational accountability approvals in general.

The book’s accountability-unit and capacity-allocation methods remain management arguments. These sources help a reader locate adjacent intellectual questions; they do not decide a firm’s authority boundary, control design, workforce choice, customer promise, or legal duty.