AI Diplomacy
What Would Happen If (different) AIs Ran a Country? (PART 2)
Right after posting, “What would happen if AI ran a country?”, I found quite an interesting article about a roulette between different AIs, also known as “AI Diplomacy”.
What AI diplomacy means is that several different AIs were put into a classic strategy game called Diplomacy. It’s this old game from the early 1900s where countries like France, Germany, and Russia compete to take over Europe
But here’s the twist: instead of humans playing/fighting, they made different AI models play against each other.
Pretty creative, isn’t it??
Call it the battle of the bots if you will!
— Intro: The Game
Here’s how it worked: each AI controlled one country and had to negotiate, make alliances, and sometimes straight-up lie to win. There were no dice rolls or random luck, just pure strategy.
The goal was the first one to take over 18 supply centers wins.
They could send each other private messages, team up, and even backstab each other!
The goal is simple: take over Europe.
WHAT EACH AI DID:
— Gemini 2.5 Pro
Gemini was actually quite good at making smart moves and building alliances.
It wasn’t all about drama
It actually had a solid plan.
It was so close to winning until o3 convinced everyone else to turn on Gemini last minute.
So instead of winning, Gemini got completely ganged up on.
Still, it was one of the only AIs other than o3 to win a game.
— The Big Guns: OpenAI (o3)
Ok, o3 didn’t just play the game, it played the players in it!
While other models were trying to make alliances, o3 was scheming and plotting everyone’s downfall !
If there was a shady move to make, o3 probably already made it five turns ago.
There’s this part in the article where o3 says in its private diary:
"Germany (Gemini) was deliberately misled. Prepare to exploit German collapse."
UM EXCUSE ME???
It pretended to be Germany’s best friend while secretly planning to destroy them.
And then it actually worked!!!!
Diabolical.
But it didn’t stop there. o3 made deals with AIs just to break them. It even convinced other AIs to gang up on Gemini just to stop it from taking the lead. And after the “alliance” helped o3 stop Gemini, guess what?
o3 turned against them too!!
If that’s not betrayal i don’t know what is.
— DeepSeek
DeepSeek didn’t just play the game it put on a show too!
This AI had the most dramatic responses like:
“Your fleet will burn in the Black Sea tonight.”
It would, however, change its vibe depending on the country it was playing.
Sometimes it was chill, and other times, it was threatening war.
It didn’t win a lot, but it came extremely close a few times.
For an AI that’s far more cheaper than o3, that’s impressive!!!
DeepSeek was like that chaotic player who adds fire to the oil to the chat for no reason and i love it hahaha!!
— Llama 4: Good but Bad??
Llama 4 wasn’t the biggest or smartest model in the game
But somehow, it was pretty good
It didn’t win any games, but it was great at making alliances and planning sneak attacks which I believe was very underrated.
When reading the article, I thought of Llama as the small kid in dodgeball who’s oddly hard to hit and somehow makes it to the final round every time
— What Was The Purpose ?
The purpose of the battle of the bots, was that it was all meant to see how different AIs behave in stressful and competitive situations
And in my opinion,
W. O. W.
I never expected AIs to be this smart and ruthless!!!
It really showed how complex they can be.
They weren’t just answering math questions or writing poems, they were forming friendships, breaking them, lying, manipulating, and making power moves!
If that’s not entertainment then I don’t know what is!!!
And let’s not forget to give props to the actual author!
LINK: We Made Top AI Models Compete in a Game of Diplomacy. Here’s Who Won.
And that’s a wrap, see you next post!