A plausible answer
A generative AI model can examine the track and describe a likely destination. It may even offer a convincing explanation, but the relevant facts are not systematically selected, checked or tied to their effect on the prediction.
The KnaiTai Open Initiative
KnaiTai
Knowledge-native AI.
Open, by design.
Making knowledge programmable for AI.
Maritime behavior understanding
A movement model can predict where the vessel is heading next. But which port is it actually trying to reach?
The recent track may point toward several plausible ports. Choosing among them requires more than movement data. The system may need to know the vessel’s size and draft, which ports can accommodate it, whether its route follows a known shipping corridor, and where the vessel has called before.
Knowledge added as context
A knowledge graph retrieval system can find information about the vessel, nearby ports and the region. That information is supplied to the model as additional context, but it may still be unclear which facts mattered and how they changed the answer.
Knowledge participates in the decision
KnaiTai selects the knowledge needed for this specific prediction and gives each item an explicit role: ruling out an unsuitable port, strengthening a likely destination or qualifying the model’s confidence.
Different kinds of knowledge help resolve the destination.
The example makes visible what each kind of knowledge contributes beyond the vessel’s observed movement.
Where is the vessel heading?
Its recent path and direction establish the initial set of plausible destinations.
What does the vessel require?
Vessel type, size, draft and berth requirements help eliminate ports that cannot receive it.
Which destinations are feasible?
Port capabilities, shipping corridors, coastal geography and protected waters shape the realistic alternatives.
What has this vessel done before?
Previous port calls and recurring behavior can strengthen one destination over otherwise similar alternatives.
The point of Knowledge-native AI: knowledge does not merely surround the model as retrieved text. It becomes an explicit and inspectable part of how the model reaches its conclusion.
(Stay tuned for KnaiTai demonstrations, tools and open-source releases.)
Assimilate
Subset
Compile
Converse
Verify