AI in maritime safety management should not be judged by how much work it can perform without a human.
The more useful question is: which parts of the work are structured, repetitive and evidence-heavy enough for AI to accelerate - while leaving material decisions under accountable human control?
1. AI should understand the controlled context
A generic chatbot can answer questions about maritime topics. An operational compliance assistant needs to understand the company's controlled SMS, vessel particulars, applicable requirements, crew context and relevant records.
Without that context, an answer may sound useful while being disconnected from the actual system of work.
2. AI can reduce the search and interpretation burden
DPAs spend time locating procedures, comparing revisions, identifying relevant requirements and connecting records. AI can help retrieve source-backed answers, summarize relevant material and surface relationships across the system.
The objective is not to hide the source. The source should remain available so the responsible person can verify the reasoning.
3. AI can draft - humans approve
AI can identify a potentially affected SMS procedure and draft a proposed revision. It can prepare a familiarisation course, summarize an audit report or assemble a scoped evidence pack.
The authorized person should remain responsible for reviewing and approving material compliance decisions.
4. AI should perform bounded tasks
The safest product pattern is not unlimited autonomy. An AI agent should have a clearly defined task, access to the appropriate controlled context and an explicit output that can be reviewed.
Examples include comparing controlled SMS revisions, summarizing a regulatory change, identifying potentially affected vessels, drafting an SMS change, preparing targeted familiarisation and assembling relevant evidence.
5. The real opportunity is the compliance graph
Over time, the most valuable intelligence layer is not the chat interface itself. It is the structured relationship between requirements, SMS procedures, vessels, people, actions, knowledge and evidence.
Once those relationships exist, the system can help answer: What changed? Where does it apply? What needs attention? Who needs to know? What is still missing? What can we prove?
The AI handles retrieval, reasoning support and bounded preparation. The human remains responsible for the decision.
Conclusion
The future of AI in safety management is not "AI replaces the DPA." It is a system in which the DPA spends less time reconstructing information and more time making accountable decisions.
The AI handles retrieval, reasoning support and bounded preparation. The human remains responsible for the decision.