We'd better be ready for AI

A global race is under way to turn artificial intelligence into an economic and strategic advantage. With the US scaling frontier-model capability at speed, China pushing scale and deployment, and the European Union trying to pair regulation with a late but sizeable infrastructure push, the UK is debating not just AI strategy, but where the compute, energy and investment conditions will come from and who will control them.

 

For engineering sectors such as naval architecture, this matters because access to compute and trustworthy models will increasingly shape who can innovate quickly, simulate more, and de-risk decisions earlier. In this context, early-stage ship design can be seen as a case in point, with the real prize being not AI-generated geometry per se, but verifiable AI embedded in professional workflows.

 

Numerous studies have shown that early design stage choices have a significant effect on cost, carbon footprint and operability. Yet, they are made under uncertainty and time pressure. Nowadays, decarbonisation with new fuels and tighter coupling between hydrodynamics, structures and operations requires better exploration of the design space and proper multi-objective optimisation as early as possible. AI can help through generative models that propose plausible hull candidates and surrogate models that screen performance quickly. However, the outputs should remain traceable, repeatable and easy to interrogate within existing CAD/CAE workflows.

 

This can significantly accelerate iteration and widen the breadth of exploration. Diffusion models and other generators can sample learned design spaces while multimodal representations can improve geometric robustness and interpretability. In parallel, physics-informed learning is reducing data hunger in some settings. However, verification is still needed. The barriers are data availability, reproducibility, clear limits of applicability, and integration into review processes, including class-facing evidence and documentation.

 

This will have direct implications for education and professional identity. Tomorrow’s naval architect will be expected to interrogate training data, recognise when a model is extrapolating beyond its competence or hallucinating, and translate AI-generated options into evidence that survives design review. In other words, “good judgement” will increasingly include knowing when not to trust automation, and how to fall back to first principles and high-fidelity tools.

 

NAOME academics and researchers are developing a range of maritime AI models for design, operation and safety. Geometry representation is a decisive factor for trustworthy generative hullform design. Our recent framework, developed with support from MarRI-UK, learns a shared latent encoding of hull geometry from complementary modalities familiar to naval architects, using surface point clouds, waterlines and buttocks. It couples them with conditional latent diffusion to generate novel, geometrically coherent hulls consistent with high-level design parameters. The benefit is practical controllability and robustness, as the generated candidates are easier to interpret, check and refine than outputs produced from a single representation.

 

Looking ahead, there is a clear trend to move from AI as a plug-in to AI-native design toolchains, linking requirements, generated geometry, fast screening, high-fidelity analysis and the design review dossier. Verification should be built in, including automatic geometry and constraint checks, uncertainty signals and clear limits of use that accompany each candidate. The next step is performance-aware generation, where diffusion models and surrogates could enable estimation of resistance, seakeeping, emissions-related outcomes and other performance metrics, with targeted CFD/FEM for confirmation. These will only scale if policy, class practice and education keep up with how tools are validated and taught.

 

Author profile

Evangelos Boulougouris, head of the Department of Naval Architecture, Ocean and Marine Engineering (NAOME), University of Strathclyde, Glasgow, UK

 

This article appeared in Opinion, TNA May-June 2026