AI and the future of it's ship building capability

[Tousansons]

Something is bothering me since I started playing at the start of Alpha. What is the point of the random (and sometimes nonsensical) AI ship design? Does it add something of value to the game other than the "easy" way to make the game create ships? Is it here to stay?

With the current AI design, naval academy can be a dice roll not only because a naval engagement is often random, but because ship design are also random for your friendlies and the enemy. The puzzle nature of the mission lose some of it's meaning if we can win due to the AI ship design doing some mistakes.

Custom battles are the same, we can't really try out cool ships if the enemy often create some unbalanced ones against them. Testing in general is also more time consuming because of that. 

How will the campaign AI work? With the current designer, is the AI able to create some sort of uniformity and "doctrine" of ships in a fleet? Can it keep the pace during 40 years? Finally, without some kind of pattern, adapting player design to fight efficiently against AI ones will be hard.

Right now, I'm still not impressed by AI design and think it is detrimental to the game, even after several patches and improvement. There is only a few cases where procedural and random add something in term of gameplay and in my opinion enemy generation is not one of them.

A solution for the long term developpement of the game is to abandon the random AI designs and settle for something simpler and easier to balance. A pool of designs per types and nations that can be expanded as the game improve, either by developpers or players input. I think this solution will add flexibility in both difficulty selection and interesting enemy ships in campaign and naval academy.

Enemy fleet could be designed to work better together in term of speed, armament, range. Like the real navies tried to do.

And most importantly, it also could lead to more drastic change in the ship designer itself, allowing greater freedom for the player without the risk of the AI being unable to create something decent.
 

PS: I think this is more related to AI and it's general effect in the game than the ship designer itself. If some of you think differently, feel free to move the topic in the shipyard.

Edited March 30, 2021 by Tousansons

[Cptbarney]

I agree, while l don't mind the concept itself if the AI has nothing to choose from or not many good designs to pick and therefore has to make something that is fine in of itself.

Regardless i would prefer a system where the player and devs, can make and submit Ship designs and components ranging from misc like fencing, searchlights, crates etc, too casemates (both the hatches and the guns themselves), turrets, guns, hulls, towers, funnels, shafts etc. That way not only do you get the devs, who can just clap out designs quicker as they don't need to fiddle with the AI to make sure it is capable of making something like it at somepoint through loads of testing.

Also this will allow players like myself to add in my own designs or designs and modules that are heavily requested, but can't be addressed by the dev's due to a large workload and/or close deadlines. Plus allows us to pave the way for mod support as well.

So in short, players and the devs make designs and then submit them into a ship library, per nation, per class, per maybe even weight class, gun size etc as well if we want to get very precise with what the AI can and will pick in-general. This also means it will create ships that fit the nation more rather than a random assortment of parts that look, interesting ultimately break immersion when see the ship in-question.

Even if the AI can't be shunted, i still think the above would be a very nice addition especially for custom battles as that would add a huge amount of replayability (And also allow me to finally set up AI tournaments with the forumites on here and discordians on the server too).

[Skeksis]

...

Edited November 9, 2021 by Skeksis

[Cpt.Hissy]

Can i dream a little?

It could be great to incorporate some kind of machine learning into ship generator. Make the random pick and stick to some values in response to player's actions, artificially make it be reluctant to accept changes as real navies were.
Then it could, starting off of a premade library or semi-randomly generated layouts, generate it's own doctrines (by picking a set of core parameters and be unwilling to change them in new generations of designs), and in the end provide different challenge every time you play.

Yes i know true machine learning takes stupid amounts of time to do something meaningful. But it could be dumbified to a system of sanely limited number of possible options and a set of synergy rules for them.

For me, in a game of this kind, making opposition "balanced", and specifically tailored to be obvious in how to beat, will not improve it, but kill the replayability and potentially make the game too boring to play even once. But that's just me, i dropped so many games that the crown considers good.

Now, that current academy has serious issues due to random enemy generation is a fact. These missions, if they were meant to be sort of tutorial, should've use standardized or fully premade enemies and starting positions. But, i think, they are not a tutorial in fact, but hastily thrown together "stuff" for customers to occupy themselves, while they work on the actual game. If that's what it is and if they're able to do it right we should see with the release of first campaign.

[SonicB]

1 minute ago, Skeksis said:

In short ‘Replayability’.

In theory we wouldn’t see the same enemy designs with each playthrough. Procedural generation systems are vastly superior to one dimensional/static pre-built enemies. All this work is for our benefit and it produces alternate histories, to play the 'what if', better since we already know what history was.

I'd suggest this is a pretty good argument for adding successful player-created designs into the AI repertoire, at least for custom battles.

[Cpt.Hissy]

"expanding library of player created designs"
is exactly how "From the depths" works. (game's totally unrelated so i think it's safe to mention), and it is very effective.
This approach so far had one major drawback though. So called power creep, the campaign set of designs grows stronger and stronger over time, as core community increases their skills and finds old ones way too weak, then game gets updated, old designs need to be adapted to the new version and they get buffed alongside.
But that game keeps seeing very drastic core mechanics changes, in something like UAD this may not become too big of an issue.

Edited March 30, 2021 by Cpt.Hissy

[disc]

Another positive value to procedural generation is that the enemy will build ships appropriate to its research level and budget in the campaign. So, if a nation has Mark 3 12in guns, coal fuel, reinforced bulkheads 1, geared turbines, etc. it can combine these into a design.

Player-designed ships would be nice too, but it would be harder to match them to campaign.

[Skeksis]

...

Edited November 9, 2021 by Skeksis

[madham82]

Like @disc stated, I think campaign won't be as crazy as we see custom battle. That said, custom needs some displacement limits up front to help balance better. For example, you set your fleet's max displacement and the enemy's. Then that determines how much you can use in total (not just for 1 ship). This would allow you to keep the existing number selections for each type of ship.  Want 1 super BB and against 2 AI equaling the same total weight as your super...done. 

Either that or custom should allow complete control over all forces (friendly and enemy). 

[Tousansons]

In ditching AI random ship design, we can still improve "replayability" by making player choices matters more than a random generator. Be it by diplomacy, events, technology, ect.

I believe there is more interesting ways to create content in a game than a random number of guns and armor thickness in a hull. 

 

 

[SiWi]

40 minutes ago, Tousansons said:

In ditching AI random ship design, we can still improve "replayability" by making player choices matters more than a random generator. Be it by diplomacy, events, technology, ect.

I believe there is more interesting ways to create content in a game than a random number of guns and armor thickness in a hull. 

 

 

except that a major part of the game is then to figure out the "one" combo you need for each timeframe to beat the designs you KNOW will come. With random design you are always "on edge" what your enemies are going to build. 

 

Personally I think a compromise could work:

you can enable "complete random AI designs" for campaigns and get just that.

Or you enable "historic blueprints" which would make the AI try to design ships roughly the sam as the real world ships.

 

I don't think it is feasible from an AI standpoint, to exspect that the AI "develops" doctrines or such.

In the moment I just hope that the game remembers the actual designs and not every battle create a different BB, even thou you already fought that one.

 

On a sidenote: "Navy strategy is building strategy" is a quote often associated with naval warfare especially since the late 19th and early 20th century. In this game however it would only work, if the player "knows" (or thinks to know) what the AI is building. On the other hand, the AI also should try to counter player builds.

A new Battleships with alot of relative smaller guns and thicc armor? build a counter ships with relative thin armor but bigger guns. (as example).

I hope that the AI will learn that at one point of the game.

 

Edited March 30, 2021 by SiWi

[Tousansons]

7 minutes ago, SiWi said:

except that a major part of the game is then to figure out the "one" combo you need for each timeframe to beat the designs you KNOW will come. With random design you are always "on edge" what your enemies are going to build. 

If the tech tree is just "better" techs allowing "better" parts, the player will build the "better" ships and in the end use the "best" combo with it's available technology against the percieved enemy menace. The type of AI ship design doesn't really come into play here, in the end the player have to build "better" if he want to win.

Real life navies usually tried to build ships with the idea of defeating their most likelly opponent. In a strategy game, intelligence of what the enemy is doing is an important part of game design. I doubt UA:D campaign will not allow the player to atleast have some informations about other nation ships at any point.

[SiWi]

1 minute ago, Tousansons said:

If the tech tree is just "better" techs allowing "better" parts, the player will build the "better" ships and in the end use the "best" combo with it's available technology against the percieved enemy menace. The type of AI ship design doesn't really come into play here, in the end the player have to build "better" if he want to win.

Real life navies usually tried to build ships with the idea of defeating their most likelly opponent. In a strategy game, intelligence of what the enemy is doing is an important part of game design. I doubt UA:D campaign will not allow the player to atleast have some informations about other nation ships at any point.

there is a vast difference between simply building the "better" ship or building the ship which perfectly fits the thread you encounter.

Of course you can try simply to put the best of all (armor, firepower, survival, speed) into a ship, but chances are that you can't do it, especially on budget (right now, money is relative irrelevant in custom battles). In which case, you probably will try to set a focus point: are you going to sacrfice armor for speed and firepower? Or the other way around? And what kind of firepower do you want? many smaller guns or few more powerful ones? If you know the enemy designs, you can make this choices far more meaningful. This gets more important, the "weaker" the navy you play is. The RN maybe can affort to have half a dozen of different BB design, for each "occasion" but China, Spain ect. won't.

 

I would hope so, but it isn't announced at the moment and I assume that the first version of the campaign will lack it.

[Cpt.Hissy]

2 hours ago, Tousansons said:

If the tech tree is just "better" techs allowing "better" parts, the player will build the "better" ships and in the end use the "best" combo with it's available technology against the percieved enemy menace. The type of AI ship design doesn't really come into play here, in the end the player have to build "better" if he want to win.

That's how it looks now, but campaign will bring in some new factors that should change it.

Persistency, for example. Once you've built something, you're stuck with it from now on, so perhaps you should think twice if you really want to build this super optimized counter to that one ship you know your enemy has - unless you also know what they will have in the future.

[Tousansons]

Literally what I wrote:

9 hours ago, Tousansons said:

In ditching AI random ship design, we can still improve "replayability" by making player choices matters more than a random generator. Be it by diplomacy, events, technology, ect.

I believe there is more interesting ways to create content in a game than a random number of guns and armor thickness in a hull. 

And:

8 hours ago, Tousansons said:

If the tech tree is just "better" techs allowing "better" parts, the player will build the "better" ships and in the end use the "best" combo with it's available technology against the percieved enemy menace

What does money change? Regardless of the type of AI ship design (random or fixed) a player is asked to build the best ship with the best technology and money available against a percieved menace.

It's not up to me deciding how this factors will come into play, good game design will make sure you can't just have a death star for five euros and will also make sure to punish a player over ambitious counter to a specific ship. Again, technology, events, ressources, etc.

My initial post is about removing the random AI ship design and replacing it with a fixed one, not deciding how the campaign will work (we have a pretty clear explanation in the front page) A fixed AI design can still create easier uniformity or flavor to a nation navy, facilitate player assessment of a threat with clearer enemy ship patterns and liberate the player ship designer to allow greater flexibility.

I get the initial concerns about replayability but in my opinion the campaign environnement should already offer enough branching with events, ressources, player built ships and diplomacy to create it, not because of a number of guns, but because of player choices.

[SiWi]

money changes things because you will not just need good ships but also numbers. Or the fact that you may simply can't build the "best" ship (or it takes too long). In other words: money changes (probably) what the "best" ship is.

Fix enemy designs take alot of change out if the Campaign, because you will know after a while the patterns and then simply build the same patterns yourself all the time, in every campaign.

 

Other factors may add to it again, but that doesn't change the fact that taking variety out in the ship designer, the main feature of the game btw, takes away from that.

[Tousansons]

3 minutes ago, SiWi said:

money changes (probably) what the "best" ship is.

Yet it is still a "best" ship with all these factors in mind. Don't you agree?

3 minutes ago, SiWi said:

Fix enemy designs take alot of change out if the Campaign, because you will know after a while the patterns and then simply build the same patterns yourself all the time, in every campaign.

Technology, events, money, player ships built is variety and create replayability. A random number of guns is just a random number of guns. Painting a ship in red instead of green add nothing.

Do I need to repeat the same things every time?

 

[Cptbarney]

17 minutes ago, SiWi said:

money changes things because you will not just need good ships but also numbers. Or the fact that you may simply can't build the "best" ship (or it takes too long). In other words: money changes (probably) what the "best" ship is.

Fix enemy designs take alot of change out if the Campaign, because you will know after a while the patterns and then simply build the same patterns yourself all the time, in every campaign.

 

Other factors may add to it again, but that doesn't change the fact that taking variety out in the ship designer, the main feature of the game btw, takes away from that.

In short, Ship designs aren't the only thing that offers replayability. Especially since if that was the case then we would be better off just having a sandbox mode really, the purpose of fixed designs is not only to allow players to add in their own designs (either using built-in assets and/or their own 3D models) but also so that the AI doesn't end up gimping itself due to poor choices it will make inevitably.

That and it hinders the designer since it's tied to that mechanic as well, meaning if it was reduced, removed (By choice or permanently) it should negate seeing ships with highly questionable loadouts and components, but somehow still being sea worthy.

It also helps speed up testing (since we can have more consistent results and even design ships towards tests and getting more reliable results from it helping the devs out in the future).

I don't mind the randomness nor the AI making auto ships if it has nothing better to choose or nothing at all to choose within a certain weight, length, armament, tech bracket. But as time goes on and we see more and more modules being put in, i would like the feature for players to start designing for fixed ships that the AI can use and set to use those only, as a higher priority, in leu of ship class or set off entirely so not only can players choose whether they want the AI to make its own ships or not.

But also, means less tweaking from the devs to sort out the auto-builder and focus elsewhere (such as the AI itself). Regardless if the devs can make a modular ship building system work in combination with the auto-designer, then thats great.

If not, then i guess the above would be a good solution, since then we can have more creative freedom and also should make mod support for models easier (since they will probs need to update the designer each time and modders would need to do the same to keep in check).

Another solution to help with stagnation would be to have a online shipyard, in which you can select from hundreds, thousands of ship designs and these being rated out of 10 (to filter out, poorly designed and meme ships) along with many filters to help choose exactly what you want.

Since with a ship library, that many players can upload too means we can choose from loads of ship types, means the devs can just focus on making parts as placeholders or generics for all nations and the AI to use. But also, allow us to design ships with a huge amount of freedom as well (for example splitting the superstructure up into 3-5 pieces and swapping those out for other bits, and allowing for interchangeable gun barrels and toggle whether smaller secondaries are armoured, semi-armoured or bare.)

If those, things legitimately can't work then thats fine, and i hope we see a greater sense of replayability in other areas of the game. Age of empires and Total war are 2 good examples of how you can have lots of replayability with the same units over and over again (hell AOE2 was kept alive for 14 years and got 2 remasters and a semi-rework as well). 

[SiWi]

1 hour ago, Tousansons said:

Yet it is still a "best" ship with all these factors in mind. Don't you agree?

Technology, events, money, player ships built is variety and create replayability. A random number of guns is just a random number of guns. Painting a ship in red instead of green add nothing.

Do I need to repeat the same things every time?

 

assuming that there is a universell best ship, build able yes. And that is more likely with no random designs. With random designs, which can dip into extremes, it is less likely that you have a universell answer, but rather you have to adjust. A ship design may be good against certain enemies but fail against others. Which is kinda the major point of this game btw.

Only because you have other factors also adding variety, doesn't make those factors as effective then random designs. Money and technology for example will hardly change so much by a different play-through, that it even really matters. Events maybe could, but thats a big if and not as half as effective then making the actual battles varied. I don't know why you insist that designs don't matter, in a game about ship designing. Of course the number of guns makes huge differences, of what that ship can do and what it can't do. Why do you even play the game if you think that 10 12 inch guns do the same as a 6 14 inch gun design? Do you really believe it is inconsequential? Changing ship color is (unless they change some spotting stat), but you really think it makes you difference what armament you have an a ship?

 

Thats my line.

[Skeksis]

...

Edited November 9, 2021 by Skeksis

[Cptbarney]

22 minutes ago, Skeksis said:

You just have to look at CCs, they’re already doing such things as scenarios via comments (and that can be difficult). Scenarios/ship blueprints exporting/importing already has a following so if UAD were to build an interface it would get used.    

QOL applications/steam workshops/hub etc. could be the next big thing, defiantly would be for us players.

These sorts of services from UAD/GameLabs should get some serious consideration, it would defiantly elevate this game and company.

Not only that, it will drastically increase replayability, customisability and also give the game some more major selling points over its competitors and could make itself into a market leader, and might even innovate this area of gaming too.

Plus, the ability to have perma ships. That can be taken into any game mode and shared with other players will be very nice. Especially if there was a particular ship or ship type that you loved for it's antics and performance in-general.

Edited March 31, 2021 by Cptbarney

[Tousansons]

7 hours ago, SiWi said:

assuming that there is a universell best ship, build able yes.

The player will build ships according to it's most likely enemy, with the best technology and best use of it's money. There is nothing universal, god like or death star. Re-read what I am writing. I'm not even talking about AI design here, but player design choices in response to a percieved threat.

7 hours ago, SiWi said:

Money and technology for example will hardly change so much by a different play-through, that it even really matters.

Technology will be different depending on the nation, probably even somewhat randomized during a playthrough (this is in the game campaign description) Depending on your wars, your losses, events and diplomacy, I don't see money being a fixed value either.

7 hours ago, SiWi said:

I don't know why you insist that designs don't matter, in a game about ship designing.

I said from the begining that fixed design will improve gameplay by offering things like better flavor to nations and why not allow in the future a better player ship designer. Random design doesn't add enough to the gameplay, this is what I said.

7 hours ago, SiWi said:

Of course the number of guns makes huge differences, of what that ship can do and what it can't do. Why do you even play the game if you think that 10 12 inch guns do the same as a 6 14 inch gun design?

A random number of gun is still just a random number of guns. Fixed ship design will also build ships with X*X inches and the player will react to them. What is your point here? I never said something about guns and calibers having no impact.

[Skeksis]

...

Edited November 9, 2021 by Skeksis

[The_Real_Hawkeye]

Complete randomness in ship design is most definitely not needed for "replayability", IMO.

Like, for the first batch of BB/BC, the Royal Navy goes the High Seas Fleet approach with an emphasis on armor an compartmentalization at the cost of gun caliber, speed and range. Then Jackie Fisher takes the helm and goes all out on "speed is armor" and build very fast, heavily armed but lightly armored BB and BC. After that, a more balanced approach is settled on and ships that are somewhat in-between are build.

I guess what I'm saying is this: Have libraries with ships that adhere to different doctrines and have the AI choose a doctrine every, I don't know 5 to 10 years and you _will_ see surprises. 

Oh, you build your ships to counter Invincible and Indefatigable? Good Morning, my name is Warspite!

[Skeksis]

...

Edited November 9, 2021 by Skeksis