Remember the thrill of Stratego? That classic board game where you moved your hidden army, trying to guess your opponent's most powerful pieces while keeping your own a secret. It's a game of bluffing, strategy, and pure deduction, much harder than chess in some ways because you never truly know what's coming. This wasn't just a game; it was a battle of wits, where information was power.

For decades, experts believed a computer could never truly master Stratego. The hidden information, the need for human-like intuition and trickery, seemed like a wall AI couldn't climb. While the world cheered for AIs conquering chess and Go, a quiet revolution was brewing behind the scenes, tackling this ultimate game of incomplete knowledge. This is the strange story of how AI finally conquered Stratego.

The Hidden

Challenge of Stratego: A Game of Secrets

Stratego isn't like chess or checkers. In those games, both players see everything on the board. You know every piece your opponent has and where it is. Stratego is different. Each player has 40 pieces, from generals to bombs, but only you know what your pieces are. Your opponent only sees their backs, a sea of identical blue or red squares.

This "fog of war" makes the game incredibly complex. You have to guess what your opponent's pieces are based on how they move, how they attack, and even how they *don't

• attack. Is that a high-ranking general moving cautiously, or a lowly scout pretending to be important? It's like a spy game, where every move is a clue, and every attack is a gamble. The goal is to capture your opponent's flag, hidden among their pieces.

Why AI Struggled with Stratego for So Long

For many years, artificial intelligence conquered games with perfect information, like chess and Go. These AIs could look millions of moves ahead and calculate the best path to victory with incredible speed. But Stratego presented a fundamentally different kind of problem. With hidden pieces, an AI couldn't just "see" the best move. It had to deal with uncertainty and deception, making it a much tougher nut to crack.

Traditional AI methods relied on knowing the full game state. In Stratego, that's impossible. Imagine trying to plan a perfect strategy when half the battlefield is invisible. Early attempts by computers often failed because they couldn't bluff, couldn't adapt to hidden threats, or couldn't learn to value the act of gathering information itself in a human-like way. The game demanded intuition and a knack for trickery, qualities thought to be uniquely human.

DeepMind Steps Up to the Board

Then came DeepMind, the company famous for building AIs that beat world champions in Go and chess. After these successes, they decided to tackle Stratego, a game that represented a significant new hurdle for AI research. Their goal wasn't just to win a few games, but to truly *master the complex strategies

• of imperfect information, pushing the boundaries of what AI could achieve.

They created an AI system called DeepNash. Unlike previous game-playing AIs that sometimes learned from human games, DeepNash started from scratch. It didn't watch grandmasters play Stratego. It didn't study human tactics or opening moves. Instead, it learned entirely by playing against itself, millions and millions of times, constantly refining its understanding of the game's hidden depths.

How DeepNash

Learned the Art of Bluffing and Deduction

DeepNash used a special kind of learning called reinforcement learning. Think of it like a child learning to ride a bike. They try, they fall, they adjust, and eventually, they get better. DeepNash did this at an incredible speed, constantly tweaking its strategies based on what worked and what didn't. It was like having billions of practice games, all against an equally intelligent opponent.

The AI developed a deep understanding of Stratego's core mechanics, including how to set up defenses, launch attacks, and most importantly, how to *bluff effectively

• and deduce opponent's pieces. It learned to trick its digital opponents, making them believe certain pieces were in different places or had different ranks. This ability to deceive and to infer was a huge breakthrough for AI, moving beyond simple calculation to more sophisticated strategic thinking.

Valuing Information Over Immediate Gains

One of DeepNash's key breakthroughs was understanding the value of information. Sometimes, it's better to make a seemingly small move, or even sacrifice a low-ranking piece, just to learn what your opponent has. The AI learned to make these kinds of calculated risks, recognizing that knowing your opponent's general location, or confirming a bomb, was more valuable than winning a minor skirmish. This strategic depth mirrored the play of top human Stratego players.

"To truly master Stratego, an AI must not only be smart, but also a master of deception, able to play the mind of its opponent, and understand the hidden value of every piece of information."

The AI's Grandmaster Performance

After countless hours of self-play, DeepNash reached a level of skill that surprised even its creators. It played against the best human Stratego players in the world and performed at a grandmaster level. This wasn't just about raw computing power; it was about the AI developing strategies that were creative, unpredictable, and deeply tactical. It consistently outperformed other Stratego programs and even the best human players.

The AI showed a remarkable ability to adapt. If its initial plans didn't work, it would quickly change its approach, exploiting weaknesses in its opponent's hidden setup. Its play was both aggressive and subtle, a combination rarely seen in computer opponents before. DeepNash didn't just win; it won with style, demonstrating a nuanced understanding of the game's psychological elements.

Beyond the Board Game: Real-World Impacts

Beating Stratego might seem like just another game for AI, but it's much more important than that. The techniques DeepNash used to conquer a game of hidden information have huge implications for the real world. Many real-world problems also involve imperfect information, where you have to make decisions without knowing everything.

Think about situations where you don't have all the facts, and you need to make smart choices:

• Negotiating a business deal where the other side has hidden goals.

• Planning military strategies where enemy positions are unknown.

• Managing supply chains with unexpected disruptions and limited visibility.

• Making medical diagnoses when a patient's full history isn't immediately clear.

• Even designing economic policies when future market conditions are uncertain.

The lessons learned from DeepNash could help AIs in these complex, uncertain environments, leading to better decisions and more efficient solutions in critical areas.

What This Means for the

Future of AI

DeepNash's victory in Stratego showed that AI can learn to handle situations where information is incomplete and deception is key. This opens up new possibilities for how AI can help us solve some of the world's toughest problems. It moves AI beyond just crunching numbers and into the realm of strategic thinking, even when faced with unknowns. It proved that AI can learn to reason and strategize in ways that mimic human intuition in games of hidden information.

This achievement reminds us that the potential of artificial intelligence is still growing. It's not just about speed or power, but about developing a deeper, more human-like understanding of complex, uncertain environments. The strange story of AI mastering Stratego is a testament to this ongoing evolution, showing us that even the most human-like games are within reach of smart machines.

The next time you set up your Stratego board, remember the hidden genius that learned to play its secret game better than almost anyone. It's a quiet victory, perhaps not as famous as other AI milestones, but one that truly changed what we thought computers could do. The game of hidden armies taught us a lot about the future of intelligent machines, and the strange ways they learn to outsmart us, even in the fog of war.