AI feedback Alpha 4 v67

[shaitan]

Preface.

I am not the kind of player who can remain engaged for very long with static missions such as offered in the Naval Academy. However, give me a custom battle setup with AI controllable forces and a spreadsheet, and days will fly by. Thus recently with release of the latest Alpha update (Alpha 4 v67) I have been spending a fair amount of time smashing fleets of AI ships together while I patiently await the arrival of the campaign mode. During this time I have observed some behaviours that I feel it would be useful to begin a discussion on. Unfortunately this means a wall of text of incoming, but hopefully someone will find the time to read it and perhaps some useful points to engage with and discuss as development of this fine game continues.

 

Perceived desirable traits.

As of the time of writing, this will be a single player game. As such, the AI opponents needs to provide a challenging and engaging game play experience whilst not being so complex as to be to computationally heavy and bogging down the simulation. To my mind, providing this engagement can be done in two complementing fashions. The first of these is verisimilitude. The AI should exhibit behaviours that mimic real world/human behaviours, to provide the illusion that hidden away behind your screen an enemy Beatty or Scheer is directing the opposing forces and doing their damndest to prevail against you.

The second element is to provide a challenge. The AI should understand the systems inherent in the game it is playing. It should utilise those systems at the least in a competent manner in order to provide the greatest tactical challenge to a competent player. This includes at least giving the illusion of pursuing a coherent tactical battle plan, with some variety of overall approaches roughly appropriate to a given engagements strategic/tactical considerations. With these traits in mind, I will now highlight what I think are the largest negative factors presented by the AI admirals as currently implemented.

 

Target selection.

I have observed ships of all classes primarily engaging the lightest visible opposing elements with all available weaponry unless engaged at very close ranges. This results in some very spectacular explosions when a battleships main battery lands a square blow on a torpedo boat, but is perhaps not entirely desirable and certainly challenges perceptions of verisimilitude with the games gunnery model. I believe that a ships primary and secondary batteries should engage opposing forces with respect to the vessels role within the fleet.

Ideally ships primary weaponry should be engaging their opposing class first, engaging down only when no ideal target presents itself, and only engaging up as a last resort. Secondary weaponry should prefer to engage down, although certain classes of secondary weaponry may be optimised for other targets.

This assumes relatively historical ship design considerations, although given the design flexibility inherent in this title, perhaps prioritising targets dependent on the calibre of the individual battery would be more appropriate? As a suggestion, perhaps a useful guideline for weapon based target selection would be:

2-5” weaponry would prioritise torpedo boats and destroyers

4-7” would prioritise light / protected Cruisers

6-9” would prioritise armoured / heavy Cruisers, possibly lightly armoured battlecruisers

10”+ should be engaging enemy battleships and battlecruisers as priority.

Torpedoes should be prioritised as engaging the heaviest viable targets, though as a weapon of opportunity they should be used against any target that presents itself. Perhaps the margin of allowable error should be judged much more strictly when choosing to engage light manouverable elements such as opposing torpedo boats.

Currently an entire fleet will engage a single enemy vessel until its destruction. While probably the correct choice for the damage model as currently implemented, as more gradual reduction in fighting ability from accrued impacts is implemented (through crew casualties and the like), it should become viable to implement a more historically accurate tactic of each ship with in a line engaging its opposing number in the enemy line, only beginning to double up when friendly forces maintain numerical superiority.

 

Formation Keeping.

This mostly comes up when a unit tries to fall out of line due to battle damage. Currently a unit seeking to fall back to the end of their squadrons battle line will do a full 180 turn to achieve this via the shortest path for their current relative position (often coming to a complete stop in front of the enemy gun line and disrupting the formation of following friendly elements, also causing them to come to a complete stop). However, ships of the era -should- be aiming to maintain constant speed. The correct course of action for a ship unable to keep pace with their squadron is to pull out of line -away- from opposing forces, before settling into a parallel course allowing friendly units to pass between themselves and the enemy. Similarly, when formations are disrupted, the lead ship should reduce speed to allow the rest of their squadron to catch up, currently once disrupted any lead element that missed the blockage will quickly become isolated and find itself facing enemy forces alone and distant from supporting friendlies.

Position selection for supporting units: when given the supporting role (as opposed to follow), squadrons seek to position themselves between the supported formations and the enemy fleet. While this is the correct choice for scouting units before the main battle lines become engaged, this positioning is suicidal once heavier enemy forces are within firing range. Ideally supporting forces should be withdrawing to a position behind the primary gun line once battle is joined, only sallying back through the line should it become necessary to fend off enemy destroyer/torpedo boat attacks. These attacks should also not be the default behaviour for destroyers and torpedo boat, these forces should be held in reserve until such times as the battle lines close to a short distance or to finish off vessels already crippled by gunfire.

 

Supporting units manouvering around primary gun line.

This remains problematic and causes many of the formation issues mentioned above. When mixed category fleets are manouvering together, the heavier elements at most should hold their course when approached by lighter elements. A single torpedo boat trying to reach the other side of a battle line should not disrupt an entire squadron of battleships. Heavier elements should hold their course and allow the lighter units to perform the bulk of the evasive manouvering. Similarly, when selecting an appropriate path, unless capable of performing a well timed turn between gaps in a battle line (probably a bit complex to ask of an AI tracking many constantly changing variables of speed, position and tactical situation), the AI needs to be aware of all ships in a formation it is trying to manouver past. Ideally, instead of just trying to avoid the lead ship in a formation and disrupting all vessels trying to follow it, a formation should choose to reduce speed and allow the interceding formation to pass before cutting behind them to reach their intended position.

 

Fleet manouver.

Probably the lowest priority to adjust/implement as current, as addressing all previous issues will drastically change the conditions in which the current implementation is working and possibly produce different outcomes. Current observation seems to indicate 3 broad class of AI/AI engagements. The first, and most aesthetic, is when the lead element of each fleet chooses to make their initial turn onto parallel courses, forming something of a classic battle line IF the majority of supporting elements have been set to follow and thus falling out of the engagement envelope of the leading gun line, allowing the classic gun duel to develop.

If lighter elements are supporting and thus interposed between the gun lines however, the second type of engagement evolves in which battleships seeking to engage light elements end up clashing at a 90 degree angle at close range, the initial stages of this engagement are often slightly farcical with Battleships within spitting distance of each other engaging distant torpedo boats with everything they have, and as lighter elements succumb finally devolving into a close range slugging match, if one side fit torpedoes they tend to win at 1890s tech, if both side fit torpedoes, everybody dies.

The third engagement resolves when lead elements turn onto opposing courses, resulting in a circling engagement that will first slaughter supporting light elements caught in the middle (who often find themselves entangled with their opposing numbers), before falling onto the following support elements at the tail of each fleet, the battleships only finally engaging their opposing numbers when all lighter shipping has been eradicated.

 

Closing Remarks.

I understand that at this stage of development, many of the core mechanical systems the AI must utilise and work around are still very much in flux. Thus until the game systems themselves stabilise at some closer to a releasable state, to much work on the AI can be seen as a waste of developer time and effort. However, as we approach the time in which the campaign mode is released to the public for testing and comment, having a functional and satisfying AI opponent will go a long way towards showing the systems in place in their best possible light and allow current and future players to remain engaged with the games future development. This engagement would generate further interest with the titles progress and hopefully contribute towards the success of the final product. All errors and misunderstandings of the systems currently on show are my own and no fault of the development team. Finally, full speed ahead and damn the torpedoes.

 

[arkhangelsk]

6 hours ago, shaitan said:

The first of these is verisimilitude. The second element is to provide a challenge.

I think the priorities should be reversed, with being a challenge (that is, doing whatever works in the game) having priority. For one thing, humans gravitate in that direction while playing anyway, so to insist on not doing this is effectively to handicap the AI.
Second, sometimes humans stubbornly do "historical" things even when rationally it is not the best choice. For example, they'll engage the BB with the main guns because it is the "historical" thing to do. When they do this, they pass up opportunities like getting a torpedo shooter (destroyer) out of the game with only a brief diversion of fire. If the AI takes some initiative here, it makes the human player more effective and makes it a truly useful aid.
Overall, I don't think the target selection right now is that bad - I've seen the ships being able to direct main armament against my battleship and secondary against transports. It is good enough that I often just let its decision stand to see what comes out of it. The irritation people feel when fire is shifted to a destroyer is likely less reasoned than instinctive.

6 hours ago, shaitan said:

Position selection for supporting units: when given the supporting role (as opposed to follow), squadrons seek to position themselves between the supported formations and the enemy fleet. While this is the correct choice for scouting units before the main battle lines become engaged, this positioning is suicidal once heavier enemy forces are within firing range. Ideally supporting forces should be withdrawing to a position behind the primary gun line once battle is joined, only sallying back through the line should it become necessary to fend off enemy destroyer/torpedo boat attacks. These attacks should also not be the default behaviour for destroyers and torpedo boat, these forces should be held in reserve until such times as the battle lines close to a short distance or to finish off vessels already crippled by gunfire.

Don't you think that until we get formation work fixed to an extremely high standard we should avoid things like boats crossing the lines over and over again?

A basic problem with AI formation work is that it has to accomodate a human player, and a human player of extremely few words at that. Pretty maneuvers are the product of homogenous divisions that are trained, and with no sudden moves by a person well versed with their limitations. You can make a turn, but if you want prettiness then you shouldn't make another one until all ships have settled back into the original relative positions. If you want a "passage-of-lines" make sure your formation has sufficient space and carefully restrict the courses and speeds for both formations.

A human player of course generally does not follow these rules. He clicks randomly on the screen and won't hesitate to stick a destroyer to the butt end of a battleship division if it'll help his own force control. Given these ugly realities, that nearly instinctive, reactive shiphandling displayed by the AI may be the realistic response.

Further, you are literally forbidden from telling it some much needed information. Consider the section "Supporting units manouvering around primary gun line." Right now, you can't even define "Primary Gun Line" for the AI. At best you can tell it to screen a division, and perhaps the AI can be improved to navigate around that division. As for another other battleship divisions and cruiser divisions, you can't even tell it that it is also part of the gun line. A lot of information a real captain would be explicitly told is denied to the AI.

Of course, it may be possible to have the AI handle its own formations better, since IT will make homogenous divisions and can understand what is own Primary Gun Line is. But if ships start making beautiful 180-degree simultaneous turns and all that on the AI side while the player ships bump into each other, heads will roll.

Edited February 10, 2020 by arkhangelsk

[shaitan]

4 hours ago, arkhangelsk said:

I think the priorities should be reversed, with being a challenge (that is, doing whatever works in the game) having priority. For one thing, humans gravitate in that direction while playing anyway, so to insist on not doing this is effectively to handicap the AI.
Second, sometimes humans stubbornly do "historical" things even when rationally it is not the best choice. For example, they'll engage the BB with the main guns because it is the "historical" thing to do. When they do this, they pass up opportunities like getting a torpedo shooter (destroyer) out of the game with only a brief diversion of fire. If the AI takes some initiative here, it makes the human player more effective and makes it a truly useful aid.
Overall, I don't think the target selection right now is that bad - I've seen the ships being able to direct main armament against my battleship and secondary against transports. It is good enough that I often just let its decision stand to see what comes out of it. The irritation people feel when fire is shifted to a destroyer is likely less reasoned than instinctive.

You may be correct in your assessment, my ordering is down to my inherent biases with regards to the limitations of an AI opponent. In my view, if a player is primarily driven by the desire for challenging game play, then they will gravitate to multiplayer titles. No AI opponent no matter how smart it appears will ever truly challenge a competent players tactical decision making for any long period of time unless it is playing on an asymmetrical rule set (such as in AI War) or is allowed to somehow cheat the rules of the game that a player must adhere to (as in many other RTS titles with omniscient AI with production buffs). Given this understanding, I believe for a single player historical title, it is more important for the longevity of a game to primarily provide an immersive experience that gives the illusion of simulating its world faithfully. A player will always have the ability to circumvent the historical wisdom of the time and experiment with abusing the rules and systems of the game for maximum performance should they choose, part of the fun of historical titles is seeing how operating in an ahistorical manner changes outcomes against historical opponents, at least for me.

4 hours ago, arkhangelsk said:

Don't you think that until we get formation work fixed to an extremely high standard we should avoid things like boats crossing the lines over and over again?

A basic problem with AI formation work is that it has to accomodate a human player, and a human player of extremely few words at that. Pretty maneuvers are the product of homogenous divisions that are trained, and with no sudden moves by a person well versed with their limitations. You can make a turn, but if you want prettiness then you shouldn't make another one until all ships have settled back into the original relative positions. If you want a "passage-of-lines" make sure your formation has sufficient space and carefully restrict the courses and speeds for both formations.

A human player of course generally does not follow these rules. He clicks randomly on the screen and won't hesitate to stick a destroyer to the butt end of a battleship division if it'll help his own force control. Given these ugly realities, that nearly instinctive, reactive shiphandling displayed by the AI may be the realistic response.

Again, you are correct, in that many of this issues interact and influence each other. That being said, while 'passage of lines' is interdependent on the ability of the AI to manouver cleanly, the core decision making that is trying to send light units in between two groups of battleships is still something I believe is flawed and was thus worth highlighting as a separate item for consideration concurrent with the aforementioned issue.

4 hours ago, arkhangelsk said:

Further, you are literally forbidden from telling it some much needed information. Consider the section "Supporting units manouvering around primary gun line." Right now, you can't even define "Primary Gun Line" for the AI. At best you can tell it to screen a division, and perhaps the AI can be improved to navigate around that division. As for another other battleship divisions and cruiser divisions, you can't even tell it that it is also part of the gun line. A lot of information a real captain would be explicitly told is denied to the AI.

Of course, it may be possible to have the AI handle its own formations better, since IT will make homogenous divisions and can understand what is own Primary Gun Line is. But if ships start making beautiful 180-degree simultaneous turns and all that on the AI side while the player ships bump into each other, heads will roll.

I agree that for AI sub command of player units, greater ability to define roles and responsibilities for AI commanded sub divisions would be extremely useful. However, my focus in this discussion was directed towards wholly AI commanded fleets (ie. the AI opponent a player will face), so in the context of this discussion I mentioned 'primary gun lines' merely as a shorthand for the leading formation of the heaviest type, which the AI uses as its reference point for all other orders issued to its subordinate units, be that to follow or support that unit. In a player fleet many of these issues can be circumvented by experience and more time spent micromanaging ones forces to avoid them stumbling on each other, thus allowing a player a distinct advantage in large fleet actions such that they may easily manouver their well ordered fleet to a position such that it can piecemeal pick off AI ships escaping their customary traffic jam.

[I Scouting Group]

Of all the points raised here, I think that those about target selection have the greatest universal relevance. I've only had the early-access version of the game since last week and have only spent a handful of hours with it, but I noticed pretty much immediately that the AI (both player ships and the enemy) do tend to concentrate their fire on a single ship.

In my opinion, this should not be the default practice for two important reasons:

1.) Many ships firing at the same target makes it difficult to evaluate fall-of-shot

One of the (many) reasons for moving from pre-and-semi dreadnought designs with a mixed battery of heavy and "medium" caliber guns was that it is difficult to distinguish between the shell splashes of the two different sizes of guns and adjust for fall of shot accordingly. This same general phenomenon (i.e. not being able to tell which splashes came from which guns) could and did arise when multiple ships were firing the same or similarly-sized shells at the same target. The hail of fire that the Queen Elizabeth-class ships were briefly subjected to when they encountered the German line at Jutland is one example of this. Another prominent example was at the Battle of the Denmark Strait. The engagement began with both Bismarck and Prinz Eugen firing at Hood, but after a few minutes later, Bismarck's Chief Gunnery Officer Adalbert Schneider requested that Prinz Eugen be ordered to shift her fire to the other ship (the Prince of Wales).

In an uneven engagement or a victorious battle in which one has achieved numerical superiority of like-sized ships, having two or more ships fire at a single target is of course unavoidable, but at that point, reduced accuracy through difficulty in spotting fall of shot and correcting the firing solution is made up for through numerical superiority.

2.) It is much easier for the enemy to shoot at you when you're note shooting at them.

During the opening phase of Jutland (the "Run to the South") an error in British fire distribution led to SMS Derfflinger being left unmolested. A ship that's not being fired at can maintain a steady course, does not have to deal with the physical and psychological effects of hits and near-misses, and has the luxury of being able to pick its targets rather than being compelled to return fire at the most immediate threat.

For these reasons, it makes the most sense for every heavy ship to engage its opposite number in line and only "double up" once enemy vessels sinking, retiring, or falling out of line necessitate doing so. I would argue, then, the the AI's default practice should be to distribute its fire evenly across similarly-sized targets with its primary batteries. If smaller targets (torpedo boats, destroyers, etc.) can be engaged with secondary batteries, it should target them. If none are present and heavy ships are in range of the secondary batteries, it should use them that way.

Assuming that folks agree with these two general premises, I would propose the implementation of a penalty for multiple ships firing at the same target. Rather than an outright accuracy penalty, I would propose a penalty to the rate at which firing accuracy increases. In theory, multiple ships firing at a target should not affect initial accuracy, assuming that the the firing ship has a good sight picture to range one. Numerous splashes from multiple ships should, however, make adjusting shot more difficult, and therefore should make the rate at which accuracy improves (either through continued shooting or closing of range, or both) slower. Two ships firing at the same target should perhaps reduce the rate at which accuracy increases by 33%, with the penalty scaling rapidly as the number of ships increases such that if four or more ships are firing at the same target, they're all locked in to their "base accuracy" for that range and target signature until some of the ships cease fire.

[arkhangelsk]

This would seem to be a bit off topic because it concerns the gunnery and crew morale models rather than the AI model. The base problem is that right now these things are not simulated. If they are in the future, then I would like the AI to decide whether to stick with concentrated fire or distributed based on what provides the best overall probabilities of suppressing the enemy fire.

To cover it briefly, however, for both the problem is that while they make superficial sense, as far as magnitudes are concerned there is no firm answer. To take an example, it's true Derflinger was unengaged, so you would expect it to have the best hit performance, but it was Lutzow that got the best hit percentage. The USNWCMR assess a 20 per cent bonus for ‘not being under equivalent fire’, but how close to the truth that was is anyone's guess.

Similarly, the effects of Fire Concentration can be very mixed. Schneider seemed to have little problem getting her battery on target anyway. And CB 3011 also awards an additional 10 per cent bonus above 12,000 yards and a 25% bonus over 15,000 yards for ‘Concentration Fire’  - a technique whereby two or three ships used the lead ship’s fire control solution (corrected for distance from the leader) and fired simultaneously. Yes, the estimated hit rate got a bonus instead of malus for having ships concentrate the fire depending on fire technique (you can't control fire technique in the game ... all you can do is tell all the ships to shoot the same target).

[shaitan]

I can't help but wonder, with regards to the preceding comments, how much the AI factors probability of armour penetration into its target selection mechanics. Given that increasing angles of incidence reduce probability of armour penetration, if this consideration were bumped up in the AIs target selection priorities list, it could lead to less focussed target selection across a a fleet. Ships relatively distant from the lead element would have their shots land at a less than ideal angle for armour penetration, reducing the value of that target in the eyes of the AI. It wouldn't go all the way to fully distributed firing techniques, but would spread out into a few groupings along an opposing line, depending on how strictly it was applied and exactly how the selection logic was written.

[RedParadize]

I intended to drop this in its own tread, but you did a nice job at explaining the problem.

Currently AI is not sophisticated, its more or less look like this:

Targeting:
1 Ship will target the enemy highest chance to hit. With exception on destroyer or when ship get too close.

Avoidance (take priority over everything else):
1 Collision avoidance: Sometimes it get messy and ship get stuck in that mode.
2 Torpedo avoidance: Trigger when torpedo are close. torpedo spotted by closer ship will not trigger this.

Main line
1:Leading main line ship go to max range then turn at a 80deg angle to their target. If target change, they change course. Leading ship will not slow down, regardless of following ship speed.
2:Following main line ship just follow the leading ship, regardless of their surrounding.

Support line
Secondary ship can be in 3 mode:
1 Screen: Ship will go toward target until it reach a certain distance (what decide it?). Then from there they will angle 80deg to their target... or do donuts on the spot for some reason.
2 Follow: Same as the ship following the main line, sometimes triggering collision avoidance and messing up formation.
3 Retreat: Self explaining.

This can easily abused. I use to do this with a simple destroyer, but now that they are easier to hit than a CA, the latest is now a better option. What you do is you send a CA toward the enemy, more on the front or the rear depending on were you want the enemy to turn. Front if you want them to turn away or back to turn them in. You can go around them too. By turning the CA in or out, you can split formation, slow the enemy, waste their ammo/torpedo and make them perpetually switch target. I did this several time with great effect yesterday. I pitched 2 AI built BB and my own CA versus 6 BB 2 CA and 4DD. I won every battle hands down without ever controlling the two Battleship.

 

Edited February 10, 2020 by RedParadize

[arkhangelsk]

6 hours ago, RedParadize said:

I won every battle hands down without ever controlling the two Battleship.

If this is literally true, the AI was doing all the maidwork with your two battleships, leaving you free to dedicate all your energies to directing your cruiser to "feel the edges" of the AI targeting algorithm. That's not the end of the world.

[RedParadize]

Well, you are free to think that predictable and easily abused AI behavior is not a problem for a strategy game.

[arkhangelsk]

Well, no need to get testy. I just think we need a sense of perspective. Now, how to defeat your exploit. One solution is indeed to use "categorical targeting". This problem can probably be avoided if the BBs will NOT shift fires to your CA no matter what.

I'll suggest the following changes to the AI:
1) The AI will use divisions of pairs (so as to exploit the formation bonuses). After all, its advantage is that it is not really bamboozled by having more subunits to handle. 15 sets of decisions will not cost it anything over 3 - the only person annoyed is the player if there is a frame rate reduction.
2) The initial targeting is to be left as is because it is a good general rule.
3) Increase the advantage delta required to shift targets (in essence, your cruiser will have to be sucked in closer to exploit the effect, and retreat longer before the AI even considers shifting target again).
4) Further, if after 1st shift, target hit rate starts to decrease, at the point where condition 3 is met again
a) Only the primary battery is shifted back
b) If more own BB than enemy BB, then one BB doesn't shift into reconsideration until CA structure is significantly down (in my experience, the AI does have some sense of shifting targets after one is weakened).

Would that work? 

[RedParadize]

@arkhangelsk Sorry, I placed word in your mouth on my last post. What I said was literally true btw. The AI is very easily abused. Before we discuss this let me explain how easy it is to game, just to make sure we are on the same page. 

Battle was a 2xBB 1xCA vs 4 BB 2xCA 4xDD. I made the CA, it need to be small, priority on speed and armor. it had 4 double torpedo launcher per side and 4 triple 6" turret.

I placed my two BB on AI right at start. The two main fleet turned right, so they were going parallel. So I went center left, torpedoed the first BB of the line, one torpedo hit and it got engined. As it moved to the back of the formation, I got closer to make them all turn around me instead of the two ally BB. Doing so, they got closer to my two allied battleship, but were still firing at me. The engined BB, now last, got isolated and killed by my two ally.

This is a picture that I took a bit later on, I think it clearly show the abuse:


 

Edited February 11, 2020 by RedParadize

[RedParadize]

Oh, note that if I feel threatened. I just need to turn away form them a bit. By doing that they will all target the two battleship again. If you have the room to do it repetitively then you can stall the enemy in a perpetual target switching and ladder fire.

[arkhangelsk]

Before we fix the AI, we need to fix the Ship Designer. Your thing is basically a Treaty cruiser in displacement, but it has

1. 16 torpedo tubes (with multiple reloads)
2. 12 6-inch guns (equivalent weightwise to 8 8-inchers so don't act all wounded there - you might even see about 9-inch guns, the hero caliber of this round)
3. 38 knot speed (speed obviously set based on the latest research on the Fast Ship Speed Modifier )
4. Up to seventeen inches of Krupp IV armor?

If it wasn't for three and four, just by adding some more persistence before switching back your cruiser should be turned into Swiss cheese in short order.

I'm not sure on what kind of response you want from the AI against your cheese design. Tactics like categorical selection won't work, because secondaries or cruiser guns should not be very useful against your SEVENTEEN inches of armor. With so many torpedoes, it cannot reasonably ignore your cruiser or the fact the secondaries aren't hurting it.

I guess the only thing left to try is to use its destroyers to try to torpedo your cruiser, but the problem of making it work aside, in this iteration, a destroyer's real survivability against an 8-9 inch gun is in the region of 2 hits (with battleship calibers it is one), so those destroyers are likely to die pretty fast after which your ship can continue to work on the enemy battleship.

Ideas?

1 hour ago, RedParadize said:

Oh, note that if I feel threatened. I just need to turn away form them a bit. By doing that they will all target the two battleship again. If you have the room to do it repetitively then you can stall the enemy in a perpetual target switching and ladder fire.

Your ridiculous ship aside, I have seen a similar symptom before (though in Alpha 2), where the AI (on my ship) just repeatedly shifted between two cruisers. For this problem at least I think the problem is that the sensitivity is set too high (the computer makes too much significance of small, temporal changes in hit chance or other criteria) and the decisions are made continuously or nearly so.

In addition to reducing sensitivity, perhaps a measure would be to make it run the comparative algorithm only periodically. I call this the normative fire order concept. When target is shifted, the computer should not even think about shifting targets until it is locked (or it goes 5-6 salvoes without a lock). Next, it should estimate the number of salvoes required to score at least one effective hit (which it can do quite well since it knows all the formulae). Three, it should complete this fire order before doing any rethink.

How does that sound?

Edited February 11, 2020 by arkhangelsk
A comment about target shifting