Kriegspiel

What the most exciting chess variant might tell us about the future of AI on the battlefield

Thank you so much for reading The Distro! There is a ChatGPT 5-sentence summary at the bottom.

---

Road Trip

I grew up playing chess. I started playing at such a young age, that I don’t remember learning how to play. I played the most chess when I visited my Pa-Pa (paternal grandfather) in Arlington, Texas every year. Pa-Pa was very strong club player, and even competed in the 1948 US Open Chess tournament. I played against my Pa-Pa hundreds of times, but I only ever won once.

We were on a road trip from Dallas to Birmingham, Alabama. I brought along a tiny little magnetic board—the pieces were the size of thumb tacks. Pa-Pa was in his mid 70s and the road trip was making him extremely tired. After a long hot day of driving we got to our hotel room halfway to Birmingham, and I begged Pa-Pa for one more game. He acquiesced, while almost falling asleep in his chair. I had the black pieces. About 15 moves in I had already scarified a knight and bishop, but I had created an open h-file and his king was trapped on the back rank. My heart was racing. If I could sneak a rook onto the h-file, I might be able to deliver a stealthy check mate. Sure enough, I managed to get a rook to the open h-file and delivered a surprise back rank checkmate.

Neither my father nor my uncle had ever bested my Pa-Pa at chess. But 12-year-old me in the cheapest room in the cheapest motel in Jackson, Mississippi, against my 75 year old Pa-Pa exhausted by an 8-hour car ride in July in a car with no A/C was able to sneak out a surprise win. I was the happiest 12-year-old boy in the world.

My grandfather, Dr. A. Edward Caroe

---

It should be common knowledge now that chess computers¹ are significantly better at chess than the best human players. It’s not even particularly close, AlphaZero (arguably the strongest chess engine) has an ELO rating of ~3500 while Magnus Carlsen, probably the greatest chess player of all time, has an ELO rating of ~2850. To give you an idea of how massive that difference is, if you go to your local chess club, the top player probably has an ELO of around 2000. Imagine your local chess champion going up against the greatest chess player in history. That is what it would be like for Magnus to play against the best engine. And the engines will only improve as time goes on. Here is Magnus playing 10 people and he can’t see the boards:

But how do chess engines work? Why are they so much better at chess than humans? Are they able to “see” so many more moves ahead? Although the algorithms for chess engines are extremely complicated, the oversimplified version is that chess engines are, for the most part, probabilistic. The beginning moves are all preprogramed based on chess theory that has been developed and codified by humans over the last 200+ years. Endgames in chess are also almost completely solved. Once a certain number of pieces are off the board, brute force computation can tell you the EXACT right move with CERTAINTY. The middlegame is the interesting part. In the middlegame, the chess engine will simulate a certain number of games, and then play the move that is most likely to lead to a win (or the one least likely to lose). The best humans can easily match the computer for about the first 12 moves or so (top grandmasters have opening lines memorized out to as many as 20 moves), but after that, the computer’s ability to simply play a million games in its “head” and then play the best probabilistic move is simply unparalleled.

Here is a great video explaining this process:

(Stockfish is very strong, but it typically gets crushed by AlphaZero)

But what does this have to do with combat?

---

Kriegspiel

In war, there is often a lot of uncertainty, especially in the opening stages of a conventional conflict. But in chess, both players know where all the pieces are and the exact limitations of the pieces. There is no uncertainty about the current state of the board. There is some mild uncertainty because you don’t know what your opponent will do, but at least you understand what options he has. Your opponent cannot just parachute a queen randomly onto the board, for example.

But Kriegspiel Chess introduces an element of uncertainty.

Kriegspiel chess, in a nutshell, is chess where you cannot see your opponent’s pieces—each player can only see their own pieces. A referee can see both boards and ensures that only legal moves are made. This video will give you an idea:

There are about 50 different variants to Kriegspiel chess.² I know it seems like you can run into all kinds of problems, but those of us who grew up playing this game have seen most of the edge cases and know how to adjudicate them. Good referees also have a distinct style that keeps the game entertaining. I can tell just by looking that the players in the video above don’t play a lot of Kriegspiel.

My uncle is one of the best Kriegspiel players that I know. In traditional chess, his chances of beating Magnus would be 0%. In Kriegspiel, I’d give him a 20% chance. Magnus would likely still have the upper hand, but there are so many sneaky moves in Kriegspiel, that my uncle actually has a fighting chance.

Kriegspiel is much more like war than conventional chess. In Kriegspiel, you must conduct reconnaissance, you must protect your pieces against where you think your opponent might move, you must be willing to take risks by moving without perfect information, you must make moves designed to confuse your opponent, and you must move your strong pieces off of their starting squares quickly, so they don’t get sniped by fast moving knights.

Great! So Kriegspiel is more like war than conventional chess, whoopdie-doo. What about AI?

---

Information

One of the reasons chess engines are so strong is because they have perfect information. An engine knows where all the pieces are and all the possible legal moves for itself and its opponent. But how would AI fair on a Kriegspiel board where it does not have perfect information? I would submit to you that you could never build a chess engine that is as dominant over humans at Kriegspiel as it is at regular chess. The element of uncertainty changes everything. The computer cannot simulate games if it doesn’t know where the pieces are. If the computer doesn’t know the strength of its opponent, how can it predict moves that make no sense? I think you could build a dominant Kriegspiel chess engine, but not to the level of a 700 point ELO advantage. I would be impressed if you could make a very strong Kriegspiel computer that will purposely make confusing moves that even itself knows is not the strongest move.

Along those same lines, AI will never be able to “solve” warfare. AI doesn’t actually know the future. The best it can do (at this point) is see patterns and make probabilistic guesses.

Information is so important in Kriegspiel (and in war) that if the human player was given a large information advantage, the best AI in the world would stand no chance. If the human player knew where the AI’s pieces were, but the AI did not know where the human’s pieces were, the game would be over very quickly. The AI is only as good as the information that it is dealing with. If it is fed garbage information, it will give a garbage result.

A crucial part of warfare has always been the ability to gain an information advantage over an opponent. This means that reconnaissance, counter-reconnaissance, and deception in all domains will continue to be as important in the future as it has been in the past. Our ability to use good sensors paired with AI to interpret data, and our ability to disrupt our opponent’s sensors will continue to be a critical part of warfare.

---

Cooperation

The best way to play Kriegspiel chess would be a human-AI combination. A human playing Kriegspiel could feed data about the game into the computer, and the computer could help predict where the opponent’s pieces are on the board. For example, if you attempt to make a move and the referee announces that the move is illegal for 3-reasons, the computer may be able to, over time, build an idea of where an opponent’s pieces are. If you discover an enemy piece in an unexpected location, the computer algorithm could make a very good guess about which piece it is, how it got there, and where other pieces are as a result.

But the computer will never be 100% accurate with the level of uncertainty presented by Kriegspiel chess. Human intuition, built over playing 100s of games of Kriegspiel, will always be a force to be reckoned with on a Kriegspiel board and on a battlefield. Sometimes, and it is hard to explain, you can sense where your opponent’s pieces are and make very strong moves that wouldn’t make sense in regular chess. Likewise, military leaders can develop an intuition known as Fingerspitzengefühl.

The conclusion here may not be earth-shattering, but it is important: AI will be useful to humans in war insofar as it makes humans better warfighters. Rather than pursue grand visions of fully autonomous armies, we should focus AI on doing narrow and specific tasks where it can excel over humans to a gigantic degree. Those tasks should then feed critical information to humans or act autonomously within a narrow band of options. AI will be important on future battlefields, but human intuition will always play a crucial role.

---

ChatGPT Summary

1. This essay discusses the superiority of chess engines over human players due to their probabilistic algorithms and computational power.

2. It then introduces Kriegspiel chess, a variant where players lack perfect information, likening it to the uncertainties of warfare.

3. The essay suggests that chess engines would not perform as dominantly in Kriegspiel chess as they do in traditional chess because of the element of uncertainty.

4. Without perfect information about the opponent's pieces and potential moves, the computational advantage of chess engines is significantly reduced.

5. The essay argues that while AI can enhance military capabilities, human intuition remains crucial, advocating for a collaborative approach between AI and human decision-making in war scenarios.

---

Substack is growing like crazy! And I’m doing my part to help cultivate #MilStack - the military corner of Substack.

There are some absolutely brilliant accounts on here that are putting out great stuff! If you find an account you like, be sure to subscribe, restack, and share their material.

Here are some of accounts that are doing great work:

Lethal Minds Journal

Bruce Ivar Gudmundsson

Franklin C. Annis

John Robb

Mark McGrath

Tyler Fox

Shady Maples

Robert A Mosher

B. A. Friedman

BCD

The Connecting File

Ryan McBeth

Mick Ryan

Marine Corps Compass Points

Zachary Griffiths

Gordon Corrigan

Big Serge

Dr Robert Lyman MBE

Grant Smith

Henry Schlottman

Tim Mak

Andrew Cockburn

ALCON Intelligence

Matthew Graham

Blowtorch

Andrew Tanner

1. I use “chess engine” and “chess computer” interchangeably.

2. There is another variant known as “Fog of War” Chess, which is Kriegspiel adjacent, but it is much easier. In “Fog of War” you can actually see some of your opponent’s pieces, depending on where your pieces are. But in true Kriegspiel, you can’t see pieces which are right next to you.