Summary of major AI problems + simple fixes to improve them

[TK3600]

Before I start, I just want to say the combat AI is generally decent. Not as good as humans, but provides sufficient challenge to the player. When I fight my own ship designs piloted by AI, I struggle. For instance, if I were to give AI an early dreadnought while I myself use a semi-dreadnought, I often find myself losing the 1v1. I consider the AI in this regard to be good enough. Yet when I play the campaign, I always stomp the AI even in places I should not. I could reliably take out a contemporary CL using a DD, despite the armament suggest I am outclassed. I can take out a BB with 3 CL, even though the said BB is 10 times cost of my CL. The difference is AI ship design is a major detriment, less so how the ship is utilized tactically by the AI. This is an important point. While I see combat AI is enhanced constantly in the patches, I rarely feel the ship design AI improve. By improving AI ship design the game would vastly improve.

Problem 1: AI use outdated large caliber guns

This is entirely introduced by particularities in tech tree. Calibres 14inch and up are unlocked as mk1 first. Humans knows to avoid them til higher marks, AI do not, hence problem. It makes no sense for new calibre to have 1890 muzzle loader stats, when we are in WWI era. If we want high calibre muzzle loaders, they ought to be unlocked earlier.

Solution:

• I advocate those weapons to be unlocked starting from MK3. For balance purpose, you could increase their research time 3x, to make up for skipping mk1 and 2. mk1 and 2 techs of high caliber guns removed.
• See this thread for example of a mod that fix it:

Problem 2: AI bias toward latest modules as opposed to most useful

Not every new component is a linear upgrade of each other. Some are niche techs like pictric acid is newer than gun cotton, but is not always the better choice. It greatly sacrifice AP performance in favor of explosive damage. Unless the AI is taught to make particular builds around pictric acid using particular hull types, it should avoid niche components, but bias toward components 'good in most builds'. Alternatively, different hull types(ex: DD, vs BB builds) should have different bias toward each type of components, but that takes more effort.

Solution:

• Bias AI into best general purpose components as opposed to newest = best
• we can leave some random chance for AI silliness, but it should not be the norm.
•  See this thread for example of a mod that fix it:

Problem 3: AI poor tech management

AI for some reason always fall behind in tech, despite not having reason to. All you need is to not spam tech priority, and max out the slider. It is very simple to do. Being behind in tech is a very severe detriment.

Solution:

• AI should prioritize to max out research slider at all times(or like 99% of times). Building ship is useless if they are outdated. One may argue AI might build less ships if that is the case. Perhaps more ships of lesser quality can win? That is not correct. AI overly aggressively scrap ships, so they will never have a 'build up'. Therefore wasting too much money building ship is not that helpful.
• The only exception I can think of to not max the tech is when country just suffered an economic collapse like war, and unable to maintain existing fleet, yet cannot scrap the fleet because it is in another war. That is the only time research slider can go down.
• AI should not utilize more than 1 priority at a time. This makes them more up to date. It is counterproductive. Ideally, no priority should be assigned. However I concede that AI country needs some flavor so they do not need to meta game 100%. Only 1 prioritization should be assigned to AI at any one time. They may queue which one to prioritize depending on year, for flavor purpose.

These changes should take less than 1 week to implement, and greatly enhance AI performance.

There are other problem of AI that severely handicap their design, but may require more than simple fixes of the above 3. Examples include terrible weight balances of AI ship designs, irrational armor distribution schemes, poor barbette placements. From my glance of game files, those are either not accounted for at all, or is done at a model per model basis. Nevertheless, those are important to fix eventually, but we have to be patient about it. Terrible weight balance in general debuff AI accuracy by roughly 40%. Irrational armor scheme makes ship cost 3x of what it should, yet still have weak spots.