Gains:
- Understand the principles of SMS and just culture and use artificial intelligence for trend analysis and system-oriented report drafting in security notifications.
- Ability to associate Dirty Dozen factors, especially complacency and automation bias, with the use of artificial intelligence and manage them as a shield
- Understanding that security data should be anonymized and that the decision/intuition/signature remains with the human.
Aviation safety is not only achieved by tightening the right bolt to the right torque; It is achieved through a culture that governs how work is done. In this unit, you will learn how to use artificial intelligence (AI) in the field of quality management, safety management system (SMS) and human factors; But we will discuss why human judgment should always remain at the center of security culture.
SMS and reporting culture
SMS (Safety Management System) is a systematic approach based on proactively detecting hazards and managing risk. At its heart are just culture (an approach in which the person who honestly reports a mistake is not punished, and negligence/intention and error are separated) and hazard reporting. A security system only works when people report it without fear.
AI is a valuable aid here, but it requires a delicate balance. Useful uses:
- Classify and trend notifications: Group recurring themes, ATA sections, conditions across hundreds of security/quality notifications.
- Report draft: Translating the facts of an incident into a structured report (with language that examines the process, not the person,).
- Facilitating root cause analysis: Generating questions for the “5 Whys” or barrier analysis.
- Quality audit checklist draft.
- Lesson learning summaries: Compiling recurring lessons from past events.
Caution: Security notifications are often personal and sensitive data. Anonymization and corporate data policy are essential when analyzing with AI. Additionally, AI should not be allowed to label an event as "whose fault"; The purpose of SMS is not to find criminals, but to fix the system. An AI used incorrectly can damage just culture.
Dirty Dozen: 12 triggers of human error
The classic of maintenance human factors is Gordon Dupont's Dirty Dozen list — 12 common human factors that lead to maintenance failure: lack of communication, complacency, lack of knowledge, distraction, lack of teamwork, fatigue, lack of resources, pressure, lack of assertiveness, stress, lack of awareness, norms (misplaced habits).
The most critical link regarding AI here is the relationship between complacency (overconfidence/complacency) and automation bias (overconfidence in automation - accepting it as true just because a system says it). AI's fluid and confident output can precisely breed complacency. The idea of "that's what the system said, no need to check" is a modern version of the Dirty Dozen.
Tip: Set up AI usage as a shield against the Dirty Dozen, not as a trigger: it's good to have a checklist reminder at the end of a tired shift; but skipping the physical examination by saying "AI already looked" is complacency itself. The tool should increase your attention, not numb it.
Assertiveness and AI
The "lack of assertiveness" in Dirty Dozen is a technician's inability to voice something he sees as wrong. Interestingly, AI has a two-way effect here. Positive: a technician can use AI to structure his concern and draft “how do I phrase this?” Negative: The authoritarian tone of the AI can lead the human to suppress their own true intuition. The right balance: AI is the input, your professional intuition and observation is the decision maker. Your sense of “something is wrong” must outweigh the AI’s output of “everything is normal” — stop and look.
three mini cases
Case 1 — Trend analysis found recurring error. When six-monthly quality notifications were grouped by YZ (anonymised), a recurring theme of “incorrectly installed fastener” stood out for a particular panel type. The quality team got to the root cause: a step in the task card was unclear. The card has been clarified; A recurring source of error has been systemically closed.
Case 2 — Automation bias trap detected. On a tired night shift, a technician wanted to skip a visual inspection based on the AI's summary of "this check looks normal." The team leader interjected: The AI had summarized the data but had not seen the piece. Physical examination found a small trace of leakage. Complacency was broken by human intervention.
Case 3 — Assertiveness support. One young technician thought a plan was unrealistic but didn't know how to say it. With AI, he structured his concern in factual, respectful language: “That task cannot be completed safely in the given time due to that prerequisite.” Structured objection considered; A step that would have been skipped under pressure was preserved.
Four copyable templates
Role: Security notification trend analyzer (with anonymised data).Task: Group recurring themes, ATA sections, and conditions in the following anonymized notifications.Rules: Blame person; DO NOT label “whose fault”; focus on systemic pattern.Notifications: [anonymous data]
Role: Incident report draft assistant (just culture language). Task: Draft an incident report from the following facts, examining the process, not the person. Rules: Do not use accusatory adjectives; Use the "what happened, what circumstances, what barrier was missing" structure. Facts: [event information]
Role: Root cause (5 Whys) facilitator. Task: Generate "5 Whys" questions for the following event and question whether each answer indicates a systemic weakness. Rules: I give the answers; you generate the right questions; Refer to the process, not personal blame. Incident: [summary]
Role: Assertiveness expression aid. Task: Suggest a sentence structure for me to express my concern below in factual, respectful and clear language. Rules: Don't be emotional/accusatory; center on security rationale.Concern: [technical concern]
Weak prompt / Strong prompt
Weak: "Who is at fault in this incident, write the report."
The culprit-seeking framework undermines just culture and obscures the true systemic cause.
Strong: "Write an incident report draft from the following facts. Don't blame the person; use the 'what happened, what circumstances were present, what barrier was missing, what can be fixed systemically' structure. Data anonymous, don't add personal information."
This prompt shifts the focus from person to system, from punishment to learning.
Table: Dirty Dozen and AI interaction
factor
AI risk
Proper use of AI
Complacency
"Already looked" complacency
Checklist reminder
automation bias
Blindly trust the output
Use as input, confirm
lack of communication
Incorrect summary transfer
Configure turnover
Lack of assertiveness
Authoritarian tone suppresses
Help expressing anxiety
fatigue
Shortcut at critical moment
Don't remind the steps, leave the signature to the person
Common mistakes
- Making AI ask the question "who is at fault?" Just spoils the culture.
- Analyzing sensitive notifications without anonymizing them. Privacy violation.
- Falling into automation bias. "The system said" does not replace the physical examination.
- Sacrificing one's own intuition to the AI's tone. Pay attention to the feeling of "something is wrong."
- Using AI to shortcut under pressure. The vehicle accelerates but does not skip a step.
In summary
Security is a culture and human judgment is at its heart. AI is a valuable aid in extracting trends from safety/quality notifications, drafting reports in just culture, and root cause analysis. But the biggest risk is that the fluidity of AI feeds complacency and automation bias. Build AI as a shield against the Dirty Dozen: increase your attention, not numb it; Leave the decision, intuition and signature to the person.
Application task
Get (or edit) a few anonymized quality/safety notifications. With the first template, ask the AI for trend analysis. Then ask yourself: Which themes did the AI find, which did it miss, did it slip into any "who's at fault" implications? Write a "systemic improvement" proposal and check that it complies with the just culture principle.
checklist
- [ ] I have anonymized the security data and complied with the data policy.
- [ ] I used AI in a system-fixing framework, not a criminal-finding framework.
- [ ] I did not skip the physical examination against automation bias.
- [ ] I put my “something is wrong” intuition above the AI output.
- [ ] I kept the just culture language in the report.
- [ ] I left the decision, interpretation and signature to the person.