Towards an adaptive AI storyteller?

[Hypolite]

I open this topic because I feel like the AI storyteller should take into account the current situation of the player to create a steeping curve of difficulty, but currently I keep experiencing and reading exactly the opposite. My sense of difficulty is based on the number of raiders that attack during eclipse period. I'm of course only writing about Cassandra Consistent, because neither Phoebe Friendly nor Randy Random would raise such questions.

On one of my first games, I had 4 colonists (after recruiting the first raider sent as a tutorial), some wounded, power issues, few metal and few food, and the 2nd or the 3rd eclipse, I got attacked by well-armed 6 raiders (not just pistols). I barely made it, killing them all. The next eclipse, 8 more raiders came up, reducing the few turrets to rubble before banging on the doors.

On a subsequent game, I was off to a much better start, having a lot of food and metal, 4 healthy colonists and more turrets than ever. In the following eclipse, I had to fend off only small groups of two to four raiders, constantly recruiting up to 8 colonists. At that point I had all the guns I needed and I still got attacked by fewer raiders than I had colonists.

Reading Sire's playthrough comments about raider attacks, I feel like he has experienced the same imbalance as I did.

Being outnumbered and outgunned is not necessarily a bad thing, it can be by design, but I feel like the challenge is not consistent with player's situation. A bad situation is made much worse by large raider attacks, while a good situation is never really threatened by small attacks.

I wanted to know how Cassandra well evaluate the current player's situation and if other testers had the same experience as I and Sire did, e.g. large raider attacks when in dire situation and small attacks when well-off.

[Tynan]

Cassandra isn't adaptive at all right now. She follows the same difficulty curve every game (internally, she receives more and more "points" to spend on threats each dark season). However, you're correct that this doesn't always create great outcomes in the game. Sometimes you'll be way ahead of her, and sometimes she'll just annihilate you.

I really need to take a pass at putting in some serious adaptive detection of the players' situations and response. Technically Cassandra works "as designed" right now because she just does her thing like a classic video game ramp-up and it's your job to respond. But I think that her play experience could be improved with some degree of adaptiveness.

There is a question of whether that should be another storyteller or not. Do people want to keep Cassie's "consistent" approach, as a yardstick to test themselves against? Or is it not even worth retaining in the game?

[Jakadasnake]

There are a lot of possibilities here. I kind of like the thought that maybe you won't be sticking to the easy, normal, and hard as far as difficulties go. Right now there are 3 storytellers? It's definitely a good idea to have one that adjusts according to player progress, but then again it will probably make the game feel even more routine. Assuming the algorithm for progress < difficulty is perfect (in an ideal world), the player is eventually just going to know exactly what to expect based on how they are doing and possibly figure out ways to manipulate how the storyteller perceives their progress.

I think it's a good idea to add an adaptive AI, but at the same time give the power gamers a wall that they can really beat their fists against. Those of us that choose that daunting difficulty where there is virtually no chance for success will undoubtedly develop that fondness out of nostalgia for the days where gaming was truly punishing.

[Hypolite]

If it was a skill-based game, a fixed difficulty curve would be expected. However, this isn't the case for RimWorld. An adaptive AI would replace a fixed one. I don't think it's a big deal if people manage to "cheat" Cassandra to have an easier life, because there's no score, and Phoebe Friendly is there for people not looking for a challenge.

And whatever hard we set the AI, an adaptive one is still the best.

[British]

And whatever hard we set the AI, an adaptive one is still the best.

But then, it depends on how we define "adaptive": it could adapt by being more difficult, more easy, or anywhere in-between.
So, do would we want to have those 3 levels (or more) of adaptiveness for each AI, or stick with each AI having a "linear adaptiveness" (so to speak) ?

I'm not sure that it's crystal clear, or that it hasn't been said/implied already, so please bear with me

[Hypolite]

Well, Cassandra Consistent could have 3 children, Edna Easy, Norman Normal and Harvey Hard. A crude adaptive example would be, for the number of raiders by attacks : Edna always sends half the number of colonists you have, Norman sends the same number, and Harvey sends double the number. That way, if you lost people to fire of wounds, the game would get slightly easier whatever the storyteller. On the other hand, if you recruited/bought people, the game would get naturally harder.

Of course, other things could be taken into account. More turrets? More raiders. More cash? More raiders. More food? More raiders.

I can even think of Imelda Impossible who would increase the number of raiders sent exponentially taking into account the players' situation. For example, 1 colonist : 1 raider, 2 colonists: 3 raiders, 3 colonists: 6 raiders, 4 colonists: 9 raiders, 5 colonists: 12 raiders, and so on.

[British]

I was also thinking about the possibility of tweaking the AI during gameplay (possibly after a pre-defined time): You've been toying with Cassandra for a while but find her dull ?
Spice-up your relationship by going harder on her (or the other around).
Or go softer if you like it slow, of course.

(Should this post be flagged with "parental advisory required" ?!)

[nomadseifer]

Since I haven't had a chance play the game yet, my thoughts are more in the abstract.

Do people want to keep Cassie's "consistent" approach, as a yardstick to test themselves against? Or is it not even worth retaining in the game?

I say definitely yes, in regards to difficulty.  I think an adaptive AI that reads the player situation and playing style and adjusts the kinds of stories it wants to tell is a fantastic idea.  But making the AI adjust its difficulty based on player progress seems a bit like hand-holding when the player is weak and success-punishment when the player is strong. 

[Hypolite]

For me, it's not so much hand-holding or success-punishing as reducing the unfairness and the boredom. Being constantly hit by waves of raiders becomes no fun after a while when you're trying to hold everything together, as having a impregnable fortress for a while without really having to fend off for your life.

If you don't want any hand-holding, you can always go for Randy Random of a higher difficulty adaptive AI storyteller. If you don't want success-punishing experience, you can always go for Phoebe Friendly. The choices are open.

[Tynan]

Okay. I'm going to prioritize making Cassie a bit more observational for the next build or the one after. I do think it would improve the game. We'll see how that pans out with you guys.

[SpaceEatingTrex]

Okay. I'm going to prioritize making Cassie a bit more observational for the next build or the one after. I do think it would improve the game. We'll see how that pans out with you guys.

I think making Cassandra Classic more adaptive to the player's situation was the right choice. Hypolite made a good post:

For me, it's not so much hand-holding or success-punishing as reducing the unfairness and the boredom. Being constantly hit by waves of raiders becomes no fun after a while when you're trying to hold everything together, as having a impregnable fortress for a while without really having to fend off for your life.

What other than time is taken into account for calculating increasing the difficulty? Time should probably be the biggest factor, but I think if the game can check the amount of:

• Colonists
• Turrets
• Weapons
• Resources

then it could derive a more appropriate difficulty for the player based on those factors. A simple system could be setup by having a "Difficulty Score" value the different factors added to, for example a pistol adding 1, shotgun adding 3, etc. with higher values corresponding to higher difficulty. However, that's a simple implementation and might not provide a lot of flexibility.

What else would be a good way to implement adaptive storyteller mechanics?

[Spike]

Cassandra isn't adaptive at all right now. She follows the same difficulty curve every game (internally, she receives more and more "points" to spend on threats each dark season).

Cassie could be made more adaptive by simply giving her a percentage of the player's points; difficulty levels could be set at Easy (50%), Medium (100%), or Hard (150%).  I'm not sure what would qualify as points, but as the game is set now it would be a mix of:

• Number of colonists
• Weapons "owned" by the player (Equipped and in racks?  Within home zone?)
• Turrets built

Of course, that's just for throwing pirates at the player.  I'm not sure how you could handle other aspects.

[British]

You might as well add the resources on your list.
I've seen a lot of videos with people having $20K+, with food and metal in huge quantities as well.
That's probably only requiring some tweaking (as Tynan is already doing), but in any case, a way to alleviate that would be for the AI to target structures, for instance, as then the player would have to rebuild...
But that's most likely what we'll see once the shitload of events get incorporated into the game.

[starlight]

Cassandra isn't adaptive at all right now. She follows the same difficulty curve every game (internally, she receives more and more "points" to spend on threats each dark season). However, you're correct that this doesn't always create great outcomes in the game. Sometimes you'll be way ahead of her, and sometimes she'll just annihilate you.

I really need to take a pass at putting in some serious adaptive detection of the players' situations and response. Technically Cassandra works "as designed" right now because she just does her thing like a classic video game ramp-up and it's your job to respond. But I think that her play experience could be improved with some degree of adaptiveness.

There is a question of whether that should be another storyteller or not. Do people want to keep Cassie's "consistent" approach, as a yardstick to test themselves against? Or is it not even worth retaining in the game?

I think the current version of Cassandra seems more suitable to be a Kassandra: a toughie.

But it wouldn't be bad to have her in the game, to check yourself aganist a pre-decided curve. (But then it would be ideal to have an option for easy / medium /hard)

The new version of the story teller would be more interesting and....... slightly difficult to make.
People should not get the feeling that they have to limit themselves becuase of story-teller limitations.

Compare with the fact that in Dwarf Fortress people are focused on staying within a certain wealth limit since going above that triggers seiges.
Not that would be something interesting you can implement.