In 10 years??

Nothing in that post (except the AI) is real...it is all fake... it is all fantasy. If you want to mastrabate with an AI, please take it private and try not to kill anyone on the way.

 

Well for AI in boat design, of course there is nothing functional to be expected from a neuronal network like midjourney that only relates 2D visuals to whatever subtitles people used to write next to them: if that is how it was trained, no wonder it confuses bow and stern in half the pics it generates. We'll start talking if someone gets an AI to analyse 3D models linked with relevant figures of merit and detailed specs, as researchers are already experimenting with in architecture (for basic tasks like office layouts in large buildings). No problem for the AI, but in the marine field i rather doubt there will be large and consistent enough databases to chew on for training anytime soon.

Where i think we shouldn't underestimate AI is in acceleration of CFD solving (perhaps also for highly detailed FEA, like composite stuff with shell elements and so on). That too is already done, but not yet on multiphase flows with moving or overset meshes afaik. The idea there is simply to zero in on a fairly likely 'predicted' solution before starting to iterate around like mad. The speedups from the approach alone are staggering, but it also allows to calculate mainly on much faster massively parrallel GPU's (whose progress still follows Moore's law more or less) instead of the classic CPU's that don't evolve all that quickly anymore.

And perhaps, if we are very lucky, as CFD trickles down from Clusters and Workstations to gamer-level computers, some big player like Google might finally put in the effort to make meshing human-friendly. AI might be helpful for that too (there is research already), but the main point would simply be scale: if cost can be spread over hundreds of thousands of engineers and students, investment can be substantial. We might see the end of the current shitshow with all these geek sorcellery micro-companies competing in obscurity with their quirky 80's GUI's, sometimes swallowed by larger CFD solver corporations who'll sit on it and lock the code and developers away even more. With meshing currently drawing/wasting 70-90% of the human work input for CFD, that would not only bring the cost down to a fraction, but also make the whole subject of CFD much more accessible to non-specialists with a wealth of practical experience to check their simulations against. Like, well, naval architects.

Iteration time matters in design: If i can have detailed and accurate lift/drag/pitch etc simulations in minutes for negligible marginal cost, perhaps even in a seaway instead of flat water, i can move from incremental developemnt guided by a few regression formula that work solely on comparable shapes to a much more open design space, leaving a bit more room for breakthroughs. Certainly i can also run more versions to check the quality. Most importantly i can learn quickly in a playful way.