AI - Can developers address how it is being handled. Can we expect a better AI than in the Planetfall version?

  • We have updated our Community Code of Conduct. Please read through the new rules for the forum that are an integral part of Paradox Interactive’s User Agreement.

druez

First Lieutenant
74 Badges
Jul 19, 2008
235
1.145
  • Crusader Kings III
  • Crusader Kings II
  • Europa Universalis IV
  • Stellaris
  • Imperator: Rome Deluxe Edition
  • Crusader Kings II: Holy Fury
  • 500k Club
  • Hearts of Iron IV: No Step Back
  • Hearts of Iron IV: By Blood Alone
  • Victoria 3 Sign Up
  • Stellaris: Nemesis
  • Stellaris: Necroids
  • Europa Universalis 4: Emperor
  • Battle for Bosporus
  • Imperator: Rome - Magna Graecia
  • Stellaris: Federations
  • Hearts of Iron IV: La Resistance
  • Europa Universalis IV: Cradle of Civilization
  • Hearts of Iron IV: Cadet
  • Deus Vult
  • Hearts of Iron IV: Colonel
  • Europa Universalis IV: Rights of Man
  • Stellaris: Digital Anniversary Edition
  • Stellaris: Leviathans Story Pack
  • Hearts of Iron IV: Together for Victory
  • Crusader Kings II: Monks and Mystics
  • Europa Universalis IV: Mandate of Heaven
  • Hearts of Iron IV: Death or Dishonor
  • Age of Wonders III
  • Crusader Kings II: Reapers Due
  • Hearts of Iron IV: Expansion Pass
  • Stellaris: Humanoids Species Pack
  • Stellaris: Apocalypse
  • Europa Universalis IV: Rule Britannia
  • Stellaris: Distant Stars
  • Europa Universalis IV: Dharma
  • Europa Universalis IV: Golden Century
  • Imperator: Rome Sign Up
  • Stellaris: Ancient Relics
  • Stellaris: Lithoids
  • Crusader Kings II: Conclave
  • Europa Universalis IV: Cossacks
  • Crusader Kings II: Horse Lords
  • Europa Universalis IV: Common Sense
  • Crusader Kings II: Way of Life
  • Europa Universalis IV: El Dorado
  • War of the Roses
  • Crusader Kings II: Sunset Invasion
  • BATTLETECH: Flashpoint
  • Stellaris: Megacorp
I'm so excited for everything I've read about this game so far. But, to me the AI is almost a make or break. Can the AI give us a competent game in a solo game? I want a competent game without the AI getting 20 different bonuses. I want an AI that plays fair, but it also good at the game.

Can we get a dev diary on how the AI will behave and the challenge level?
 
  • 3Like
  • 2
Reactions:
you are not the only one with concerns about the AI.
the more is revealed about AOW4 and the more i am excited to see it released as everything shown so far is great!
but a bad AI can really destroy a well done game.

for that matter, planetfall AI was horrible as it was both scared of attacking decent sized armies led by the player even with overwhelming total forces AND sucked at modding its troops correctly. (or just at modding as alot of the AI's troops were always missing some key mods.)

i made a thread about it here:
 
  • 3Like
Reactions:
Perhaps we can make a survey after 3 months of playing, how the community sees the AI and what would be nice to add/modify/remove for a better game. Perhaps we are even happy, let's see ;) But if not such a survey would be a good path to happiness :)
 
  • 2
Reactions:
The AI is going to be bad because the AI is always bad. They're probably going to do what they can but people just has to accept (as I have, after playing strategy games for like 30 years) that AI is always bad.
 
  • 3
  • 2Like
  • 1
Reactions:
The AI is going to be bad because the AI is always bad. They're probably going to do what they can but people just has to accept (as I have, after playing strategy games for like 30 years) that AI is always bad.

I echo this sentiment, the AI is a huge undertaking to fix and the crowd that'll sit at highest difficulty and still want more isn't that big I'm afraid. I expect we'll see AI mods that nudges the AI more into an aggressive position and harder for the more hardcore strategists.

We'll need more complex AI from outside sources to start posing a challenge and able to mimic human players, I doubt game developers will ever be able to afford to create good enough AI themselves alongside having to make a game. The former is still in infant state though and the AI still have to be programmed to learn a specific game for a long time, so we're still a couple of decades or so away before that may become mainstream and affordable by developers. But strategy games is where these AI have been tested the most, mainly real time, but turn based should work if we look at boardgame AI like chess.


Otherwise I do also have the concern for release. Planetfall's release AI had similar problem of turtling a lot. It could PvP itself just fine cause I'm guessing the AI being set on strategies and not adapting makes it easier to deal with itself, but against a player there's so many variables and unknowns about human behavior, able to adapt and change strategy on the fly, which the AI being mainly reactionary don't know how to respond to, nevermind long term goals that it doesn't lock into but can adapt and change as the game goes on.

Hopefully the more apparent issues is something that can be addressed quickly, as I recall the AI became a fair focus on Planetfall and the last time I played it, albeit being fairly rusty, the AI did pose some challenge and was fairly aggressive, even opting to sacrifice colonies to sneak around and steal my capital with its ruler. It does still have the problem of not really able to regain itself, throwing out weak armies, didn't really build garrisons and turtled up after losing one major battle but spreading too thin to defend anything.

I hope the new siege system alongside the new way of how battles are done (limited to 3v3 of strongest armies rather than allowing you to 4v3 or 5v2 the AI) and separate unit queue will address some of the problems when dealing with an AI turtling, not able to make a good enough comeback and how it weights and decides to be aggressive should they fix the initiate passiveness that's been reported which I got a feeling we'll see on release but hopefully quickly addressed in patches.
 
The AI is going to be bad because the AI is always bad. They're probably going to do what they can but people just has to accept (as I have, after playing strategy games for like 30 years) that AI is always bad.
I don’t agree. The AI is not always bad, for example it is not always bad in tactical combat in AOW. The AI on the strategic map in Planetfall is bad in the sense that is too passive. And we should try to improve this 'badness'; that is the way to go instead of accepting that AI is always bad.
 
Last edited:
  • 3Like
  • 2
  • 1
Reactions:
The AI is going to be bad because the AI is always bad. They're probably going to do what they can but people just has to accept (as I have, after playing strategy games for like 30 years) that AI is always bad.
And this comment does help us how? We all want to get as good a game as it is possible. I had thousands of hours fun with AoW3. The AI is good enough there to provide me with fun and that is all I ask for. A survey could help make it even better after lots of people have played it.
 
  • 2Like
Reactions:
that's right. the AI is good enough in AOW3. but fares poorly in planetfall.
but not only on the strategic map but also on the tactical map. the moment you had a good enough army you could take on 4 AI armies with one because the AI never regroup its troops.
i stopped playing planetfall when i realised how stupid and passive the AI were. it ruined me the game.

a good way to improve the AI would be:
_be more aggressive after a declaration of war. don't hesitate to attack. preferably weak area's first.
_avoid declaring wars when it's too far away or it doesn't have any alies nearby you. or if its strenght is definitively not on par.
the Ai shoulodn't declare war for the sake of it but when it is ready to attack you.
_consider its allies into its own battle power.
_coordinate itself with other allied AI.
_attack weak settlements with overwelming forces if nearby. (no more sending a few troops to take out a weak city just to getwrecked because as a player i can better manage my troops in battle.)
_regroup its troops on the battlefield before charging mindlessly.
_waiting behind walls instead of getting aggroed by one spell and going out of the walls.

at least it will not have to deal with a modding system.


Perhaps we can make a survey after 3 months of playing, how the community sees the AI and what would be nice to add/modify/remove for a better game. Perhaps we are even happy, let's see ;) But if not such a survey would be a good path to happiness :)
some game journals already made a review about the game and one of its main criticizme was that the AI was too passive. i got nasty planetfall flashback and was urged to make a post about it.
better to assume that something isn't working the best so that it gets "fixed" for when the game launches.
 
  • 2Like
Reactions:
I had thousands of hours fun with AoW3. The AI is good enough there to provide me with fun and that is all I ask for.
Same here. Almost 3000 AOW3-hours. AI is good enough there to provide fun (which cannot be said about PF, sadly).
However there are at least 2 areas where the AI on the strategic map in AOW3 can improve imo.

1. Different armies not moving together
This issue makes armies vunerable because of the Adjacent hex-rule.
Good thing is, this rule is replaced by the Reinforcement-rule in AOW4 as was explained in the last stream!

2. AI does not take into account when it plays in a team/if it's allied
Simply put: AI does not play as a team. Although I think it is extremely difficult to make AI really play as a team, there is a rather simple general principle:
if an AI-leader plays in a team, it can take more risks, because it is about winning the war as a team of allies, and not about preventing loosing individual armies.


This can be implemented (I think) with a different army comparison where it is about weighted losses. This means:
AI is more willing to accept losses, because losses are relatively less impactful for a team of AI's.

This should make AI more willing to attack & to move to human targets, making it more aggressive. This makes sense when it plays in a team. This makes also sense on higher difficulties AI; where it usually has more resources to build larger armies. I explained it in a bit more detail in this thread incl. example.

A survey could help make it even better after lots of people have played it.
Sure. I hope both devs and players keep an eye on the topic. I became a bit less pessimistic last week because of the Reinforcement Rule and Jordi explaining in this thread that the AI can easier understand province-improvements and how to best use them.
 
I echo this sentiment, the AI is a huge undertaking to fix and the crowd that'll sit at highest difficulty and still want more isn't that big I'm afraid. I expect we'll see AI mods that nudges the AI more into an aggressive position and harder for the more hardcore strategists.

We'll need more complex AI from outside sources to start posing a challenge and able to mimic human players, I doubt game developers will ever be able to afford to create good enough AI themselves alongside having to make a game. The former is still in infant state though and the AI still have to be programmed to learn a specific game for a long time, so we're still a couple of decades or so away before that may become mainstream and affordable by developers. But strategy games is where these AI have been tested the most, mainly real time, but turn based should work if we look at boardgame AI like chess.
It would be great if resources are available to train a neural-net AI based on alphago architecture.
However this is a significant investment of resources.
 
I don’t understand why people always ask this and pressure devs into wasting a DD on it.

what are you expecting them to say? “AI is actually way worse than previous titles”?
 
  • 4
  • 3Like
  • 1Haha
Reactions:
I don’t understand why people always ask this and pressure devs into wasting a DD on it.

what are you expecting them to say? “AI is actually way worse than previous titles”?
Umm because AI in a strategy game is one of the most important features of the game, especially in solo play. I would like for them to discuss what they learned from their previous game and what they are doing to make this one play better. I mean based on what you say, what is the point of them talking about anything?
 
  • 2
  • 1Like
Reactions:
Umm because AI in a strategy game is one of the most important features of the game, especially in solo play. I would like for them to discuss what they learned from their previous game and what they are doing to make this one play better. I mean based on what you say, what is the point of them talking about anything?

One thing I don't often see brought up in these kinds of discussions is actually how important the AI is to the common player. It's easy for more experienced folks who have thousands of hours in these kinds of games to say the AI is bad or not challenging, but unless its particularly egregious do you think most people would even notice? Now, I'd like to say I have decent experience, I'm just not very good, so for me I found the AI in AoW3 was serviceable. In terms of difficulty it had the scaling I'd have wanted, my main complaint was more about its diplomatic behaviors which it seems the devs have made much better in PF and now in AoW4, so I'm fine with it.

AI is very difficult to create and improve upon, and Triumph is a small studio you must understand. They might not see it as worth putting too much effort into, and most devs don't. Total War is the same, they've had the same sort of AI for years, but it is serviceable.
 
  • 1Like
Reactions:
GPT4 says, in response to the question of how the AI in AOW4 can be improved:
"
As an AI language model, I don't have access to the specific implementation details of Age of Wonders 4 (AOW4) AI, but I can suggest some general approaches that can improve the AI in strategy games like AOW4.

  1. Enhance the decision-making process: The AI in AOW4 can be improved by making it more capable of assessing risks and benefits associated with different options available to it. This can be done by using advanced algorithms like reinforcement learning, decision trees, and Monte Carlo tree search.
  2. Improve the learning algorithms: By using machine learning techniques like deep learning, the AI in AOW4 can learn from its past experiences, which can help it make better decisions in future games. This can be achieved by training the AI on a large dataset of human player moves and strategies, which can help it learn how to play the game more effectively.
  3. Increase the complexity of the AI: The AI in AOW4 can be made more challenging by increasing its ability to adapt to different situations and to come up with creative solutions to problems. This can be achieved by increasing the number of possible actions available to the AI, as well as by making its decision-making process more sophisticated.
  4. Implement multi-agent systems: AOW4 can implement multi-agent systems, where multiple AI agents work together to achieve a common goal. This can simulate real-world scenarios and create a more realistic experience for players.
  5. Improve the AI's awareness: AOW4 can be improved by giving the AI better awareness of the game world and its surroundings. This can be achieved by using advanced algorithms like computer vision and natural language processing, which can help the AI understand the context of the game and respond more appropriately to different situations.
Overall, improving the AI in AOW4 requires a combination of advanced algorithms and techniques, as well as a thorough understanding of the game mechanics and player behavior.
"
 
  • 3Haha
Reactions:
When asked which is the most promising approach, it says:
"
It's difficult to say which approach is the most promising as it depends on the specific needs and goals of the game developers and the game itself. However, a combination of approaches is often used to improve the AI in strategy games like AOW4.

If I were to pick one approach that has shown great promise in recent years, it would be the use of deep learning algorithms. Deep learning has proven to be very effective in improving the performance of AI systems in a variety of domains, including game playing. By training the AI on a large dataset of human player moves and strategies, deep learning algorithms can learn how to play the game more effectively and make better decisions in future games.

Additionally, combining deep learning with other techniques such as reinforcement learning and Monte Carlo tree search can further improve the performance of the AI in AOW4. These techniques can help the AI make better decisions in different situations and increase its ability to adapt to different scenarios.

In conclusion, the most promising approach to improving the AI in AOW4 is likely to be a combination of advanced algorithms such as deep learning, reinforcement learning, and Monte Carlo tree search. However, the specific approach or combination of approaches used will depend on the specific needs and goals of the game developers and the game itself.
"
 
When asked which is the most promising approach, it says:
"
It's difficult to say which approach is the most promising as it depends on the specific needs and goals of the game developers and the game itself. However, a combination of approaches is often used to improve the AI in strategy games like AOW4.

If I were to pick one approach that has shown great promise in recent years, it would be the use of deep learning algorithms. Deep learning has proven to be very effective in improving the performance of AI systems in a variety of domains, including game playing. By training the AI on a large dataset of human player moves and strategies, deep learning algorithms can learn how to play the game more effectively and make better decisions in future games.

Additionally, combining deep learning with other techniques such as reinforcement learning and Monte Carlo tree search can further improve the performance of the AI in AOW4. These techniques can help the AI make better decisions in different situations and increase its ability to adapt to different scenarios.

In conclusion, the most promising approach to improving the AI in AOW4 is likely to be a combination of advanced algorithms such as deep learning, reinforcement learning, and Monte Carlo tree search. However, the specific approach or combination of approaches used will depend on the specific needs and goals of the game developers and the game itself.
"
The next question would be: Why is video game AI not yet written with deep learning algorithms:

"
Video game AI can be written using deep learning algorithms, and in fact, there are many examples of games that already use deep learning for their AI systems. However, it is also true that not all games use deep learning for their AI, and there are several reasons for this.

One reason is that deep learning algorithms can be computationally expensive, especially during training, which can make them impractical for use in some games, particularly those with real-time constraints. In addition, game developers may choose not to use deep learning algorithms for their AI because they require large amounts of labeled data, which can be difficult and expensive to obtain.

Another reason is that game developers may opt for simpler, rule-based AI systems that are easier to implement and understand. These systems can be effective in many games, particularly those with well-defined rules and objectives.

Finally, game developers may also choose not to use deep learning algorithms because they may be more difficult to tune and balance compared to rule-based systems, which can affect the overall gameplay experience.

Overall, deep learning algorithms can be a powerful tool for video game AI, but their use depends on the specific needs and constraints of the game in question.
"

This is probably the answer here too. They chose rule based systems as deep learning algorithm tooling is not yet sufficiently mature to easily create complex game AI with. There may be examples of games, such as Starcraft 2, where this is done. But these are still research like examples that likely took a lot of effort on the developers part. If they did use it they may very well run into all sorts of weird problems that makes the game less fun, such as the AI running units around endlessly. Rule based AI is then a safer choice from a business perspective.

TLDR: Wait for AoW 5.
 
  • 1Like
Reactions:
Ah, so the group that made a deep learning chat bot also made it prefer a deep learning AI approach. Somehow not surprised even if it wasn’t intended.

of corse we have no knowledge on player behavior note the years needed to fine tune the necessary information to make it viable.

proves it knows only that its a game and has AI. Personal I don’t like the idea of a learning AI for a video game, if the strategies change dynamically then it could go from fun to not and back again, making every time you start the game a gamble for weather or not it’s a good time.

this would be especially true if you have the algorithm focused on victory for max difficulty and then attempt to gimp it for lower difficulties.

and even if the live version of the game has locked in AI the developers would still needed to constantly retrain the AI, decided if this version has each level of difficulty right each time they update the balance of the game. Much less when a DLC is added
 
  • 1
Reactions:
Umm because AI in a strategy game is one of the most important features of the game, especially in solo play. I would like for them to discuss what they learned from their previous game and what they are doing to make this one play better. I mean based on what you say, what is the point of them talking about anything?
AI isn’t a feature, it’s hardware.

If you haven’t learned with the release on any sequel ever made here’s how it usually goes:

“In the previous title AI did x things wrong, in this title we’ve programmed it to take advantage of the flanking system, use stacks more effectively, blah blah blah blah blah”

Whether it actually works or not is highly variable. The devs can say whatever they want but AI programming is hard as hell so very often proposed solutions fall flat.
It’s all so boring… I’d rather they actually talk about all the actual game features we know next to nothing about or do more dev streams.

in any case the AI looked pretty damn good in streams, and Explorminate even praised the autobuild feature for cities. So, so far so I don’t think we have much to worry about
 
  • 2
Reactions:
The next question would be: Why is video game AI not yet written with deep learning algorithms:

"
Video game AI can be written using deep learning algorithms, and in fact, there are many examples of games that already use deep learning for their AI systems. However, it is also true that not all games use deep learning for their AI, and there are several reasons for this.

No. Videogame AI cannot be written using deep learning.

Deep learning AI can learn to play some videogames like a player, but almost no videogames have a large enough training dataset to begin the process of training them.

AlphaStar learned to play Starcraft 2 by watching nearly a million ladder matches played between humans and then played many millions more simulated matches between candidate agents, and it is not "in Starcraft 2" it's in a datacentre at Google connecting to the ladder as if it's a human player.

So tell the class: How is a deep learning AI going to watch a million matches of a game that isn't out yet?

Edit: Also these AIs aren't actually "smart". They can be beaten by patterns outside their training data. AlphaGo was good enough to beat grandmasters at the game, but someone absolutely clowned on it by doing something so dumb that even someone who had only had the rules explained to them wouldn't fall for it (just making a box around the whole board whilst making some distracting moves in one corner). Because it was so dumb and wrong as a strategy it wasn't in the training set because why even? but because it wasn't in the training data AlphaGo had no way to respond to it or even notice what was happening.
 
Last edited:
  • 2
  • 1Like
Reactions: