Maritime workers face ‘verification tax’ on AI output, report says
Yet, almost none of their employers measure the cost.
Maritime professionals spend hours each week checking artificial intelligence (AI) output, and almost none of their employers measure the cost.
According to a new Marcura report, the ‘verification tax’ is the time and effort needed to check AI-generated work before anyone acts on it.
In a survey of 60 maritime professionals, 55% said they spend at least an hour a week checking or correcting AI outputs.
Yet 85% said AI still saves them more time overall once that checking is counted, and 73% said it lets them do things they could not do without it.
Asked how often they find errors, 47% said sometimes, 18% often and 8% almost always, with 22% saying rarely.
For 15%, an AI error has already caused a real financial, client, reputational or operational impact, and another 12% said one was caught only before it did damage.
"We don't have benchmarking data," said Rajesh Gopinathan, head of business transformation at Seaspan, which is now capturing data from existing processes so that future changes can be measured against time taken and quality of work.
The report argues the tax only becomes a problem when checking outweighs the benefit.
"If I can complete a task manually in ten minutes, but AI generates an answer in one minute that then requires fifteen minutes of checking, correction and source verification, there has been no productivity gain," said Abbas Alipanah Kordlar, senior AI development specialist at Digital Exchange.
Not all checking is waste, the report says. It separates ‘warranted’ verification, where an error could change a company's contractual position or affect sanctions or port-restriction decisions, from "residual" verification driven by distrust, opacity or unclear accountability.
Meanwhile, respondents are split on how far trust can substitute for checking. If an AI output had proved reliable over time, 47% said they would check much less or stop, whilst 38% would still do most or all of the checking.
On important business decisions, 70% said they would always check the AI's recommendation. Yet 60% would accept an occasional error on everyday tasks in exchange for the time saved.
Evidence from outside the industry suggests mature processes can eliminate checking. Ardent Partners found touchless invoice processing, with no human intervention, averages 32.6% across accounts payable functions but reaches 49.2% among top performers.
Their exception rates run at 14% on average, 9% for the best and 22% for the rest. However, the report cautions that the comparison is not directly equivalent to maritime AI workflows.
