Every maritime AI deployment sits on one side of a fundamental line: the system that advises a human who decides, or the system that acts on its own authority. It's the single most important architectural distinction in the whole stack — because it defines who is responsible when the system is wrong, and it's the line the regulation is drawn around.
This article is about that line: what decision support is, what automation is, why the distinction is the safety case's foundation, and why the products that blur it are the dangerous ones.
Key findings
- Decision support advises — the human decides, the AI informs. Automation acts — the AI decides within its envelope. Established
- The line defines accountability — a decision-support error traces to the human who decided; an automation error traces to the system's authority. Established
- The blur is the hazard — an advisory system that acts, or an automation that pretends to advise, is the worst of both. Maritime-GPT analysis
- Trust calibration differs by side — decision support needs the human engaged; automation needs the human supervising. Established
- The MASS degrees of autonomy formalize the spectrum — from decision-support through supervised autonomy to fully autonomous. Established
Two different products
A decision-support system — Voyage Intel Pro's weather analysis, a chartering copilot's recap, an engine monitor's anomaly flag — produces a recommendation a human acts on; its value is the quality of the advice and the human's authority is preserved. An automation system produces an action — the autopilot holds course, the autonomous system chooses a manoeuvre — and its value is the quality of the action. They're different products, designed differently, certified differently, and responsible differently. Established
The accountability line
The line matters because accountability attaches to it. A decision-support error — the advice was wrong — traces to the human who decided to follow it (or failed to verify it); the system was an advisor. An automation error — the action was wrong — traces to the system's authority; the human's role was supervision. The legal and safety architecture is built on which side of the line the system sits — and a product that blurs it (an advisor that silently acts, an automation that hides behind 'recommendation') is an accountability hole. Maritime-GPT analysis
The trust calibration differs
The human-factors demand is different on each side. For decision support the need is engagement — the officer must stay cognitively active, treating the AI as input to their judgement. For automation the need is supervision — the officer must monitor without over-trusting, ready to take control. A system designed for one side deployed as the other fails the trust calibration: the decision-support that encourages passive reliance, or the automation that demands constant active judgement, is mismatched to its role. Established
Where the products sit
- <strong>Pure decision support.</strong> Weather analysis, market intelligence, compliance retrieval — advice with no actuation. Established
- <strong>Advisory-plus.</strong> Route recommendations, watchkeeping alerts — advice the human applies to the ship. Established
- <strong>Bounded automation.</strong> Autopilot, engine governors — acting within a defined function. Established
- <strong>Supervised autonomy.</strong> MASS operations — the system decides within its envelope under remote supervision. Established
What we know
- Decision support advises; automation acts — the authority is the difference.
- Accountability attaches to which side of the line the system sits.
- Trust calibration is different — engagement for support, supervision for automation.
What we don't
- How to certify hybrid systems that blur the line.
- Whether advisory systems drift into de-facto automation through reliance.
- The regulatory treatment of systems that both advise and act.
Being tested
- Clear authority-boundary design in maritime AI products.
- Reliance-pattern measurement for advisory vs automated systems.
- Hybrid-system certification approaches.
Maritime-GPT analysis
The decision-support/automation line is the maritime AI stack's foundation — not because the technology differs, but because the accountability does. Everything we build stays on the advisory side deliberately: Voyage Intel Pro's analysis informs the master who decides, and that boundary is the product's safety case. The industry's honest architecture keeps the line explicit — the system that advises is one product, the system that acts is another, and the ones that blur them are the ones that produce casualties. Maritime-GPT analysis
This is a research feature, not operational guidance. Simulation results are reported as simulation; demonstrations are not deployments; company claims are attributed as such. Verify safety-critical decisions against official nautical publications and flag-state requirements.
Sources & evidence
The human-factors foundation for the advise/act line.