Konstantinos Pechlivanis
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Konstantinos Pechlivanis has a total of 6 co-authors. Here are the top co-authors by number of shared publications:
- Dimitrios Ziakkas
Purdue University
9 shared publications - Spiridon Chazapis3 shared publications
Abner Del Cid Flores
Purdue University
1 shared publicationDimitrios Ziakkas
Purdue University
1 shared publication- Don Harris
Coventry University
1 shared publication Zafeiria Chazapi
1 shared publication
Latest Publications
- Open Access
- 19 August 2026
- Published by Elsevier in SSRN Electronic Journal
The substitute is not the solution: unaccountable informal knowledge transmission as an emerging safety concern in commercial aviation
Aviation’s informal knowledge channel is thinning as an experienced cohort retires. European safety planning identifies this loss as a hazard. This research examines the resulting vacuum, and argues that the substitute is itself a safety concern rather than a remedy. Two corpora of public aviation video from one platform by two discovery paths: a browse corpus of 100 high-reach videos from twenty stratified channels, and a search corpus of 110 videos surfaced by twenty-two occupationally framed queries. Both were segmented into coding units, analysed against a common codebook, and rated on an identical four-domain instrument, and were never pooled. Technical accuracy was high in both (median 5 of 5; browse mean 4.57, search mean 4.86). The concern identified here is not inaccurate content. It is systemic. The observable practice of accountable claim-making, comprising the flagging of speculation, the weighing of evidence and the anchoring of claims to investigative record, collapsed more than fivefold on the search path, within an instructionally framed query set, while taught human-factors content persisted. The source-anchored end of authority construction hollowed out, leaving warrants grounded in personal standing. No shared pedagogical standard was observable, and one unanchored query returned no relevant results. We characterise this as an unmonitored transmission pathway in which competence is conveyed without the correction mechanisms that made workplace informal learning tolerable; no such mechanism was identified within the video content itself, comment-level correction not having been analysed. We argue that it warrants attention as a hazard in its own right.
- 19 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessThe substitute is not the solution: unaccountable informal knowledge transmission as an emerging safety concern in commercial aviation
Aviation’s informal knowledge channel is thinning as an experienced cohort retires. European safety planning identifies this loss as a hazard. This research examines the resulting vacuum, and argues that the substitute is itself a safety concern rather than a remedy. Two corpora of public aviation video from one platform by two discovery paths: a browse corpus of 100 high-reach videos from twenty stratified channels, and a search corpus of 110 videos surfaced by twenty-two occupationally framed queries. Both were segmented into coding units, analysed against a common codebook, and rated on an identical four-domain instrument, and were never pooled. Technical accuracy was high in both (median 5 of 5; browse mean 4.57, search mean 4.86). The concern identified here is not inaccurate content. It is systemic. The observable practice of accountable claim-making, comprising the flagging of speculation, the weighing of evidence and the anchoring of claims to investigative record, collapsed more than fivefold on the search path, within an instructionally framed query set, while taught human-factors content persisted. The source-anchored end of authority construction hollowed out, leaving warrants grounded in personal standing. No shared pedagogical standard was observable, and one unanchored query returned no relevant results. We characterise this as an unmonitored transmission pathway in which competence is conveyed without the correction mechanisms that made workplace informal learning tolerable; no such mechanism was identified within the video content itself, comment-level correction not having been analysed. We argue that it warrants attention as a hazard in its own right.
- Open Access
- 16 August 2026
- Published by Elsevier in SSRN Electronic Journal
Measuring a lexical safeguard: how reliably can an automated screen enforce a content rule in language-model-generated training material? A two-corpus measurement study
Language models increasingly generate training material in safety-critical domains, including aviation, where content rules are enforced by a prompt instruction and an automated lexical screen. That screen is described to the humans behind it as a check, yet its reliability is asserted rather than measured.,This paper measures one deployed screen against two corpora, reported as a diagnostic-accuracy study. The screen enforces two rules in an aviation training generator: items must stop before resolving their scenario and supply no model answer. Study 1 assessed detection and generalisation on 48 authored items, including evasive paraphrases, plus 18 items held out after the patterns were frozen. Study 2 ran the screen over 4,667 safety-report synopses from the NASA Aviation Safety Reporting System (2023-2026), scoring a random subsample of 100 against the adjudicated consensus of two independent blinded expert coders (Cohen’s kappa = .94).,As shipped, the screen caught 43% of authored violations (95% CI [27, 61]) at 100% specificity. Broadening raised in-sample recall to 97% [83, 99], significantly above both the held-out estimate of 50% [24, 76] (p = .002) and the out-of-distribution estimate of 22% [14, 32] (p < .001). Missed violations were figurative or idiomatic, not keyword-bearing. The safeguard is a high-precision tripwire, not an independent cross-check, and the human reviewer remains the control.,Two limitations apply: the production violation rate was not estimated, and both coders were study authors. Safeguards on generated content should be described to users as their measurements warrant, which emerging European regulation now makes explicit.
- 16 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessMeasuring a lexical safeguard: how reliably can an automated screen enforce a content rule in language-model-generated training material? A two-corpus measurement study
Language models increasingly generate training material in safety-critical domains, including aviation, where content rules are enforced by a prompt instruction and an automated lexical screen. That screen is described to the humans behind it as a check, yet its reliability is asserted rather than measured.,This paper measures one deployed screen against two corpora, reported as a diagnostic-accuracy study. The screen enforces two rules in an aviation training generator: items must stop before resolving their scenario and supply no model answer. Study 1 assessed detection and generalisation on 48 authored items, including evasive paraphrases, plus 18 items held out after the patterns were frozen. Study 2 ran the screen over 4,667 safety-report synopses from the NASA Aviation Safety Reporting System (2023-2026), scoring a random subsample of 100 against the adjudicated consensus of two independent blinded expert coders (Cohen’s kappa = .94).,As shipped, the screen caught 43% of authored violations (95% CI [27, 61]) at 100% specificity. Broadening raised in-sample recall to 97% [83, 99], significantly above both the held-out estimate of 50% [24, 76] (p = .002) and the out-of-distribution estimate of 22% [14, 32] (p < .001). Missed violations were figurative or idiomatic, not keyword-bearing. The safeguard is a high-precision tripwire, not an independent cross-check, and the human reviewer remains the control.,Two limitations apply: the production violation rate was not estimated, and both coders were study authors. Safeguards on generated content should be described to users as their measurements warrant, which emerging European regulation now makes explicit.
- Open Access
- 11 July 2026
- Published by Elsevier in Journal of Air Transport Management
- Vol. 137
Archetype-based analysis of single-pilot operations risk: Evidence from small jet incident and accident narratives
- 11 July 2026
- Published by Elsevier in Journal of Air Transport Management
- Vol. 137
Open AccessArchetype-based analysis of single-pilot operations risk: Evidence from small jet incident and accident narratives
- Open Access
- 30 April 2026
- Published by Elsevier in SSRN Electronic Journal
Aviation Accidents: Countering the Misinformation Dynamic
Aircraft accidents generate acute information scarcity and competing narratives. Speculation proliferates while verified facts remain sparse. Brandolini's Law formalizes this asymmetry: refuting false claims demands disproportionate effort. Drawing on the case studies of Ukraine International Airlines Flight PS752 and Germanwings Flight 9525, this analysis examines how misinformation anchors public understanding before investigations clarify causation, why rigorous corrections often fail, and how open-source intelligence occasionally interrupts this dynamic.Two contrasting outcomes emerge. Post-incident misinformation follows predictable patterns rooted in cognitive biases, platform amplification, and institutional delay. Brandolini's asymmetry is conditional, not invariant. Practical recommendations address three audiences: regulators, investigators, and safety professionals. The 72-hour post-incident window is critical. Rapid transparent communication and credible OSINT engagement can narrow asymmetry, yet current institutional practice systematically misses this window.
- 30 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessAviation Accidents: Countering the Misinformation Dynamic
Aircraft accidents generate acute information scarcity and competing narratives. Speculation proliferates while verified facts remain sparse. Brandolini's Law formalizes this asymmetry: refuting false claims demands disproportionate effort. Drawing on the case studies of Ukraine International Airlines Flight PS752 and Germanwings Flight 9525, this analysis examines how misinformation anchors public understanding before investigations clarify causation, why rigorous corrections often fail, and how open-source intelligence occasionally interrupts this dynamic.Two contrasting outcomes emerge. Post-incident misinformation follows predictable patterns rooted in cognitive biases, platform amplification, and institutional delay. Brandolini's asymmetry is conditional, not invariant. Practical recommendations address three audiences: regulators, investigators, and safety professionals. The 72-hour post-incident window is critical. Rapid transparent communication and credible OSINT engagement can narrow asymmetry, yet current institutional practice systematically misses this window.
- Open Access
- 17 February 2026
- Published by Elsevier in SSRN Electronic Journal
The Viral Deviation: A Quantitative Analysis of Social Media Engagement Metrics on Non-Compliant Flight Deck Behaviors
BackgroundNormalization of Deviance (NoD) is a well-documented organizational failure mode in aviation, yet its propagation through digital ecosystems remains unexplored. EASA highlights a critical mentorship vacuum due to pilot retirements, and social media platforms increasingly serve as informal training environments. This study introduces the concept of “Algorithmic Normalization of Deviance”, hypothesizing that recommendation algorithms systematically reward and amplify "gray-zone" cockpit behaviors that erode sterile flight deck protocols.MethodWe conducted a stratified netnographic analysis of 150 high-visibility aviation videos (N=150) across Instagram Reels, split between algorithmically recommended content (n=100) and curated influencer content (n=50). A novel coding framework operationalized compliance (Compliant vs. Gray Zone vs. Deviant) and incentive structures. Engagement metrics were analyzed to quantify the algorithmic advantage of non-compliant behavior.ResultsResults reveal a "Viral Deviation" effect: Gray-zone recordings (i.e., technically legal but procedurally borderline) comprised 57% of the sample and achieved significantly higher engagement than compliant content (d=0.24), particularly within the algorithmic discovery feed (p<.05). Structural incentive analysis demonstrates that monetization pressures correlate with increased gray-zone production, effectively creating a parallel, shadow curriculum that reinforces drift.ConclusionSocial media algorithms function as an unmonitored feedback loop that accelerates the normalization of deviance outside regulatory oversight. We propose a shift from prohibition to risk-stratified governance, recommending that regulators distinguish between handheld (high-risk) and mounted (managed-risk) devices, and integrate digital algorithmic literacy into ab-initio pilot training to counter this new systemic threat.
- 17 February 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessThe Viral Deviation: A Quantitative Analysis of Social Media Engagement Metrics on Non-Compliant Flight Deck Behaviors
BackgroundNormalization of Deviance (NoD) is a well-documented organizational failure mode in aviation, yet its propagation through digital ecosystems remains unexplored. EASA highlights a critical mentorship vacuum due to pilot retirements, and social media platforms increasingly serve as informal training environments. This study introduces the concept of “Algorithmic Normalization of Deviance”, hypothesizing that recommendation algorithms systematically reward and amplify "gray-zone" cockpit behaviors that erode sterile flight deck protocols.MethodWe conducted a stratified netnographic analysis of 150 high-visibility aviation videos (N=150) across Instagram Reels, split between algorithmically recommended content (n=100) and curated influencer content (n=50). A novel coding framework operationalized compliance (Compliant vs. Gray Zone vs. Deviant) and incentive structures. Engagement metrics were analyzed to quantify the algorithmic advantage of non-compliant behavior.ResultsResults reveal a "Viral Deviation" effect: Gray-zone recordings (i.e., technically legal but procedurally borderline) comprised 57% of the sample and achieved significantly higher engagement than compliant content (d=0.24), particularly within the algorithmic discovery feed (p<.05). Structural incentive analysis demonstrates that monetization pressures correlate with increased gray-zone production, effectively creating a parallel, shadow curriculum that reinforces drift.ConclusionSocial media algorithms function as an unmonitored feedback loop that accelerates the normalization of deviance outside regulatory oversight. We propose a shift from prohibition to risk-stratified governance, recommending that regulators distinguish between handheld (high-risk) and mounted (managed-risk) devices, and integrate digital algorithmic literacy into ab-initio pilot training to counter this new systemic threat.
- 7 January 2026
- Published by Taylor & Francis in Theoretical Issues in Ergonomics Science
- p. 1-24
The evolutionary de-crewing of airline cockpits: a historical and analytical perspective
- 7 January 2026
- Published by Taylor & Francis in Theoretical Issues in Ergonomics Science
- p. 1-24
The evolutionary de-crewing of airline cockpits: a historical and analytical perspective
- Open Access
- 24 October 2025
- Published by Elsevier in Journal of Air Transport Management
- Vol. 131
Prompt engineering in generative AI applications for commercial single pilot operations (SiPO): An emerging competency
- 24 October 2025
- Published by Elsevier in Journal of Air Transport Management
- Vol. 131
Open AccessPrompt engineering in generative AI applications for commercial single pilot operations (SiPO): An emerging competency
- 19 November 2023
- Published by Springer Nature
- p. 228-239
Extended Reality in Aviation Training: The Commercial Single Pilot Operations Case
- 19 November 2023
- Published by Springer Nature
- p. 228-239
Extended Reality in Aviation Training: The Commercial Single Pilot Operations Case
- Open Access
- 4 November 2023
- Published by Elsevier in Decision Analytics Journal
- Vol. 9
Artificial intelligence applications in aviation accident classification: A preliminary exploratory study
- 4 November 2023
- Published by Elsevier in Decision Analytics Journal
- Vol. 9
Open AccessArtificial intelligence applications in aviation accident classification: A preliminary exploratory study