The AI thread

[Pequod]

On 6/28/2026 at 11:11 PM, BraveSirRobin said:

Are you talking about the IL2 GB AI that everyone is currently complaining about or the IL2 1946 AI that blatantly cheated?

Did the AI blatantly cheat in IL-2 1946? Absolutely. But make no mistake: just because the AI in IL-2 Great Battles uses the same flight model as the player doesn't mean it doesn't cheat.

Some people say it can't see through clouds; others insist it can. I've been torn apart by the AI in terrible weather, weaving through clouds, when I could barely see my own wingmen at 50 meters. Its situational awareness is superhuman. It can spot you in situations where a real pilot would struggle, such as when you're on his six o'clock low or against a complex ground background. It has instantaneous positional awareness and flawless target identification.

A simulation "cheats" in almost everything. It can't reproduce reality exactly because doing so would be computationally impossible. It doesn't have to be exact—it has to be convincing. The end result simply has to feel believable.

The old IL-2 1946 helped the AI pilot with a more forgiving flight model, but its decision-making (built around a large number of scripts) produced more realistic behavior. The same is true of WoFF and Falcon 4.0.

In IL-2 Great Battles, the AI pilot follows the same flight rules as the human player, but its behavior is dreadful, whether friend or foe. The moment I start noticing repetitive behavior patterns or nonsensical decisions, immersion disappears instantly.

Programming a believable pilot is far more difficult than programming a realistic aircraft.

Edited July 2 by Pequod
Mistake

[Xtremist]

I find the AI very suicidal, they do nothing to avoid collision, and keep firing and dogfighting while in flames. Even the cargo planes are dogfighting, they turn after you like they had guns. 

[Jarg1]

On 6/20/2026 at 4:54 PM, belphegor said:

Regarding the effectiveness of the B-29 defensive guns, does anyone know how many MiG-15s were shot down by B-29s in the Korean War?  As far as I can tell, in the April 12, 1951 battle, no MiG-15s were lost, or at least not to the B-29s' guns.  The B-29 had a pretty advanced remote control and lead computing gun system, but it was designed with prop plane speeds and ranges in mind; it wouldn't surprise me if it just wasn't very effective once jets came onto the scene.

 

On 6/20/2026 at 6:38 PM, KillerBee said:

I offer this link concerning B-29s in Korea: https://nuclearcompanion.com/data/b-29-superfortress-chronology-korean-war/#june-1950

This might provide some insight.  There are many more Korean War resources out there for sure that may confirm or deny.
 

[Edit] If you’re a mission builder, I suggest you explore the compass direction those B-29s are flying to and from targets on and the effective combat ranges of the various intercept aircraft involved before committing to a mission design.  Remember, there are no B-29s in Korea![/Edit]

Another page at the same site - Known B-29 Superfortress Losses in Korea (1950-1953) - Nuclear Companion: A nuclear guide to the cold war

It lists 16 B-29 losses to enemy aircraft, and claims of 27 aircraft destroyed including 16 MIG-15s, with another possible 17 MIG-15s destroyed and 11 damaged.  If this is accurate it suggests that attacking B-29s was a very dangerous proposition!

 

[357th_KW]

Some of the criticism in this thread is pretty far off base - like "they only turn left".  That is easily proved false by setting a quick battle, the AI will employ a variety of moves, both horizontal and vertical, and the more complicated the scenarios you build for them, the more varied and organic feeling their actions become. 

That said, I think there are some places where we desperately need to see some improvements over the existing in game AI.  A few people have already brought up some of these points already, but I think it's worth touching on them further:

AI should have limited situational awareness, similar to a human.  This is actually already possible at a simple level, even in IL-2:GB.  I've implemented AI with this code on my multiplayer PvP/PvE server, The Cold Blue, in IL-2:GB.  You can surprise them sometimes, but if you don't pounce on them quickly, eventually they will spot you and react.  Unfortunately they still have no awareness or recognition of clouds - something which appears to have been addressed already in IL-2: Korea.  The code unfortunately eats into your already limited resources.  It's also very time consuming and difficult to setup and configure, so built-in optimized code for this that you could adjust in the aircraft properties in the mission editor would be a huge improvement.

AI needs to have some sort of morale system.  It's frustrating when all enemies only fight to the death and will chase you all the way back to your base with no regard for their own safety.  Again, this is actually already achievable in IL-2:GB and I've implemented it on my server.  I've made a basic morale system where enemies will react to wingmen being shot down and sometimes choose to run.  They also won't chase you forever and will give up and turn back if you've outrun them.  This is an area where it would be nice to have some additional tools - there's no easy way to make them respond to damage short of being shot down (the OnDamaged report in the IL-2 mission builder has never worked properly, and really reacts almost identically to OnKilled, making it impossible to manage that behavior).  Again, having some sort of built-in, customizable morale system in the mission editor would be a huge step forward.

There needs to be a way to dynamically spawn and assign AI wingmen to players in multiplayer missions.  There is currently no way to do this which is a big downside in multiplayer.  Real combat flying is a team sport, and has been almost from the beginning.  Unfortunately most PvE or PvP multiplayer largely involves humans flying solo.  Giving players the option to spawn an AI wingman or flight with them would give a much more authentic feel to the multiplayer experience.  It would be a game changer for attack/bomber pilots, where having a formation to work with would really increase the amount of impact you can have, as well as increasing survivability.

 

Edited July 3 by 357th_KW

[gecko]

This would be a very key area where an old piece of sim-design wisdom is very appropriate: It doesn't matter what it's doing; it matters what it looks like it's doing. 

If someone thinks they will actually be able to notice if the AI modeling is or isn't taking into account all of the factors some have listed, well, we have institutions for people like that. This is an area where the more complexity that is built in, the higher the likelihood of unintended consequences. A visual example of this was posted in the screenshots section, where a truck was shot up by 50 cals, blasted into the air, and landed balanced on its front bumper. I think we're all really happy to see the level of complexity that has been added to ground targets, and a more physics-based approach. It probably works well most of the time, but clearly some of the laws of physics aren't applied in all the ways they need to be. The complexity to do that, much like the level of granularity some are saying the AI must have, is not realistic. This begs the question, is it better to have "physics-based" ground objects and uber-detailed AI modeling that occasionally (or quite possibly, regularly) produce ridiculous, immersion-breaking results, or much simpler mechanics that rely on rules and probabilities to ensure that a reasonable variety of realistic outcomes always happen? I'd argue that the capabilities of our software are still a long way from being able to achieve the former, so in that regard, I agree with BSR. It isn't achievable. Where I disagree, is that it is necessary. Especially when the call from players is not for a perfect AI - just one that is better than the rather poor one in Great Battles. To say improvement beyond mediocrity is impossible because perfection is not achievable is silly. A simplified model that works well is better than a complex one that does not, especially when it's in an area that the player cannot directly observe. That it's possible to create one that does a noticeably better job than Great Battles has already been demonstrated by other development teams. It requires prioritization, and a willingness to accept the current limitations and then decide to produce something that works well within them.

My own impression is that the AI already has most, if not all, of the capabilities it needs to be able to do a very good job, but that the rules and probabilities for how and when they are used need adjustment. I think if this is done thoughtfully, major improvement is possible just using what is already there. Perfect? No. Better? Definitely. Some of what 357th is saying would be a great start.

[Avimimus]

12 hours ago, Jarg1 said:

Another page at the same site - Known B-29 Superfortress Losses in Korea (1950-1953) - Nuclear Companion: A nuclear guide to the cold war

It lists 16 B-29 losses to enemy aircraft, and claims of 27 aircraft destroyed including 16 MIG-15s, with another possible 17 MIG-15s destroyed and 11 damaged.  If this is accurate it suggests that attacking B-29s was a very dangerous proposition!

I do wonder how many of those kills can be confirmed based on post-cold war archival analysis?

17 August 1943 Schweinfurt–Regensburg mission fighter escorts claimed 21 Luftwaffe fighters destroyed and the bomber gunners claimed 288 while Luftwaffe records showed 40 fighter aircraft lost.

[Donik]

Extremely well said @gecko!! I'd also encourage everyone to dedicate an hour or so on @357th_KW The Cold Blue server. The AI improvements he's implemented there is immediately noticeable if you have experience with the AI behaviors.

[Stonehouse]

12 hours ago, 357th_KW said:

Some of the criticism in this thread is pretty far off base - like "they only turn left".  That is easily proved false by setting a quick battle, the AI will employ a variety of moves, both horizontal and vertical, and the more complicated the scenarios you build for them, the more varied and organic feeling their actions become. 

That said, I think there are some places where we desperately need to see some improvements over the existing in game AI.  A few people have already brought up some of these points already, but I think it's worth touching on them further:

AI should have limited situational awareness, similar to a human.  This is actually already possible at a simple level, even in IL-2:GB.  I've implemented AI with this code on my multiplayer PvP/PvE server, The Cold Blue, in IL-2:GB.  You can surprise them sometimes, but if you don't pounce on them quickly, eventually they will spot you and react.  Unfortunately they still have no awareness or recognition of clouds - something which appears to have been addressed already in IL-2: Korea.  The code unfortunately eats into your already limited resources.  It's also very time consuming and difficult to setup and configure, so built-in optimized code for this that you could adjust in the aircraft properties in the mission editor would be a huge improvement.

AI needs to have some sort of morale system.  It's frustrating when all enemies only fight to the death and will chase you all the way back to your base with no regard for their own safety.  Again, this is actually already achievable in IL-2:GB and I've implemented it on my server.  I've made a basic morale system where enemies will react to wingmen being shot down and sometimes choose to run.  They also won't chase you forever and will give up and turn back if you've outrun them.  This is an area where it would be nice to have some additional tools - there's no easy way to make them respond to damage short of being shot down (the OnDamaged report in the IL-2 mission builder has never worked properly, and really reacts almost identically to OnKilled, making it impossible to manage that behavior).  Again, having some sort of built-in, customizable morale system in the mission editor would be a huge step forward.

There needs to be a way to dynamically spawn and assign AI wingmen to players in multiplayer missions.  There is currently no way to do this which is a big downside in multiplayer.  Real combat flying is a team sport, and has been almost from the beginning.  Unfortunately most PvE or PvP multiplayer largely involves humans flying solo.  Giving players the option to spawn an AI wingman or flight with them would give a much more authentic feel to the multiplayer experience.  It would be a game changer for attack/bomber pilots, where having a formation to work with would really increase the amount of impact you can have, as well as increasing survivability.

 

Another one is AI using flak avoidance when engaged by flak and also by the 1950s ground attack tactics for FEAF aircraft had moved on quite a bit from WW2. Things like attacking using multi axis co-ordinated attacks to split AAA across the maximum number of targets and directions come to mind. I have a feeling that AAA while deadly in game is not as deadly as it should be. Probably because AI at present would not cope and also perhaps radar laid guns are not fully implemented yet. By the 1950s both sides used radar to guide a lot of the AAA used. About 20% of FEAF aircraft losses were to enemy AAA.

[Dash,Polder]

It has to concede to the games needs, if you make it too deadly no one wants to play it.   Of the 335 F-51 losses 10 were to enemy air, 172 to ground fire.  17 to 1 ratio losses, but a game that does that sold for air combat won't sell very well, that ratio will be pushed toward a reversed in order to fill servers and move units, and I'm OK with that.  No one enjoys dying to a bot at those rates, against another player is a different story, cause when you do turn the tables and win its a completely different thing. 

[FlightBud]

Extremely disappointed with IL2 Korea… just played Berloga and fought some AI planes and it’s pitiful… they’ve made absolutely no improvement over GB and it’s just so boring… I expected more out of Korea. This honestly feels like an expansion pack with better graphics but performance that’s so bad that you need to turn down settings until graphics are actually worse than GB…

[Stonehouse]

14 hours ago, Dash,Polder said:

It has to concede to the games needs, if you make it too deadly no one wants to play it.   Of the 335 F-51 losses 10 were to enemy air, 172 to ground fire.  17 to 1 ratio losses, but a game that does that sold for air combat won't sell very well, that ratio will be pushed toward a reversed in order to fill servers and move units, and I'm OK with that.  No one enjoys dying to a bot at those rates, against another player is a different story, cause when you do turn the tables and win its a completely different thing. 

It does I agree but it also needs to be balanced somehow so that it feels like a threat and in enough density to be immersive even if no deadlier than now. This plus wanting to have flak in sensible locations without killing frames was a big driver for me to create the flak generators mod for IL2 GB. I find flying enroute to the mission objective over towns and railyards and airfields, ports etc that should be defended by flak but isn't a real immersion killer.

[AndreiTomescu]

1 hour ago, FlightBud said:

I expected more out of Korea. This honestly feels like an expansion pack with better graphics but performance that’s so bad that you need to turn down settings until graphics are actually worse than GB…

Korea it's not released yet. This is just an early glimpse. It would be fair to make an opinion after release, not just yet. Also if you read the dev's blogs, you'll know what to expect, and then , if expectations aren't met, you'll have every right in the world to be disappointed. Even angry.

To hyper resume the blogs, the hole game will be worked on till the end of the year.

I hear you: now it's not The next Generation, Star Trek. We all wanted and hoped for that, also hyped by the devs blogs that presented things as such. (the next gen). But they have the ground to be so. It will take a heap of work from them and some good amount of patience from us. Are you ready for that? Am I ready for that? Well, what else to do ? Constant pressure and bad vibes will slow the process, imho. I'm telling you from my own experience.

This early release is actually a stupid idea, since most players don't read/understand, are disappointed, and make bad opinion about something that's not even released. Rockstar way, imho, is the smart way.

But also, you might be quite a performant flyer, and then of course not even the improved AI aircraft will present any challenge to you. Probably that PvP is for, actually.

I ain't saying that the AI is much different from GB now, 'cause it ain't. Fine ideas were developed here on how to improve. Also the SP content it's almost barren. And the game is full of annoying bugs. And to get decent image you need a powerful rig. And that's expensive. And the optimization, especially for VR sucks. And it crashes to desktop, quite often. And the trees are still turning. Did I cover it all ? Guess not.

But the game is not released yet. 

Let's give these guys a fighting/working chance, before being disappointed. It's their big interest to please the players, since the product must sell well. 

Just my opinion.

[Panzerlang]

1 hour ago, AndreiTomescu said:

Korea it's not released yet. This is just an early glimpse. It would be fair to make an opinion after release, not just yet. Also if you read the dev's blogs, you'll know what to expect, and then , if expectations aren't met, you'll have every right in the world to be disappointed. Even angry.

To hyper resume the blogs, the hole game will be worked on till the end of the year.

I hear you: now it's not The next Generation, Star Trek. We all wanted and hoped for that, also hyped by the devs blogs that presented things as such. (the next gen). But they have the ground to be so. It will take a heap of work from them and some good amount of patience from us. Are you ready for that? Am I ready for that? Well, what else to do ? Constant pressure and bad vibes will slow the process, imho. I'm telling you from my own experience.

This early release is actually a stupid idea, since most players don't read/understand, are disappointed, and make bad opinion about something that's not even released. Rockstar way, imho, is the smart way.

But also, you might be quite a performant flyer, and then of course not even the improved AI aircraft will present any challenge to you. Probably that PvP is for, actually.

I ain't saying that the AI is much different from GB now, 'cause it ain't. Fine ideas were developed here on how to improve. Also the SP content it's almost barren. And the game is full of annoying bugs. And to get decent image you need a powerful rig. And that's expensive. And the optimization, especially for VR sucks. And it crashes to desktop, quite often. And the trees are still turning. Did I cover it all ? Guess not.

But the game is not released yet. 

Let's give these guys a fighting/working chance, before being disappointed. It's their big interest to please the players, since the product must sell well. 

Just my opinion.

Ya! 🙂
 

 

[Steini18778]

2 hours ago, AndreiTomescu said:

Korea it's not released yet. This is just an early glimpse. It would be fair to make an opinion after release, not just yet. Also if you read the dev's blogs, you'll know what to expect, and then , if expectations aren't met, you'll have every right in the world to be disappointed. Even angry.

To hyper resume the blogs, the hole game will be worked on till the end of the year.

I hear you: now it's not The next Generation, Star Trek. We all wanted and hoped for that, also hyped by the devs blogs that presented things as such. (the next gen). But they have the ground to be so. It will take a heap of work from them and some good amount of patience from us. Are you ready for that? Am I ready for that? Well, what else to do ? Constant pressure and bad vibes will slow the process, imho. I'm telling you from my own experience.

This early release is actually a stupid idea, since most players don't read/understand, are disappointed, and make bad opinion about something that's not even released. Rockstar way, imho, is the smart way.

But also, you might be quite a performant flyer, and then of course not even the improved AI aircraft will present any challenge to you. Probably that PvP is for, actually.

I ain't saying that the AI is much different from GB now, 'cause it ain't. Fine ideas were developed here on how to improve. Also the SP content it's almost barren. And the game is full of annoying bugs. And to get decent image you need a powerful rig. And that's expensive. And the optimization, especially for VR sucks. And it crashes to desktop, quite often. And the trees are still turning. Did I cover it all ? Guess not.

But the game is not released yet. 

Let's give these guys a fighting/working chance, before being disappointed. It's their big interest to please the players, since the product must sell well. 

Just my opinion.

Normally i would agree, but dev's have told us since the beginning of IL2 that an AI developer is hard to find. Since 14 years AI had only a very slow progress. There was no single korea DD about AI the last 2,5 years, not a single statement. They told us a lot about tiny functions, all interessting, but about AI there is silence only. They do know that they can not promise any major progress here. They have to keep silence to sell their product. After release they will keep telling it will be better, this is what they tell us since 14 years now. If they have a talented developer now, will he start working in august? Lets be realistic.

19 hours ago, gecko said:

This would be a very key area where an old piece of sim-design wisdom is very appropriate: It doesn't matter what it's doing; it matters what it looks like it's doing. 

If someone thinks they will actually be able to notice if the AI modeling is or isn't taking into account all of the factors some have listed, well, we have institutions for people like that. This is an area where the more complexity that is built in, the higher the likelihood of unintended consequences. A visual example of this was posted in the screenshots section, where a truck was shot up by 50 cals, blasted into the air, and landed balanced on its front bumper. I think we're all really happy to see the level of complexity that has been added to ground targets, and a more physics-based approach. It probably works well most of the time, but clearly some of the laws of physics aren't applied in all the ways they need to be. The complexity to do that, much like the level of granularity some are saying the AI must have, is not realistic. This begs the question, is it better to have "physics-based" ground objects and uber-detailed AI modeling that occasionally (or quite possibly, regularly) produce ridiculous, immersion-breaking results, or much simpler mechanics that rely on rules and probabilities to ensure that a reasonable variety of realistic outcomes always happen? I'd argue that the capabilities of our software are still a long way from being able to achieve the former, so in that regard, I agree with BSR. It isn't achievable. Where I disagree, is that it is necessary. Especially when the call from players is not for a perfect AI - just one that is better than the rather poor one in Great Battles. To say improvement beyond mediocrity is impossible because perfection is not achievable is silly. A simplified model that works well is better than a complex one that does not, especially when it's in an area that the player cannot directly observe. That it's possible to create one that does a noticeably better job than Great Battles has already been demonstrated by other development teams. It requires prioritization, and a willingness to accept the current limitations and then decide to produce something that works well within them.

My own impression is that the AI already has most, if not all, of the capabilities it needs to be able to do a very good job, but that the rules and probabilities for how and when they are used need adjustment. I think if this is done thoughtfully, major improvement is possible just using what is already there. Perfect? No. Better? Definitely. Some of what 357th is saying would be a great start.

We know that the AI can be much better with a good mission design. @AEthelraedUnraed told us years ago and you are probably right. But what is the point when i start a campaign on my own computer and won't have any of those observations? If you have those experiences then share them with developer. Something is wrong and you can help solving it.

When the game is released in 4 weeks with the commander mode we will see what will happen when AI hits in. The commander mode depends only on a well made AI. 

Edited July 4 by Steini18778

[AndreiTomescu]

@Steini18778 what you say is 110% true. 

A bit of time ago, when I had the chance to help a bit some really talented campaign designers, like Sandmarken, Jaegermeister and Vendigo, I've noticed that the ducking same AI that prevented me from playing (and not without trying) any stock Career suddenly became quite acceptable, if not good. (I'm a medium/poor flyer).

So there is some magic to be done!

Regarding the lack of AI improvement: you are right again. But these/our complains have been here for ages, with quite modest changes/results.

My point is to change the approach. See, they have quite bold ambitions now. They want to go on the larger market. This might require those changes we dream of, and more so , that the game needs to stay modern and even  go larger. 

So, let's support that. Imho, they are trying. Not always succeeding, but they are.

I would encourage, not criticise, at this stage of...."work in progress".

I do realise, the devs i'm sure realise, most of us realise that failing this project equals the competition catching up. And one can't use the wild card of a previous awesome title (1946) more than once. So unless they are ..... dreamers, they know they must provide now.

About my experience: when a client puts extreme pressure on me, I tend to be more cautious and choosing the "safest" options instead of the best tailored for him.

When one has the "i really trust you're doing your best for my best" approach, I have the tendency to try and move mountains for that case.

I agree totally with your observations, and even embrace a bit your frustration, but i'm still hoping.

Don't worry about them devs knowing our opinions: the forum is closely watched.

So my take is: they promised what they promised in the blogs. If that announcements are going to be fulfilled, till the end of the year, then it's fair & square, imho.

To help getting that, I would go for the supportive approach, instead of the critical one, since the latter already failed. 😃

 

[Steini18778]

36 minutes ago, AndreiTomescu said:

@Steini18778 what you say is 110% true. 

A bit of time ago, when I had the chance to help a bit some really talented campaign designers, like Sandmarken, Jaegermeister and Vendigo, I've noticed that the ducking same AI that prevented me from playing (and not without trying) any stock Career suddenly became quite acceptable, if not good. (I'm a medium/poor flyer).

So there is some magic to be done!

Regarding the lack of AI improvement: you are right again. But these/our complains have been here for ages, with quite modest changes/results.

My point is to change the approach. See, they have quite bold ambitions now. They want to go on the larger market. This might require those changes we dream of, and more so , that the game needs to stay modern and even  go larger. 

So, let's support that. Imho, they are trying. Not always succeeding, but they are.

I would encourage, not criticise, at this stage of...."work in progress".

I do realise, the devs i'm sure realise, most of us realise that failing this project equals the competition catching up. And one can't use the wild card of a previous awesome title (1946) more than once. So unless they are ..... dreamers, they know they must provide now.

About my experience: when a client puts extreme pressure on me, I tend to be more cautious and choosing the "safest" options instead of the best tailored for him.

When one has the "i really trust you're doing your best for my best" approach, I have the tendency to try and move mountains for that case.

I agree totally with your observations, and even embrace a bit your frustration, but i'm still hoping.

Don't worry about them devs knowing our opinions: the forum is closely watched.

So my take is: they promised what they promised in the blogs. If that announcements are going to be fulfilled, till the end of the year, then it's fair & square, imho.

To help getting that, I would go for the supportive approach, instead of the critical one, since the latter already failed. 😃

 

You are right here and we all should hope and support. But this is what i am doing since 14 years know. I am a developer myself and made a lot of community stuff for the original IL2. I know what is easy possible and what is hard to do. You dont have to build the perfect AI to make an interessesting game, it is about compromise like they did for the B29 formations now. I fly since 1995 now and have seen many good AI since then. Why IL2 struggle so much with AI has to be a core problem and when it would be easy to fix they would have done it 2012.

 

[FlightBud]

5 hours ago, AndreiTomescu said:

Korea it's not released yet. This is just an early glimpse. It would be fair to make an opinion after release, not just yet. Also if you read the dev's blogs, you'll know what to expect, and then , if expectations aren't met, you'll have every right in the world to be disappointed. Even angry.

To hyper resume the blogs, the hole game will be worked on till the end of the year.

I hear you: now it's not The next Generation, Star Trek. We all wanted and hoped for that, also hyped by the devs blogs that presented things as such. (the next gen). But they have the ground to be so. It will take a heap of work from them and some good amount of patience from us. Are you ready for that? Am I ready for that? Well, what else to do ? Constant pressure and bad vibes will slow the process, imho. I'm telling you from my own experience.

This early release is actually a stupid idea, since most players don't read/understand, are disappointed, and make bad opinion about something that's not even released. Rockstar way, imho, is the smart way.

But also, you might be quite a performant flyer, and then of course not even the improved AI aircraft will present any challenge to you. Probably that PvP is for, actually.

I ain't saying that the AI is much different from GB now, 'cause it ain't. Fine ideas were developed here on how to improve. Also the SP content it's almost barren. And the game is full of annoying bugs. And to get decent image you need a powerful rig. And that's expensive. And the optimization, especially for VR sucks. And it crashes to desktop, quite often. And the trees are still turning. Did I cover it all ? Guess not.

But the game is not released yet. 

Let's give these guys a fighting/working chance, before being disappointed. It's their big interest to please the players, since the product must sell well. 

Just my opinion.

I really hope you are right because I agree with you and want the game to succeed so that I can enjoy it for years to come.

I fly in VR and at this point, the game would need to go from what I currently consider a disaster and make massive improvements by release. My assessment of the game isn’t based on the many small annoying bugs - those can be fixed - my assessment comes from the unplayable VR, terrible optimization and performance, and an AI which everyone knew is terrible in GB, and to just come out with the same terrible AI in what is supposed to be a next-gen game is less than adequate. Are they going to make magic in the next month and come out with actually decent AI by release or am I going to continue to chase planes in a perpetual loop that defy the laws of physics?

And no, I haven’t read and listened to every blog… but I shouldn’t have to… if I pay $140 for a founders edition I expect something that’s decent… so I really really hope that the early release is just some unfinished glimpse of what a great game will be, but I’m realistic and know that that’s not how things work and suspect we’ll get a slightly more bug-fixed game with the same terrible performance.

After 1000 hours in GB that I’ve thoroughly enjoyed I really had my hopes up. Again, I hope I’m wrong and this game becomes the great sim it had the potential to be.

 

[gecko]

4 hours ago, Steini18778 said:

We know that the AI can be much better with a good mission design. @AEthelraedUnraed told us years ago and you are probably right. But what is the point when i start a campaign on my own computer and won't have any of those observations? If you have those experiences then share them with developer. Something is wrong and you can help solving it.

When the game is released in 4 weeks with the commander mode we will see what will happen when AI hits in. The commander mode depends only on a well made AI. 

What I was meaning is that the AI has various capabilities, for instance, deciding to return to base, but what triggers that decision needs work. It has been possible in missions to improve AI behavior in the way 357th has, but the developers need to incorporate that kind of thing internally so it is available by default. Presumably if they are doing it in their own code, what is possible would exceed what talented end users have been able to achieve.

[AEthelraedUnraed]

On 7/4/2026 at 10:13 AM, Steini18778 said:

We know that the AI can be much better with a good mission design. @AEthelraedUnraed told us years ago and you are probably right. But what is the point when i start a campaign on my own computer and won't have any of those observations? If you have those experiences then share them with developer. Something is wrong and you can help solving it.

When the game is released in 4 weeks with the commander mode we will see what will happen when AI hits in. The commander mode depends only on a well made AI. 

Yes, well, there are several types of AI. There's the flying and combat AI, which in GB despite some obvious shortcomings I think is pretty decent. Then there's a tactical AI, which is essentially nonexistent and is supposed to be handled by the mission designer. The advantage is that an experienced mission writer can have the AI do whatever he wants, whether that's a full kamikaze fight to the death, RTB at the first sign of trouble, or something in between. The obvious disadvantage is that it requires an experienced mission writer, and even then it's hard (and time-intensive) to get it right.

I hope that they're able to find some kind of middle ground in Korea, where the default AI in quick missions etc. does do some tactical decision making such as retreat against a superior enemy, while a mission designer is still able to override this.

[AndreiTomescu]

28 minutes ago, AEthelraedUnraed said:

Then there's a tactical AI, which is essentially nonexistent and is supposed to be handled by the mission designer. The advantage is that an experienced mission writer can have the AI do whatever he wants, whether that's a full kamikaze fight to the death, RTB at the first sign of trouble, or something in between. The obvious disadvantage is that it requires an experienced mission writer, and even then it's hard (and time-intensive) to get it right.

So this is the "magic" that made the difference in GB between Career (👎) and a good scripted campaign (👍) !!!

Thank you, @AEthelraedUnraed, really thank you, you really made me happier now that I finally know why the same AI can be both lame and game. 

[Jade_Monkey]

I fought a 4v4 skirmish with P80 vs Mig15 and was pleasantly surprised how the migs were using their energy and fighting on the vertical instead of just turning or getting stuck in some loop.

 

On a separate skirmish, yaks vs P51, I had about 5 of them of my tail and with a gentle turn they couldn't make the deflection shots even if their lives depended in it, but they chased me to no end.

All this to say there is some hope but also there are some remnants of GB issues that need to be ironed out.

[AndreiTomescu]

4 hours ago, AEthelraedUnraed said:

The obvious disadvantage is that it requires an experienced mission writer, and even then it's hard (and time-intensive) to get it right.

I hope that they're able to find some kind of middle ground in Korea, where the default AI in quick missions etc. does do some tactical decision making such as retreat against a superior enemy, while a mission designer is still able to override this.

Sooo....what you say is that it would be possible for the devs to implement a kind of... "AI BlackSix" that would fulfill the tactical AI part in a semi dynamic/ career mode and that might bring the new career mode at the aprox level of a good scripted campaign?

@Jade_Monkey: while patiently but anxiously waiting for "The Fall of Seoul" scripted campaign 🙂 , could you tell me where did you find those skirmish missions? And if you made them yourself, could you pretty please share them, as to have some SP content upon the near future (4th aug) ? Thank you!

[AndreiTomescu]

P.S. I found where to get skirmish missions. Stupid me.

But i'm still hoping for that "The fall of Seoul".....the Velike Luki version is one of my favourites from GB

[bsones]

Random question: Does the AI accumulate G fatigue like the player? It kind of seems like they don't, but maybe they are just better at managing it than I am. 

[AEthelraedUnraed]

2 hours ago, AndreiTomescu said:

Sooo....what you say is that it would be possible for the devs to implement a kind of... "AI BlackSix" that would fulfill the tactical AI part in a semi dynamic/ career mode and that might bring the new career mode at the aprox level of a good scripted campaign?

Well, an "AI BlackSix" is what they tried in the GB career. This quickly becomes very complex since in a mission there's just too much going on. If you're making just one mission, it's doable to finetune the AI so that it behaves like you want, but if you programmatically need to come up with a mission logic that works in every case, it becomes very hard.

What I rather hope for, is an actual AI that periodically evaluates what to do. E.g. if they're clearly losing a battle, they should try to disengage. If they're clearly outnumbered, they should not engage a superior enemy. If they spot the enemy early and they have the opportunity, they should try to engage from the sun. Things like that, that are not directly related to combat performance, but do change the way they approach battle.

The good thing is that IL2's mission system already supports such things. Ideally, the above describes how the AI would behave with a waypoint's priority setting set to "low". That gives a mission writer the opportunity to use the "medium" or "high" settings to override the AI behaviour as desired, to force certain events to happen.