Has AI Conquered the Board Game 'Stratego'? How to Read Hidden Information

Most of the AI news we encounter frequently involves defeating human champions in games like Go or chess. However, these games share one common feature: all the pieces on the board are visible. Since you know all the pieces the opponent holds on your turn, the AI only needed to calculate well to win.

But imagine this. What if you were playing a card game and had no idea what cards your opponent was holding? Or what if you had to play a board game without knowing what pieces your opponent is hiding? In situations like these, beyond just calculating well, you must read the opponent’s psychology and even use ‘bluffs.’ Recently, surprising news arrived that AI has finally crossed the human barrier in this difficult realm. ‘DeepNash,’ an AI developed by Google DeepMind, has conquered the board game ‘Stratego.’

Why is this news important?

Think about our daily lives. Many of the decisions we make in reality are made with insufficient information. No one can be 100% sure how the stock market will perform tomorrow or which route to take to avoid traffic today. In this sense, the ability to make the best choices under conditions of imperfect information was a wall that AI had to overcome to move one step closer to the human realm.

Existing AIs overwhelmed humans in environments where all information was transparently disclosed, like chess or Go, but they could not move past an amateur level in environments like Stratego, where information must be hidden and bluffs used (Source: Game has been particularly challenging for AI to master, scientists say, Source: [2206.15378] Mastering the Game of Stratego with Model-Free…). However, the emergence of DeepNash is a major event signifying that AI has begun to handle the ‘uncertainty’ of the complex real world.

In simple terms, what kind of game is it?

Stratego is a ‘game of imperfect information’ where the identity of the opponent’s pieces is unknown (Source: DeepMind’s newest AI thrashes human gamers at Stratego).

To put it in perspective: if chess is a direct confrontation with all cards laid out, Stratego is like conducting espionage while not knowing the opponent’s hand. Until you directly attack the enemy piece and engage in battle, you don’t know if that piece is a bomb or a powerful general. Therefore, you must constantly suspect whether the opponent is trying to deceive you or has set a trap (Source: This AI Finally Beat the Best Humans at One of the Last Board Games…).

Simply put, this AI did not memorize countless rules, but realized through self-play which moves increase its winning percentage. Within the astronomical 10^33 possibilities for initial setups and the vast 10^535 possible game states, DeepNash learned how to read the opponent’s moves and strike at their weaknesses through hundreds of millions of games (Source: This AI Finally Beat the Best Humans at One of the Last Board Games…, Source: Mastering the Game of Stratego with Model-Free).

How far have we come?

DeepNash has already demonstrated overwhelming skill against expert-level human gamers (Source: DeepMind’s Latest AI Trounces Human Players at the Game ‘Stratego’). While past AIs merely overwhelmed opponents with fast calculations, this DeepNash shows a significant difference in that it demonstrated psychological judgment to infer invisible information and strike at the opponent’s blind spots.

However, this is only an achievement within established rules. Although Stratego is very complex, the real world we live in contains far more exceptions and variables than the game. Nevertheless, DeepNash has clearly proven that AI systems have reached a ‘new frontier’ (Source: DeepMind’s newest AI thrashes human gamers at Stratego).

What lies ahead?

DeepNash’s success will be an important foundation for AI to make more flexible decisions in real life in the future. AI’s capabilities are expected to improve significantly in areas requiring decision-making in environments lacking information, such as logistics optimization, complex negotiations between companies, or other environments with more variables. From the perspective of an AI reporter, AI is evolving beyond a simple calculator to a level where it can judge situations ‘intuitively’ and respond like a human. Before long, we may enter an era where AI helps us ponder the complex and uncertain problems of our daily lives.

References

1. Snap! - - Spooky Space, Cute AI, AI Masters Stratego - Spiceworks…

2. [Vue HN 2.0

   With most information hidden, the game Stratego had…](https://vue-hackernews-ssr-5cavbdjcta-ew.a.run.app/item/49933740)

3. This AI Finally Beat the Best Humans at One of the Last Board Games…
4. DeepMind’s newest AI thrashes human gamers at Stratego
5. Game has been particularly challenging for AI to master, scientists say
6. Mastering the Game of Stratego with Model-Free

7. [AI beats us at another game: STRATEGO

   DeepNash… - YouTube](https://www.youtube.com/watch?v=3vO45gcEbRs)

8. [2206.15378] Mastering the Game of Stratego with Model-Free…
9. stratego.io
10. DeepMind’s Latest AI Trounces Human Players at the Game ‘Stratego’
11. With most information hidden, the game Stratego had stumped…