MBA Thesis  •  References

Bibliography and Appendixes

The reference list below is rendered in Steinbeis house format. It is alphabetised by first author. Two formatting rules from the Steinbeis Guidelines govern the entries. Italics mark the title of a book, a journal, or a standalone online work. Internet sources and reports carry the date of last retrieval in parentheses at the end.

Bibliography

Foundational canon: theory and method

This group holds the human-automation research tradition that the paper builds on, the paradox theory that types each tension, and the method literature that justifies the research design. These sources are cited across Chapters II, III, V, and VI rather than tied to a single cluster.

  • Bainbridge, L. (1983). Ironies of automation. Automatica, 19(6), pp. 775-779.
  • Christensen, C. M. (1997). The Innovator's Dilemma: When New Technologies Cause Great Firms to Fail. Boston: Harvard Business School Press.
  • Endsley, M. R. (1995). Toward a theory of situation awareness in dynamic systems. Human Factors, 37(1), pp. 32-64.
  • Kaber, D. B., & Endsley, M. R. (1997). Out-of-the-loop performance problems and the use of intermediate levels of automation for improved control system functioning and safety. Process Safety Progress, 16(3), pp. 126-131.
  • Lee, J. D., & See, K. A. (2004). Trust in automation: Designing for appropriate reliance. Human Factors, 46(1), pp. 50-80.
  • March, J. G. (1991). Exploration and exploitation in organizational learning. Organization Science, 2(1), pp. 71-87.
  • McGrath, J. E. (1982). Dilemmatics: The study of research choices and dilemmas. In: McGrath, J. E., Martin, J., & Kulka, R. A. (Eds.), Judgment Calls in Research (pp. 69-102). Beverly Hills: Sage.
  • Ming, V. (2026). Robot-Proof Reasoning: Well-Posed and Ill-Posed Problems in the Age of AI. Place of publication to be confirmed: Publisher to be confirmed.
  • Parasuraman, R., & Riley, V. (1997). Humans and automation: Use, misuse, disuse, abuse. Human Factors, 39(2), pp. 230-253.
  • Parasuraman, R., Sheridan, T. B., & Wickens, C. D. (2000). A model for types and levels of human interaction with automation. IEEE Transactions on Systems, Man, and Cybernetics: Part A: Systems and Humans, 30(3), pp. 286-297.
  • Sheridan, T. B., Verplank, W. L., & Brooks, T. L. (1978). Human and Computer Control of Undersea Teleoperators. Cambridge, MA: Man-Machine Systems Laboratory, Department of Mechanical Engineering, Massachusetts Institute of Technology.
  • Smith, W. K., & Lewis, M. W. (2011). Toward a theory of paradox: A dynamic equilibrium model of organizing. Academy of Management Review, 36(2), pp. 381-403.
  • Snyder, H. (2019). Literature review as a research methodology: An overview and guidelines. Journal of Business Research, 104, pp. 333-339.
  • Tranfield, D., Denyer, D., & Smart, P. (2003). Towards a methodology for developing evidence-informed management knowledge by means of systematic review. British Journal of Management, 14(3), pp. 207-222.

Bibliography

Cluster sources

This group holds the sources cited in the descriptive and analytical treatment of the five clusters: Trust Miscalibration, Automation Bias, Automation-Induced Complacency, Automation-Induced Skill Atrophy, and Cognitive Debt Accumulation. Several entries also supply the capability-cost snapshots that open the paper.

Appendix A

Paradox Summaries

This appendix summarizes each cluster's paradox in one place. For every cluster, it gives the business tension in plain terms, the two-pole tension underneath it, the yes/no question put to the evidence, the verdict, the foundational source, and the stakes of getting it wrong. All five resolve to paradox, because neither pole is defensible alone. This is the analytical result behind the paper's title.

A.1 Trust Miscalibration

  • Business tension: Trust miscalibration is the gap between the confidence humans place in agentic systems and what they actually deliver.
  • Stakes: Get the trust level wrong, and every decision built on the AI inherits the error.
  • Recurring tension: Miscalibrated trust (load-bearing) and Informed reliance.
  • Classification question: When a human operator leans on AI output, can that operator rely on it without checking it against what the system can actually do under deployment conditions?
  • Verdict: Paradox. Neither pole holds alone. Overtrust produces misuse, and distrust produces disuse. Both forces are real, so the tension is managed by holding them together rather than choosing one.
  • Anchor source: Lee and See (2004) set out the framework for when people trust automation too much and too little.

A.2 Automation Bias

  • Business tension: Automation bias is the human tendency to defer to automated or agentic systems over one's own judgment, accepting their outputs with less scrutiny than one would apply to one's own work.
  • Stakes: Skip the second check, and the rare errors that slip through are the costly ones: they reach real decisions, surface only after the damage is done, and leave the operator holding the blame.
  • Recurring tension: Cognitive passivity (load-bearing) and Operator vigilance.
  • Classification question: Is deferring to a high-accuracy AI recommendation, without independent checking, a defensible operator practice?
  • Verdict: Paradox. The failure deepens as the system improves. The more accurate the AI, the more an operator relaxes verification, so vigilance and reliance must be kept in balance.
  • Anchor source: Lyell and Coiera (2017) reviewed, across many clinical studies, how often clinicians wrongly defer to automated decision support.

A.3 Automation-Induced Complacency

  • Business tension: Automation-induced complacency is the tendency to lower active monitoring of an automated system that has proved dependable, sliding from close supervision into a passive sign-off given with barely a glance.
  • Stakes: The more reliable the automation, the less alert anyone is the moment it fails.
  • Recurring tension: Operational over-reliance (load-bearing) and Processing speed.
  • Classification question: Is sustained human vigilance a coherent requirement over highly reliable automated systems in safety-critical and high-tempo operations?
  • Verdict: Paradox. The same reliability that makes the system worth deploying is what dulls the attention meant to supervise it.
  • Anchor source: Parasuraman and Riley (1997) named the conditions under which people use, misuse, and disuse automation.

A.4 Automation-Induced Skill Atrophy

  • Business tension: Automation-induced skill atrophy is the erosion of hands-on competence in tasks that once only humans could perform.
  • Stakes: Hand the skill to the machine long enough, and on the day it fails, no one in the room remembers how the work was done.
  • Recurring tension: Skill atrophy (load-bearing) and Skill maintenance.
  • Classification question: Can operators keep the skills they need for failure recovery while automation reliably does the task for them?
  • Verdict: Paradox. Reliable automation erodes the skills needed to recover when it eventually fails, so those skills have to be maintained alongside the automation that makes them feel unnecessary. This is Bainbridge’s classic “irony of automation”.
  • Anchor source: Bainbridge (1983) set out the ironies of automation: skills fade when people stop practicing them.

A.5 Cognitive Debt Accumulation

  • Business tension: Cognitive debt accumulation is the lack of a true, shared understanding of a product or system that a person or team experiences when an agentic system delivers finished, working artifacts at a much faster pace than anyone could produce on their own.
  • Stakes: Ship work nobody fully understands today, and every later fix and audit gets harder.
  • Recurring tension: Decontextualized automation (load-bearing) and Scalable consistency.
  • Classification question: Is shipping AI-generated work at scale a defensible way to sustain a team's capacity to keep evolving that work?
  • Verdict: Paradox. The faster the system ships working output, the less the team understands what it ships, which quietly raises the cost of every future fix.
  • Anchor source: Storey (2026) was the first to frame the long-term cognitive cost of relying on AI for knowledge work.

Appendix B

Corpus

The corpus holds 62 pieces of evidence across the five clusters: 13 in Trust Miscalibration, 13 in Automation Bias, 12 in Automation-Induced Complacency, 13 in Automation-Induced Skill Atrophy, and 11 in Cognitive Debt Accumulation. Of these, 61 classify as paradoxes and 1 as a dilemma, a paradox share of 98.4 percent. The sole dilemma lies within Automation Bias. At the cluster level, all five clusters resolve as paradoxes, so the cluster-level verdict is 5 of 5 paradoxes.

The corpus draws on 26 academic papers, accounting for 42 percent of the corpus. The academic share varies by cluster: 77 percent in Trust Miscalibration, 15 percent in Automation Bias, 50 percent in Automation-Induced Complacency, 38 percent in Automation-Induced Skill Atrophy, and 27 percent in Cognitive Debt Accumulation. The lower-academic clusters rely on practitioner and primary media sources, weighted using the source-quality framework described in Chapter III.

Throughout this paper, the corpus was checked at five levels. The first level is the validity of each source. The second is the fidelity of each extraction from a source to an evidence record. The third is the soundness of each dilemma-or-paradox classification. The fourth is the coherence of each cluster's synthesis. The fifth is the consistency of the grouping into the corpus.

Appendix C

Canonical Vocabulary Tables

These tables reproduce the rows of the controlled vocabularies that the source-to-cluster spine draws from. Each table lists only the entries used in this analysis.

C.1 Recurring Tensions

Each recurring tension names a tension shape that recurs across business domains. All five tensions are paradox-typical, both-and axes. Each manifests as a paired AI-side and Human-side duality, shown for individual cases in the Equilibrium Mappings of Chapter IV and in the traces of Appendix A. The load-bearing pole is named first.

ClusterRecurring Tension (load-bearing and counter-pole)What it trades off
Trust MiscalibrationMiscalibrated trust and Informed relianceReading AI reliability wrong in either direction is costly: overtrust lets bad outputs into real decisions, and undertrust leaves a working tool unused.
Automation BiasCognitive passivity and Operator vigilanceDeferring to a fast, polished recommendation saves effort, and it displaces the independent check that catches the system's errors.
Automation-Induced ComplacencyOperational over-reliance and Processing speedLeaning on reliable automation keeps operations fast, and it dulls the sustained attention meant to supervise the system.
Automation-Induced Skill AtrophySkill atrophy and Skill maintenanceLetting the system do the task reliably erodes the human skill needed to recover when it fails.
Cognitive Debt AccumulationDecontextualized automation and Scalable consistencyAutomating a judgment makes it consistent at scale, and it strips the context a human would have weighed.

C.2 Business Domains

The evidence groups into the business domains below, which anchor the Dichotomy Probe. They are industry-agnostic.

Business DomainDescriptionTypical contexts
AI-assisted decision-makingHumans make the final decision with AI providing input, analysis, or recommendations. The AI augments judgment, and the human keeps authority and accountability.clinical diagnosis, portfolio advisory, credit scoring, hiring screening, triage
AI-augmented knowledge workIndividual-cognition-scale work supported by AI, such as essay, analytical, research, and creative writing, and solo-developer coding. It foregrounds the individual cognitive scaffold over collective workflow.essay and analytical writing, research writing, strategy or policy memos, solo-developer coding, creative writing
AI-augmented operationsHigh-tempo operational work where AI augments human execution, from decision support, with humans driving and AI advising, to supervisory control, with AI driving and humans approving. The largest domain in the evidence base.developer copilots, customer support, content moderation, incident response, sales enablement
Safety-critical automationLife- or safety-critical systems where automation failures carry severe physical consequences. It spans the full range of automation and is the home of the classical automation literature.autonomous vehicles, aviation autopilot, surgical robotics, industrial control, production cloud infrastructure

C.3 DEM Categories (Smith and Lewis, 2011)

Four paradox-generating activities, after Smith and Lewis (2011). The clusters use three of these four categories: three classify as Learning paradoxes, one as a Performing paradox, and one as an Organizing paradox. Belonging is listed for completeness, because Chapter II introduces all four.

CategoryDefinitionUse in this paper
LearningTensions that surface as systems change, renew, and innovate, building on as well as discarding the past to create the future.Three clusters: Trust Miscalibration, Automation-Induced Skill Atrophy, and Cognitive Debt Accumulation. The human cost of outsourcing cognition is a learning tension: the capability the team gains today erodes the capability it needs tomorrow.
BelongingTensions of identity between the individual and the collective, as people and groups seek both similarity and distinction.Not a dominant category for any cluster.
OrganizingCompeting process designs to reach a desired outcome: collaboration and competition, control and flexibility, routine and change.Automation-Induced Complacency. The tension is between sustained oversight and operational reliance, a competing-process-design choice.
PerformingTensions from the plurality of stakeholders and the resulting competing strategies and goals.Automation Bias. The tension is between decision speed and decision accuracy, a competing-goals choice.

C.4 Levels of Automation (Sheridan, 1978)

The evidence sits at the automation intervals below, drawn from the four-level reduction of Sheridan's ten-point scale: Level 1 Human, Level 2 AI-assisted, Level 3 Human-in-the-loop, Level 4 Autonomous.

IntervalMeaningSheridan range
AI-assisted human decisionTensions within the AI-assisted zone: the scope and quality of AI recommendations to a human decision-maker.2 to 5
AI-assisted to Human-in-the-loopShifting decision authority from human to AI with human validation. The most common interval in the evidence, covering automation bias and complacency.2 to 5, then 6 to 9
Human-in-the-loop AI decisionTensions within the human-in-the-loop zone: veto latency, oversight quality, meaningful human control.6 to 9

C.5 Thesis Scope

This paper covers the human side of the AI-and-human relationship. Its tensions are cognitive failures inside the operator of an agentic system, and they group on the human side of that pairing. The table below places this paper within the wider thesis, which also covers the system, organization, and society sides, all of which are out of scope here. The project refers to this placement as the Tier.

Supercluster NameTierDescriptionEquilibrium Mapping ConvergenceIn Scope?
SC1: Outsourcing Human CognitionHumanCognitive failuresHuman sideIn Scope
SC2: Governing Algorithmic SystemsSystemAI misalignment in organizationsAI sideOut of Scope
SC3: Deploying and Scaling AI SystemsOrganizationAI misalignment in organizations governance and process.N/AOut of Scope
SC4: Societal Implications of AISocietyAI, Civic-justice and existential concerns.N/AOut of Scope

C.6 AI-side Equilibrium Mapping poles

This table presents the AI-side pole of each cluster's Equilibrium Mapping (the full mappings appear in Chapter IV). 🅐 is the load-bearing pole (left), the one whose failure mode typically jeopardizes agentic deployment; 🅑 is the counter-pole (right).

Cluster🅐 Load-bearing pole🅑 Counter-pole
Trust MiscalibrationOn-task reliability gapCapability-grounded transparency
Automation BiasVerification overloadDecision accuracy
Automation-Induced ComplacencyComplacency inductionReliable operational capability
Automation-Induced Skill AtrophyComplacency inductionOperational efficiency
Cognitive Debt AccumulationDecontextualized automationScalable consistency

C.7 Human-side Equilibrium Mapping poles

This table gathers the Human-side pole of each cluster's Equilibrium Mapping (the full mappings appear in Chapter IV). 🅒 is the load-bearing pole (left), the one whose failure mode typically risks the agentic deployment; 🅓 is the counter-pole (right).

Cluster🅒 Load-bearing pole🅓 Counter-pole
Trust MiscalibrationMiscalibrated trustInformed reliance
Automation BiasCognitive passivityOperator vigilance
Automation-Induced ComplacencyFunctional sedationEffective command
Automation-Induced Skill AtrophySkill atrophySkill maintenance
Cognitive Debt AccumulationCognitive passivityDeliberate reasoning