In the case of autonomous surface vehicles, artificial intelligence is more likely to be used in the algorithm of the systems of these types of ships.

yep, Just like the subject of many scholars studying now, applied DL and RL to MASS technology research such as maritime traffic flow prediction, navigation situation estimation, ship anomaly behavioral identification, collision trajectory prediction, behavioral decision-making and path planning and so on.

AI includong DL and RL could play a crucial role in maritme traffic flow prediction by applying certain algorithms and adding reasonability to expert systems. We just were yesterday at Continental ContiTech coorporation where they produced high velocity and pressure materials that could assist in transport of heavy items above the sea surface and applying autonomous systems could be an innovation.

would Artificial intelligence be restrained or regulated by another Artificial intelligence; to prevent take over or hijacking? What would be the ethical, moral, engineering provisions to sand box the platform properly? Is the question in security, control or efficiency/accuracy?

Is there still an operator involved and if so, does he/she understand the choices made by the algorithm ?

AI is the promising alternative for Ocean Engineering. Classification or Regression issues handled by using Deep Learning algorithms, hyperparamters automatically optimized by RL etc. For my concern, deep unsupervised learning or deep semi-supervised learning with RL are good options. But the critical thing or the first thing is that could you get such number of data or find the suitable problem. As I investigated, for normal concern or limited number of data, the traditional machine learning is superior. Additionally, there has been lots of open source algorithms. If determining the best one, we should be first familar with our concern (data) and know the theory about the algorithm. And then just try it and start to tune the hyperparamter (the most boring and tough thing).