When AI flags 'routine': the judgment gap that watchkeeping training must close
AI watchkeeping tools can reduce cognitive load on the bridge — but they also introduce a new failure mode when officers defer to the algorithm over their own observation.
THE NEWS
According to The Maritime Executive, a recently published op-ed raises a pointed operational question: when an AI-assisted watchkeeping system classifies a contact as routine, and a junior officer's independent observation suggests otherwise, what is the correct course of action? The piece frames this not as a technology question but as a training and command culture question — one the industry has not yet answered systematically.
The op-ed does not report an incident. It poses a scenario to surface a structural tension in how AI decision-support tools are being integrated into bridge operations. The central concern is that the presence of an algorithmic alert — or the absence of one — may be subtly reshaping how watchkeepers weigh their own perceptions, potentially in ways that current competency frameworks do not address.
The argument, as described, is that AI alerts should function as a trigger for judgment exercises, not as a substitute for them. The implication is that maritime training programs need to evolve to treat AI outputs as one input among several, rather than as a default arbiter.
WHY IT MATTERS
For the Brazilian offshore sector, this discussion arrives at a moment of practical relevance. Vessels operating in the Campos and Santos basins — supply boats, anchor handlers, platform support vessels, and shuttle tankers — are increasingly equipped with decision-support systems that process AIS data, radar returns, and environmental inputs to assist watchkeepers. The question the op-ed raises is not hypothetical for these crews.
The core tension is one of automation complacency, a phenomenon well-documented in aviation and now migrating into maritime operations as bridge technology matures. The risk is not that AI systems malfunction — it is that they perform well enough, often enough, that officers begin to treat their outputs as authoritative rather than advisory. A junior officer who has seen the system correctly classify hundreds of contacts may hesitate to override it on the occasion when their own observation diverges. That hesitation is the failure mode.
This has direct implications for how Brazilian maritime training institutions and operators structure bridge resource management (BRM) curricula. Current IMO STCW competency standards were developed in an era when the primary concern was ensuring officers could operate without automation. The emerging challenge is ensuring they can operate critically alongside it — distinguishing between situations where the algorithm's confidence is well-founded and situations where independent seamanship judgment must take precedence. These are different cognitive skills, and they require different training scenarios.
For operators with large Brazilian-flagged fleets — including those managing offshore support vessels and shuttle tankers under long-term contracts — there is a practical crew management dimension here as well. If AI watchkeeping tools are deployed inconsistently across a fleet, officers rotating between vessels will encounter different decision environments. A watchkeeper trained to trust one system's alert logic may apply that same deference to a different system with different calibration, different sensor inputs, or different confidence thresholds. Standardization of how AI outputs are communicated to crews, and how crews are trained to interrogate those outputs, becomes a fleet management concern, not just a technology procurement one.
The regulatory dimension is also worth tracking. NORMAM and the broader IMO e-navigation framework are still developing the standards that will govern AI-assisted bridge systems. Brazil's maritime authority, through its participation in IMO working groups, will have input into how these standards take shape. The op-ed's framing — that AI alerts should be treated as judgment drills rather than verdicts — is consistent with the direction that several flag states and classification societies are taking in their guidance documents. Brazilian operators and training centers that engage with this framing early will be better positioned when formal requirements arrive.
Finally, there is a human factors dimension specific to the offshore environment. Unlike deep-sea voyages, offshore support operations in Brazilian waters involve high-traffic, constrained maneuvering areas around platforms and FPSOs, frequent DP transitions, and watch patterns that can accumulate fatigue. These are conditions that may amplify the tendency to defer to automated systems precisely when independent judgment is most needed. The offshore context does not make the op-ed's argument more alarming — it makes it more concrete.
CONTEXT
The broader industry conversation about human-machine teaming on the bridge has been accelerating since several classification societies began publishing guidance on autonomous and semi-autonomous vessel operations. The question of how to preserve — and test — the judgment capacity of officers working alongside AI systems is appearing with increasing frequency in maritime safety literature and in flag state policy discussions.
For Brazilian offshore professionals, the most analogous precedent may be the evolution of DP competency standards, where the industry learned over time that operators needed not just technical proficiency with the system but the ability to recognize when the system's behavior warranted manual intervention. The watchkeeping AI question follows a similar logic: the technology is a tool, and the measure of its safe deployment is whether the humans working with it retain the capacity and the confidence to override it when the situation demands.
Source: THE MARITIME EXECUTIVE