As organisations embrace AI summaries of documents to save time, implementing a traffic light system for documents could provide a green light to cutting through red tape.
The rapid and widespread integration of AI Large Language Models into our daily lives has fundamentally changed how we consume information, with AI being used to summarise everything from reports and vessel notifications to IMO regulatory updates and meeting summaries. While the cognitive labour saved from distilling a multi-page charterparty or a technical manual into a series of neat bullet points, is important, we are risking a dangerous erosion of due diligence.
AI is an exceptional tool for synthesis, but it is not an adequate replacement for professional and technical judgment. The risk isn’t just about ‘hallucinations’ (where AI invents facts), but also the likelihood of missing specific contractual obligations, tone or nuance that fall away during the compression process. While this won’t be an issue every time, there will be instances in which the consequences for getting it wrong could be expensive or even catastrophic.
A simple traffic light system (red, amber and green) of document governance may provide an easy way in which to support human decision making around document interaction.
Red light for high risk
There are certain documents where the ‘big picture’ simply isn’t enough. When a file is designated as red, it signifies that the document carries significant legal, safety, or operational risk. In these instances, relying on an AI-generated summary would be considered a breach of professional responsibility as the specific wording of the document is highly relevant.
A summary, by definition, must omit details to achieve brevity. In a high-stakes salvage agreement, a complex Bill of Lading, or new MARPOL amendments, those ‘details’ are often where the actual liability lives. A summary might tell you that a document is ‘standard,’ while the original text contains a specific indemnity clause or a geographical limit that changes the vessel’s risk profile entirely. For red-category documents, the mandate is clear: read the original text. The AI may be used to help navigate the document, but details will need to be verified as specificities are relevant. The final interpretation must come from a human eye scanning the primary source of information.
Amber approach for informed verification
The amber or yellow category represents the middle ground which includes the vast majority of technical reports, white papers, vetting feedback, and internal safety management policies. Here, AI summaries are incredibly useful for triaging information, but should never be the final word.
When working with a yellow document, the summary serves as an outline for the topic. It allows individuals to identify which sections are most relevant to their operations without needing to dedicate hours reading information that is not pertinent. However, once relevant sections are identified, the reader must return to the original text to confirm the exact application of details. In some cases, users may benefit from asking the AI to flag any mentions of updates or revisions to existing rules, so that they can then check the original text to see the details of what they need to incorporate into their thoughts and actions.
Going green for efficiency
Documents that would involve the green traffic light are ones where the details are largely expected to be understood by the team, or where the overarching message is the only critical takeaway. General maritime industry news, high-level circulars, or internal team announcements often fall into this category.
For these documents, the risk of missing a minor detail is outweighed by the efficiency gained from a rapid summary. Here AI is able to cut through the administrative burden without increasing the risk to any organisation.
Bespoke traffic light system
Implementing this documentation system cannot be left to individual discretion, particularly for those new to the organisation. We cannot expect a junior operator, new team member or a recently qualified cadet (all of whom are focused on providing value at high speed) – to correctly assess whether a specific memorandum carries the weight of a red or yellow classification. Rather, individual companies must provide guidance for the types of documents that fit each of the categories, with this guidance linked to corporate governance and compliance frameworks.
Categorisation should be handled by senior stakeholders or automated via metadata tagging based on the document’s source and type. It may be useful to flag certain stakeholders – such as flag states, regulators, insurers, law firms, etc – as automatically falling into the red or yellow categories. Some companies can programme the AI summary to warn the reader that the document being summarised is from a named ‘red light’ organisation on the company’s internal list and thus the details must be checked – which may be a useful way to interrupt a habit of summarising text.
Embedding these guardrails is a governance must, in the same way that guidance is provided for other safety or risk related operations. Companies also have a vested interest in protecting newcomers from making high-stakes errors and ensuring that the workforce’s collective reliance on technology remains a useful, rather than a liability.
AI’s ability to summarise is one of its greatest strengths and can be a useful tool for the modern maritime professional. But as our sector embraces the speed of the summary, we must never forget the value of the source and the importance of human judgement when making decisions. After all, the devil is often in the details.
——–
If you want to get future digitalisation and AI in maritime news and insights delivered straight to your inbox, subscribe to our monthly newsletter.