Playing Burning Banners vs AI
I decided to play Campaign 16 - Assault in the North against ChatGPT to learn how effective the AI could be at learning the system and playing the game AND teaching myself the Advanced Rules for Magic, Blessings, and Monsters. I love the art and vibe of this game, and getting to play more of it would be great!
Campaign 16 is a Scrolls (stand alone) scenario set in the year 599 near the end of the War of the Burning Banners. The Campaign starts in Spring 599 with the turn order of:
Fjordland, Army of the Night, then Oathborn.
The Oathborn and Fjordlanders are closing in on the Spire and attempting to put the nail in the coffin containing the Army of the North. I decided how to lay out the initial builds, focusing on putting pressure on the Spire with the Stormcaller leading the Oathborn and expanding in the Southwest with Astrid leading the Fjordland, while defending the Spire with Lilith and putting pressure on the Dwarves in the Western mountains with Kali from the Army of the Night.
After Setup
I took the role of the resistance, an alliance of the Oathborn and the Fjordland, and assigned the Army of the Night to ChatGPT.
Spring 599
As the Fjordland took their first turn, Astrid led a liberation of Adakirk and Felsoft along Hallgeirr Bay. An group of Berserkers attacked a Basilisk and were wounded.
I described the board state, took pictures of the board and the hand of cards in the Army of the Night hand, and uploaded the Undying Rules to ChatGPT.
ChatGPT asked for more information about its heroes, and told me it could not read its hand. I provided a description of the hand and cards in it and the heroes. It assessed my position in the southwest as extended, and described taking action to disrupt the West while consolidating defense in layers around the Spire.
The strategy made sense, but in specific execution there were logic gaps and it was clear some of the rules were not understood. Some suggested activities would be huge gambles or impossible given the situation. Initially stacking rules were misunderstood, and positioning was described that would be ineffective.
I adjusted these actions to try to match the intent expressed by ChatGPT.
The AI built around the Spire and attempted to put pressure on the Dwarven strongholds in the North and West. Kali was more successful in the Mountains by taking Barzirak for the Night.
The Oathborn spent the Spring closing on the Spire with Stormcaller, Oathtakers, and Storm Giants. An army of dwarven pikemen also marched from Highgarden toward Barzirak.
Summer 599
As the Summer began, Fjordland armies moved up the Njalselva River, taking Khorikar and Norstead and setting up to be able to assault the Spire in the Autumn.
Summer revealed some issues with the AI as an opponent. It wanted to delay, keeping spells in reserve and repositioning. I had to fill in the tactical gaps, but not as much as I would have in a multi-handed solo game. The Army of the Night attacked weaker Alliance armies and tried to block off access to the Spire. Kali continued to be a standout Hero and routed another Oathborn army in the Glitterfeld near Barzirak.
The Oathborn Storm Giants assaulted Lilith, who escaped a defeat of her Vampire Knights and fled to Barlas Lake with a unit of Ghosts. They built a catapult unit and recruited another hero, Dara Firemane. Most Army of Night defenses have been removed, but the dwarves failed to put a dent in the Spire itself.
Autumn 599
The Fjordlanders rebuilt the hamlet of Far Tumed and used it to build a battering Ram siege Engine {I think I broke the rules here}. The Fjorland recruited Lieve Eriksdottir, and she threatened to liberate The Bjornfoss from the Army of the Night in the West. Despite the presence of the battering ram and catapults, the Fjorlanders lost a unit of Berserkers to a frontal assault on the Spire and a second army was weakend.
In desperation, the Army of the Night elected to conduct a Ritual of Speed and assault the Storm Giants with Lilith and her army of ghosts. In a shocking upset, she eliminated the Giants!
Dara Firemane rallied a second unit of Giants together. After Oathtakers cleared away the giant bats, Dara cast a Ritual of Power on the Stone giants, and they overwhelmed Osterloch who got 6 successes, but the Giants had a whopping 8! The game ends with the Oathborn in possession of the tower at the end of 599. The key was obviously the siege engines passive ability that canceled the magical fortifications of the Spire (reducing the number needed for a success from a 7 to a 5 because both the Oathborn and Fjordland engines were adjacent to the Spire.
The final roll of 6 successes by Osterloch (on 6 white dice and 2 black dice added by a spell) and 8 successes by Storm Giant army (on 3 black and 6 white dice, pumped by the Ritual) were both very strong rolls. The battles can be swing-prone, but the narrative comes through easily.
Analysis of ChatGPT in this role
Positives:
- Strategy and narrative explanation by the AI were strong.
- A narrative of the Campaign developed naturally, as it would with 2+ players.
- I gained experience using the rules for Magic.
Negatives:
- Tactically, the AI was weak to just wrong about some rules. Perhaps this could improve on further plays of the same system, but I am not holding my breath at this point.
- Timing and purchase/placement decisions were left almost entirely to me.
- Monsters were not interacted with much, and only by me.
Conculsion
I would use ChatGPT again, but with low expectations. I am eager to try again with the AI ‘Gemini’ and what differences there are.
Comments
Post a Comment