Gains:
- Being able to distinguish where artificial intelligence saves time in the maritime workflow (route, fuel, weather, documentation, machinery, legislation) and where maneuver and safety decisions are left to the captain and chief engineer, depending on the risk level.
- Ability to apply a discipline that verifies each artificial intelligence output through the steps of connecting it to the source, confirming it with an independent tool, and passing it through expert interpretation.
- Ability to understand that, due to the safety-critical nature of maritime, artificial intelligence output does not replace competent expert approval and that the final responsibility lies with the human.
You are on the bridge of a container ship. In front of you is an electronic map (ECDIS - Electronic Chart Display and Information System, a digital navigation screen that replaces a paper map), radar on one side, and a route recommendation sent by the weather forecast service on the other. An artificial intelligence (hereafter AI — the computer's ability to learn patterns from big data and produce predictions and suggestions) produced this suggestion: "Shift your route 12 miles south, save 18 hours, save 9 tons of fuel." The suggestion is tempting. But on that route there is another ship, a shallow and a fishing fleet. Who will approve this proposal? The answer is one and the backbone of this module: captain. AI calculates, suggests, signals; But the person who is legally responsible for the safety of life, property and environment at sea is the captain, and this responsibility cannot be transferred to any algorithm.
This module was written for deck officers (captain, first officer, navigational officers) and marine engineering engineers (chief engineer, second engineer, engine officers). The aim is to teach AI to be used end-to-end, from route and fuel optimization to weather evaluation, from navigation and cargo documentation to machinery monitoring and predictive maintenance, from safety regulatory research to bridge decision support; but at the same time, it is to make the AI understand where it should stand and why its output must be verified.
Why is maritime a special field?
Maritime is one of the safety-critical sectors (where an error can lead to loss of life, environmental disaster or millions of dollars in damage) where the cost of error is highest. A land vehicle may stop on the side of the road; A 300-meter tanker cannot stop, cannot turn immediately, or slow down before moving a few kilometers. The team is away from the nearest port for days; A wrong decision cannot be corrected immediately from the outside. Moreover, the field is tightly regulated by an international body of rules — SOLAS (International Convention for the Safety of Life at Sea), COLREG (rules for preventing collisions at sea), MARPOL (convention to prevent pollution of the seas from ships), ISM Code (International Code of Safety Management).
Therefore, the role of AI in shipping is clear from the beginning: AI is a decision support tool, not a decision maker. In safety-critical areas such as engineering and navigation, AI output is not a substitute for approval from a competent expert (captain, chief engineer, class surveyor). AI; It is an assistant that processes big data, produces scenarios, writes drafts, and marks patterns. The last word always belongs to the person.
Caution: "The AI said so" is not a defence. The question to ask in a post-accident investigation (marine accident investigation, P&I club, flag state) is: “With what independent source did the captain/engineer verify this recommendation?” If there is no verification record, the responsibility lies entirely with the human — the AI bears no burden.
Where does AI save time at sea, where does it leave it to humans?
To position AI correctly, divide the work into two: expeditable work and human decisions.
- Can be accelerated (AI helps with confidence): comparing multiple weather forecast scenarios, fuel/time estimation of route alternatives, drafting logbooks and port papers, flagging anomalies in machinery sensor data, drafting finding relevant clause in long regulatory text, extracting patterns from maintenance history.
- Human (AI makes suggestions, not decisions): final approval of the route, how many miles to pass a ship (COLREG decision), whether to stop the machine, escape to port shelter in case of dangerous weather, whether a maintenance can be postponed, emergency manoeuvre.
This distinction will be repeated in each unit throughout the module. Just because the AI shows a recommendation as “safe” does not prove that the recommendation is safe at sea; it merely offers a hypothesis.
Validation discipline: three-step filter
Before dumping each AI output into the sea, pass it through three filters:
- Connect it to the source. What data is the recommendation based on? Is the weather data up to date, which model? Which map (ENC - Electronic Navigational Chart, official digital map) is the route recommendation based on? Output of unknown origin is unverifiable output.
- Confirm with independent tool. Check the route recommendation in ECDIS for no-go areas (shallow/dangerous area that should not be crossed); fuel estimation with ship performance curve; machine anomaly by a second sensor or manual measurement.
- Get expert commentary. Even if the number is correct, does it make sense in a maritime context? If "18 hours of gain" requires passing through a storm front, that gain is bogus.
Tip: Tell the AI "mark where you are not sure and write down what data needs to be verified" with each request. A good output confesses its own uncertainty; Requiring this makes verification easier.
three mini cases
Case 1 — Unverified route. On a bulk carrier, the officer of the watch enters the "shortcut" route suggested by the AI without checking it in ECDIS. The route passes over a bank whose depth changes with tide; In low water, UKC (under keel clearance) drops to almost zero. The captain realizes it at the last moment and corrects the route. The AI had calculated the time correctly; but depth and tide confirmation was not made. Lesson: speed gains cannot trump security validation.
Case 2 — Malfunction caught early. On a tanker's bow engine, YZ marks cylinder exhaust temperatures when one cylinder deviates from the others by 30°C. The chief engineer manually verifies the indicators, checks the fuel injector and finds a clogged injector. It is replaced before the port, so there will be no problems during navigation. AI pointed out; engineer diagnosed and repaired it.
Case 3 — Fake legislation article. An officer asks the AI about ballast water rules; The AI produces a convincing but non-existent “item number” (hallucination — information that the AI fabricates as real). If the officer wrote this in the report without verifying it, there would be serious problems in the audit. No matter when compared to the official IMO text. Lesson: in legislation, AI only tells “where to look”; You read the provision from the official text.
Weak prompt / Strong prompt
Weak prompt:
Give me the best route.
"Best" is undefined; ship type, cargo, weather, no restrictions. AI produces a generic and insecure answer.
Powerful prompt:
Your role: senior navigation officer advisor. Ship: 180 m bulk cargo, draft 10.5 m, service speed 13 knots. Route: Gibraltar -> Piraeus. I have a 72 hour wind/wave forecast (below). Task: Draft 2-3 route alternatives; For each write down the estimated time, fuel and highest wave height experienced. Constraint: I will confirm UKC and no-go areas in ECDIS; you note "confirmation required" for shoals/tide. Tick any number you are not sure about.
The context, constraint, and validation request make the output both useful and secure.
Role and responsibility table
Quest
Role of AI
ultimately responsible
Generating route alternatives
Scenario, prediction
captain
Route confirmation and UKC
Warning/draft
Captain/navigational officer
fuel optimization
account, suggestion
Captain/company
Machine anomaly detection
marking
chief engineer
maintenance decision
Priority recommendation
chief engineer
Regulatory research
Redirect to source
Officer + official text
COLREG maneuver
(Out of decision)
Watch officer/captain
Common mistakes
- Thinking that AI is the decision maker. AI suggests; The captain/chief engineer makes decisions and bears the responsibility.
- Not connecting the output to the source. Which weather model, which map, which sensor? Unsourced output cannot be verified.
- Mistaking a hallucination for reality. AI can invent a non-existent item, port rule, or depth; Confirmation with official source is required.
- Sacrificing security verification for speed. Time/fuel savings never outweigh UKC and no-go confirmation.
- Uploading confidential/commercial data to the vehicle indiscriminately. Freight manifest, route, commercial contracts are sensitive; approved vehicle and masking required.
In summary
Maritime is a safety-critical domain and AI is a decision support assistant here, not a decision maker. AI; saves time on route, fuel, weather, documentation, machine monitoring and regulatory research; but route approval, manoeuvring, engine stopping and emergency decisions rest with humans—mostly the captain and chief engineer. Each output must pass a three-step filter (attribute to source, verify with independent tool, pass through expert review). AI output is not a substitute for competent expert approval.
Application task
Describe your own ship type (or an imaginary ship). Have the AI draft two route alternatives using the "Strong prompt" pattern above. Then filter the output in three steps: write down the source of each number, indicate which secondary tool (ECDIS, performance curve) you will use, and list the “confirmation required” points that the AI flagged.
checklist
- [ ] I positioned AI as a decision support tool; I make the decision.
- [ ] I connected each output to the source (weather model, map, sensor).
- [ ] I confirmed the recommendations with an independent tool (ECDIS, performance curve, manual measurement).
- [ ] I verified the legislation/depth/rule information from the official source; I checked for hallucinations.
- [ ] I prioritized security verification over speed/fuel gains and protected sensitive data.