Category Archives: On This Day

Gentleman Johnny

On June 13th, 1777, “Gentleman Johnny” Burgoyne and Major General Guy Carleton inspected the forces of Great Britain assembled at Saint-Jean-sur-Richelieu and about to embark upon and invasion of the American colonies from Canada. The force consisted of approximately 7,000 soldiers and 130 artillery pieces and was to travel southward through New York, by water and through the wilderness, to meet up with a second force moving north from New York City. The act of capturing the Hudson River Valley and, in particular, the city of Albany, would divide New England from the mid-Atlantic colonies, facilitating the defeat of the rebel forces in detail.

Poor communication may have doomed the plan from the start. The army which Burgoyne counted on moving up from New York City, under the command of General Howe, was committed to an attack on Philadelphia, to be executed via a southern approach. Thus, when it needed to be available to move north, Howe’s army would be separated from the upstate New York theater not only by distance, but also by George Washington. Burgoyne did not receive this important information and set out on his expedition unaware of this flaw.

Nonetheless, Burgoyne began well enough. As he moved southward, the colonial forces were unaware of his intent and strength and friendly native forces screened his army from observation. He captured Crown Point without opposition and successfully besieged and occupied Fort Ticonderoga. Following these successes, he embarked on an overland route from Skenesboro (at the southern reaches of Lake Champlain) to Fort Edward, where the Hudson River makes its turn south. This decision seemed to have been taken so as to avoid moving back northward, a retrograde movement necessary to use Lake George’s waterway. It may well also indicate Burgoyne’s lack of appreciation for the upstate New York terrain and its potential to allow a smaller colonial force to impede his movements.

Live Free or Die

In order to deal with the enemy blocking his path, Burgoyne sent forth his allied Indian forces to engage and run off the colonials. Having done so, they proceeded to loot and pillage the scattering of colonial settlements in the area. This had the perverse effect of driving otherwise-neutral locals into the rebel camp. As the fighting portion of his army made the trek to Fort Edward rather rapidly and uneventfully, Burgoyne discovered he had two serious issues. First, he finally received communication from Howe informing him that the bulk of the New York army, the forces with whom Burgoyne was planning to rendezvous, were on their way by sea to south of Philadelphia. Second, the movement through the wilderness had delayed his supply train, unsuited as it was to movement through through primal woodland.

Burgoyne’s solution was to again pause and to attempt to “live off the land” – requisitioning supplies and draft animals from the nearby settlers. Burgoyne also identified a supply depot at Bennington (Vermont) and directed a detachment to seize its bounty. What he didn’t know is that the settlers of Vermont had already successfully appealed to the government of New Hampshire for relief. New Hampshire’s General John Stark had, in less than a week’s time, assembled roughly 10% of New Hampshire’s fighting-age population to field a militia force of approximately 1,500.

When Burgoyne’s detachment arrived at Bennington, they found waiting for them a rebel militia more than twice their number. After some weather-induced delay, Stark’s force executed an envelopment of the British position, capturing many and killing his opposing commander. Meanwhile, reinforcements were arriving from both sides. The royal force arrived first and set upon the disarrayed colonial forces who were busy taking prisoners and gather up supplies. As Stark’s forces neared collapse, the Green Mountain Boys, under the command of Seth Warner, arrived and shored up the lines. The bloody engagement continued until nightfall, after which the royalists fell back to their main force, abandoning all their artillery.

Stark’s dramatic victory had several effects. First, it provided a shot in the arm for American morale, once again showing that the American militia forces were capable of standing up to the regular armies of Europe (Germans, in this case). Second, it had an opposing dilatory impact on the Indian tribe’s morale, causing the large, native force that Burgoyne had used for screening purposes to abandon him. Third, it created a folklore that persists in northern New England to this day. Stark became a hero with his various pre- and post- battle utterances preserved for the ages. Not the least of these was from a letter penned well after independence. Stark regretted that his ill health would prevent him from attending a Battle of Bennington veterans’ gathering. He closed his apology with the phrase, “Live free or die: Death is not the worst of evils,” which has been adopted as the official motto of the State of New Hampshire.

Saratoga

The delays put in the path of Burgoyne’s march gave the Colonials time to organize an opposition. New England’s rebellion found itself in a complex political environment, pitting the shock at the loss of Ticonderoga against the unexpected victory at Bennington. The result was a call-to-arms of the colonial militias which were assembled into a force of some 6,000 in the vicinity of Saratoga, New York. General Horatio Gates was dispatched by the Continental Congress to take charge of this force, which he did as of August 19th. His personality clashed with some of the other colonial generals including, perhaps most significantly, Philip Schuyler. Among the politicians in Philadelphia, Schuyler had taken much of the blame for the loss of Ticonderoga. Some even whispered accusations of treason. Schuyler departure deprived Gates with the knowledge of the area he was preparing to defend, hindering his efforts. Burgoyne focused his will to the south and was determined to capture Albany before winter set in. Going all-in, he abandoned the defense of his supply lines leading back northward and advanced his army towards Albany.

On September 7th, Gates moved northwards to establish his defense. He occupied terrain known as Bemis Heights, which commanded the road southward to Albany, and began fortifying the position. By September 18th, skirmishers from Burgoyne’s advancing army began to meet up against those of the colonists.

Having scouted the rebel lines, Burgoyne’s plan was to turn the rebel left. That left wing was under the command of one of Washington’s stars, General Benedict Arnold. Arnold and Gates were ill-suited for each other leaving Arnold to seek allies from among Schuyler’s old command structure, thus provoking even further conflict. Arnold’s keen eye and aggressive personality saw the weakness of the American left and he realized how Burgoyne might exploit it. He proposed to Gates that they advance from their position on the heights and meet Burgoyne in the thickly-wooded terrain, a move that would give an advantage to the militia. Gates, on the other hand, felt his best option was to fight from the entrenchments that he had prepared. After much debate, Gates authorized Arnold to reconnoiter the forward position where he encountered, and ultimately halted, the British advance in an engagement Freeman’s Farm.

Game Time

In my previous article, I talked about some new stuff I’d stumbled across in the realm of AI for chess. The reason I was stumbling around in the first place was a new game in which I’ve taken a keen interest. That game is Freeman’s Farm, from Worthington Games, and I find myself enamored with it. Unfortunately it is currently out of print (although they do appear to be gearing up for a second edition run).

How do I love thee? Let me count the ways. There are three different angles from which I view this game. In writing here today, I want to briefly address all three. To me, the game is interesting as a historical simulation, as a pastime/hobby, and as an exercise in “game theory.” These factors don’t always work in tandem, but I feel like they’ve all come together here – which is why I find myself so obsessed (particularly with the third aspect).

The downside for this game, piling on with some of the online reviews, is with its documentation. Even after two separate rule clarifications being put out by the developer, there remain ambiguities aplenty. The developer has explained that the manual was the product of the publisher and it seems like Worthington values brevity in their rule sets. In this case, Worthington may have taken it a bit too far. To me, though, this isn’t a deal-breaker. The combination of posted clarifications, online discussion, and the possibility of making house rules leave the game quite playable and one hopes that much improvement will be found in the second edition. Still, it is easy to see how a customer would be frustrated with a rule book that leaves so many questions unanswered.

Historical War Gaming

This product is part of the niche category that is historical wargaming. Games of this ilk invite very different measures of evaluation than other (and sometimes similar) board games. I suppose it goes without saying that a top metric for a historical wargame is how well it reflects the history. Does it accurately simulate the battle or war that is being portrayed? Does it effectively reproduce the command decisions that the generals and presidents may have made during the war in question? Alternatively, or maybe even more importantly, does is provide insight to the player about the event that is being modeled?

On this subject, I am not well placed to grade Freeman’s Farm. What I will say is that the designer created this game as an attempt to address these issues of realism and historicity. In his design notes, he explains how the game came about. He was writing a piece on historical games for which he was focusing on the Saratoga Campaign. As research, he studied “all the published games” addressing the topic, and found them to be lacking something(s).

I’ll not bother to restate the designer’s own words (which can be accessed directly via the publishers website). What is worth noting is that he has used a number of non-conventional mechanisms. The variable number of dice and the re-rolling option are not exactly unique, but they do tend to be associated with “wargame lite” designs or other “non-serious” games. Likewise, the heavy reliance on cards is a feature that does not cry out “simulation.” That said, I am not going to be too quick to judge. Probability curves are probability curves and all the different ways to roll the dice have their own pros and their own cons. Freeman’s Farm‘s method allows players to easily understand which factors are important but makes it very difficult to calculate exactly what are the optimal tactics. Compare and contrast, for example, to the gamey moves required to get into the next higher odds column on a traditional combat results table.

Playing for Fun

All the above aside, a game still needs to be playable and fun to succeed. We seem to be living through a renaissance in the board gaming world, at least among the hobbyists that have the requisite depth of appreciation. A vast number of well-designed and sophisticated board games are produced every year covering a huge expanse of themes. More importantly, ideas about what makes a game “fun” and “playable” have evolved such that today’s games are almost universally better than the games of a generation or two ago. Gone are the days (such as when I was growing up) when slapping a hot franchise onto a roll-the-dice-and-move-your-piece game was considered effective and successful game design. You can still buy such a thing, I suppose, but you can also indulge yourself with dozens and dozens of games based on design concepts invented and refined over the last couple of decades.

From this standpoint, Freeman’s Farm also seems to have hit the mark. It is a nice looking game using wooden blocks and a period-evoking, high quality game board. I’ve read some complaints on line about boxes having missing (or the wrong) pieces in them. This would definitely be a problem if you don’t notice it and try to play with, for example, the wrong mix of cards. The publisher seems to be responsive and quick to get people the material they need.

Game Theory

The real reason I’m writing about this now is because the game has a form that seems, at least to my eyes, to be a very interesting one from a theoretical standpoint. Contrast to, say, chess, and you’ll notice there are very few “spaces” on this game’s board. Furthermore, the movement of pieces between spaces is quite restricted. This all suggests to me that the model of the decision making in this game (e.g. a decision tree) would have a simplicity not typical for what we might call “serious” wargames.

Given the focus of that last post, I think this game would get some interesting results from the kind of modeling that AlphaZero used so successfully for chess. Of course, it is also wildly different from the games upon which that project has focused. The two most obvious deviations are that Freeman’s Farm uses extensive random elements (drawn cards, rolled dice) and that the game is not symmetric and non-alternating. My theory, here, is that the complex elements of the game will still adhere to the behavior of that core, and simple, tree. To make an analogy to industrial control, the game might just behave like a process with extensive noise obscuring its much-simpler function. If true, this is well within the strengths of the AI elements of Alpha Zero – namely the neural-net enhanced tree search.

Momentum and Morale

A key element to all three of these aspects revolves around rethinking on how to model command and control. It is novel enough that I wrap up this piece by considering this mechanism in detail. In place of some tried-and-true method for representing command, this game uses blocks as form of currency – blocks that players accumulate and then spent over the course of the game. Freeman’s Farm calls this momentum; a term helps illustrate its use. From a battlefield modeling and simulation standpoint, though, I’m not sure the term quite captures all that it does. Rather, the blocks are a sort of a catch-all for the elements that contribute to successful management of your army during the battle, looking at it from the perspective of a supreme commander. They are battlefield intelligence, they are focus and intent, and they are other phrases you’ve heard used to describe the art of command. Maybe the process of accumulating blocks represents getting inside your enemy’s OODA and the spending blocks is the exploitation of that advantage.

Most other elements in Freeman’s Farm only drain away as time goes by. For example, your units can lose strength and they can lose morale, but they can’t regain it (a special rule here and there aside). You have a finite number of activations, after which a unit is spent – done for the day, at least in an offensive capacity. Momentum, by contrast, builds as the game goes on – either to be saved up for a final push towards victory or dribbled out here and there to improve your game.

Now, I don’t want to go down a rabbit hole of trying to impose meaning where there probably isn’t any. What does it mean to decide that you don’t like where things are going and your going to sacrifice a bit of morale to achieve a better kill ratio? Although one can construct a narrative as to what that might mean (and maybe by doing so, you’ll enjoy the game more), that doesn’t mean it is all simulation. The point is, from a game standpoint, the designer has created a neat mechanism to engage the players in the process of rolling for combat results. It also allows a player to become increasingly invested in their game, even as it is taking away decisions they can make because their units have become engaged, weakened, and demoralized.

I’m going to want to come back and address this idea of modeling the game as a decision tree. How well can we apply the either decision trees and/or neural networks to the evaluation of game play? Is this, indeed, a simple yet practical application of these techniques? Or does the game’s apparent simplicity obscure a far-more complex reality that prevents the application of these computer learning techniques by being applied by someone who doesn’t have Deep Mind/Google’s computing resources? Maybe I’ll be able to answer some of these questions for myself.

Related

In the fall of 1915, after ten years of analysis, Albert Einstein presented his gravitational field equations of general relativity in a series of lectures at the Royal Prussian Academy of Sciences. The final lecture was delivered on November 25th, 104 years ago.

Yet it wasn’t until a month or so ago that I got a bug up my butt about general relativity. I was focused on some of the paradox-like results of the special theory of relativity and was given to understand, without actually understanding, that the general theory of relativity would solve them. Not to dwell in detail on my own psychological shortcomings, but I was starting to obsess about the matter a bit.

Merciful it was that I came across The Perfect Theory: A Century Of Geniuses And The Battle Over General Relativity when I did. In its prologue, author Pedro G. Ferreira explains how he himself (and others he knows in the field) can get bitten by the Einstein bug and how one can feel compelled to spend the remainder of one’s life investigating and exploring general relativity. His book explains the allure and the promise of ongoing research into the fundamental nature of the universe.

The Perfect Theory tells its story through the personalities who formulated, defended, and/or opposed the various theories, starting with Einstein’s work on general relativity. Einstein’s conception of special relativity came, for the most part, while sitting at his desk during his day job and performing thought experiments. He was dismissive of mathematics, colorfully explaining “[O]nce you start calculating you shit yourself up before you know it” and more eloquently dubbing the math “superfluous erudition.” His special relativity was incomplete in that it excluded the effects of gravity and acceleration. Groundbreaking though his formulation of special relativity was, he felt there had to be more to it. Further thought experiments told him that the gravity and acceleration were related (perhaps even identical) but his intuition failed to close the gap between what he felt had to be true and what worked. The solution came from representing space and time as a non-Euclidean continuum, a very complex mathematical proposition. The equations are a thing of beauty but also are beyond the mathematical capabilities of most of us. They have also been incredibly capable of predicting physical phenomena that even Albert Einstein himself didn’t think were possible.

From Einstein, the book walks us through the ensuing century looking at the greatest minds who worked with the implications of Einstein’s field equations. The Perfect Theory reads much like a techno-thriller as it setts up and then resolves conflicts within the scientific world. The science and math themselves obviously play a role and Ferreira has a gift of explaining concepts at an elementary level without trivializing them.

Stephen Hawking famously was told that every formula he included in A Brief History of Time would cut his sales in half. Hawking compromised by including only Einstein’s most famous formula, E = mc2. Ferreira does Hawking one better, including only the notation, not the full formula, of the Einstein Tensor in an elaboration on Richard Feynman’s story about efforts to find a cab to a Relativity conference as told in Surely You’re Joking, Mr. Feynman. The left side of that equation can be written as, Gμν. This is included, not in an attempt to use the mathematics to explain the theory, but to illustrate Feynman’s punch line. Feynman described fellow relativity-conference goers as people “with heads in the air, not reacting to the environment, and mumbling things like gee-mu-nu gee-mu-nu”. Thus, the world of relativity enthusiasts is succinctly summarized.

The most tantalizing tidbit in The Perfect Theory is offered up in the prologue and then returned to at the end. Ferreira predicts that this century will be the century of general relativity, in the same way the last century was dominated by quantum theory. It is his belief we are on the verge of major new discoveries about the nature of gravity and that some of these discoveries will fundamentally change how we look at and interact with the universe. Some additional enthusiasm shines through in his epilogue where he notes the process of identifying and debunking a measurement of gravitational waves that occurred around the time the book was published.

By the end of the book, his exposition begins to lean toward the personal. Ferreira has an academic interest in modified theories of gravity, a focus that is outside the mainstream. He references, as he has elsewhere in the book, the systematic hostility toward unpopular theories and unpopular researchers. In some cases, this resistance means a non-mainstream researcher will be unable to get published or unable to get funding. In the case of modified gravity, he hints that this niche field potentially threatens the livelihood of physicists who have built their careers on Einstein’s theory of gravity. In fact, it wasn’t so long ago that certain aspects of Einstein’s theory were themselves shunned by academia. As a case in point, the term “Big Bang” was actually coined as a pejorative for an idea that, while mathematically sound, was too absurd to be taken as serious science. Today, we recognize it as a factual and scientific description of the origin of our universe. Ferreira shows us a disturbing facet of the machinery that determines what we, as a society and a culture, understand as fundamental truth. I’m quite sure this bias isn’t restricted to his field. In fact, my guess would be that other, more openly-politicized fields exhibit this trend to an even greater degree.

Ferreira’s optimism is infectious. In my personal opinion, if there is to be an explosion of science it may come from a different direction that which Ferreira implies. One of his anecdotes involves the decision of the United States to defund the Laser Interferometer Space Antenna (LISA), a multi-billion dollar project to use a trio of satellites to measure gravitational waves. To the LISA advocates, we could be buying a “gravitational telescope,” as revolutionary in terms of current technologies as radiotelescopy was to optical telescopes. The ability to see further away and farther back in time would then produce new insights into the origins of the universe. But will the taxpayer spend billions on such a thing? Should he?

Rather than in the abstract, I’d say the key to the impending relativity revolution is found in Ferreira’s own description of the quantum revolution of the past century. It was the engineering applications of quantum theory, primarily to the development of atomic weapons, that brought to it much of the initial focus of interest and funding. By the end of the century, new and practical applications for quantum technology were well within our grasp. My belief is that a true, um, quantum leap forward in general relativity will come from the promise of practical benefit rather than fundamental research.

In one of the last chapters, Ferreira mentions that he has two textbooks on relativity in his office. In part, he is making a point about a changing-of-the-guard in both relativity science and scientists, but I assume he also keeps them because they are informative. I’ve ordered one and perhaps I can return to my philosophical meanderings once I’m capable of doing some simple math. Before I found The Perfect Theory, I had been searching online for a layman’s tutorial on relativity. Among my various meanderings, I stumbled across a simple assertion; one that seems plausible although I don’t know if it really has any merit. The statement was something to the effect that there is no “gravitational force.” An object whose velocity vector is bent (accelerated) by gravitational effects is, in fact, simply traveling a straight line within the curvature of timespace. If I could smarten myself up to the point where I could determine the legitimacy of such a statement, I think I could call that an accomplishment.

The Pride and Disgrace

Handful of Senators don’t pass legislation.

On August 10th, 1964, the Gulf of Tonkin Resolution was enacted. It was the quick response from America’s politicians to the attack upon a U.S. naval vessel off the coast of Vietnam by torpedo boats from the communist regime of North Vietnam. That attack had occurred on August 2nd with an additional “incident” taking place on August 4th. The resolution (in part) authorized the President “to take all necessary steps, including the use of armed force, to assist any member or protocol state of the Southeast Asia Collective Defense Treaty requesting assistance in defense of its freedom.” This would come to justify the deployment of U.S. troops to directly engage the enemies of South Vietnam.

Before this time, South Vietnam was fighting to eliminate a communist insurgency driven by remnants of the Việt Minh. The communist guerillas had agreed to resettle in the North as part of the Geneva Agreement which ended France’s war in Vietnam in 1954. The United States saw their continued attacks on the government of South Vietnam as a violation of that peace. In particular, their support obtained from across national borders was considered to be a part of a strategic plan by the Soviet Union and China to spread communism throughout all of the countries of Southeast Asia.

Even before the withdrawal of France, the United States had supported the anti-communist fight with money, matériel, and military personnel (in the form of advisors). After the French exit, a steady increase in commitment from the U.S was evident. Nevertheless, the signing of the Gulf of Tonkin resolution marks a milestone in American’s involvement making it arguable the start of the U.S. War in Vietnam. Although it would take almost another year for U.S. forces to become clearly engaged, Presidents Johnson’s reaction to the Gulf of Tonkin incidents seems to have set the course inevitably toward that end.

I’m sittin’ here.
Just contemplatin’.
I can’t twist the truth.
It knows no regulation.

The anniversary and the nature of it got me to thinking about how one might portray the Vietnam War as a game. Given the context, I’m thinking purely along the lines of the game focused at the strategic level, taking into account the political and international considerations that drove the course of the conflict.

From a high-level perspective, one might divide America’s war in Vietnam into 4 distinct phases. In the first, the U.S. supported Vietnam’s government with financial and military aid, and with its advisors. While U.S. soldiers were, in fact, engaging in combat and being killed, it wasn’t as part of American combat units, allowing the U.S. to convince itself that this was a conflict purely internal to South Vietnam. Through the presidencies of Eisenhower, Kennedy, and Johnson, the amount of aid to, and the number of Americans in, Vietnam increased. However, the big change, and the transition of the second phase, can be located after the passage of the Gulf of Tonkin Resolution and the subsequent deployment of the U.S. Marines directly and as a unit.

At first, U.S. direct involvement carried with it a measure of popular support and was, from a purely military standpoint, overwhelmingly successful. Johnson and the military were wary of pushing that involvement in ways that would turn the public opinion against them. The U.S. feared casualties as well signs of escalation that be interpreted as increasing military commitment (for example, extending the service time for draftees beyond twelve months), but in general this was a period of an increasing U.S. buildup and, generally, successful operations. Nevertheless, progress in the war defied a clear path toward resolution.

The third phase is probably delineated by the 1968 Tet offensive. While still, ultimately, a military success from the U.S. standpoint, the imagery of Viet Cong forces encroaching on what were assumed by all to be U.S.-controlled cities turned opinion inexorably against continuing engagement in Vietnam. The next phase, then, was what Nixon called “Vietnamization,” the draw-down of American direct involvement to be replaced with support for the Army of the Republic of Vietnam (ARVN). Support was again in the form of money, equipment, and training as well as combat support. For example, a transition to operations where ARVN ground units would be backed by U.S. air power.

The final phase is where that withdrawl is complete, or at least getting close to that point. Where joint operations were no longer in the cards. Clearly this phase would describe the post-Paris accords situation, after Nixon’s resignation, as well as encompassing the final North Vietnamese operation that rolled up South Vietnam and Saigon.

From a gaming perspective, and a strategic-level gaming perspective at that, the question becomes what decisions are there for a player to make within these phases and, perhaps more importantly, what decisions would prompt the transition from one phase to another.

The decision to initially deploy U.S. troops, made by Johnson in early 1965, seems to have been largely driven by events. Having Johnson as president was probably a strong precondition. Although he ran against Goldwater on a “peace” platform, the fact that he saw his legacy as being tied into domestic policy probably set up the preconditions for escalation. A focus on Vietnam was never to be part of his legacy, but given the various triggers in late 1964 and early 1965, his desire to avoid a loss to communism in Vietnam propelled his decision to commit ground troops. You might say his desire to keep it from being a big deal resulted in it being a big deal.

Where this all seems to point is that any strategic Vietnam game beginning much before Spring of 1965 must restrict the player from making the most interesting decision; if and when to commit U.S. ground troops and launch into an “American” war.

Amusingly, if you subscribe to the right set of conspiracy theories, the pivotal events might really be under control of a grand-strategic player after all. Could it be that the real driver behind Kennedy’s assassination was to put a President in office who would be willing to escalate in Vietnam? Was the deployment of the USS Maddox on the DESOTO electronic warfare mission meant to provoke a North Vietnamese response? How about the siting of aviation units in Vietnam at places like Camp Holloway, which would become targets for the Viet Cong? Where actual aggression by the North wasn’t sensational enough, were details fabricated? This rather far-flung theorizing would not only make the resulting game that much harder to swallow, but it is also difficult to see how any fully-engineered attempt to insert American into Vietnam could have moved up the timetable.

So it would only make sense to start our game with our second phase, which must come after our Gulf of Tonkin incident and the 1968 presidential election, at a minimum.

The remaining game will still be an unconventional one, although we do have some nice examples of how it could be done from over the years. Essentially, the U.S. will always be able to draw upon more military power and, ultimately, sufficient military power to prevail in an particular engagement. Yet while it is possible, though insufficient planning, to achieve a military loss as the U.S. it is probably not going to be possible to achieve a military victory. On the U.S. side, the key parameters are going to be some combination of resources and “war weariness.”

Our Vietnam game would rule out, either explicitly or implicitly, a maximal commitment to victory by the United States. American planners considered options such as unfettered access to Laos and Cambodian, an outright invasion of North Korea, or even tactical nuclear weapons. The combination of deteriorating domestic support and the specter of overt Chinese and Soviet intervention would seem to be a large enough deterrent to prevent exercise of these options. This is one of the reasons that rules (those that I’ve come across, any way) simply forbid, for example, crossing units into North Vietnam.

The other reason is one of scope. If a ground invasion of North Vietnam is on the table, then the map needs to include all the potential battlefields in the North in addition to the actual battlefields of the South. Likewise extended areas within Cambodia and Laos need to be available to the player. Continuing on, if U.S. ground forces are going to be straying that close to North Vietnam’s northern border, might it not be necessary to include China in as well? Perhaps having learned a lesson from Korea, our player would react to Chinese direct intervention by taking the fight onto Chinese sovereign territory. It doesn’t take long before we have to consider adding Germany, Korea, Cuba, and any other hotspot of the time as a potential spillover for escalation in Vietnam. Besides the problem of the game expanding without limitations, we have another design concern. A Vietnam game narrative adhering closely to the historical path has the advantage of actual battles, strategies, and events on which to model itself. If all the forces of NATO, the Warsaw Pact, and China are fair game, we are now in the realm of pure speculation.

If for no other reason than to maintain the sanity of the designer, it seems that rules which quickly push the U.S. into its historical de-escalation policy is the right way to tie off such a game on the other end. I will save the consideration of how that might work to another time.

ABC Easy as 42

Teacher’s gonna show you how to get an ‘A’

In 1989, IBM hired a team of programmers out of Carnegie Mellon University. As part of his graduate program, team leader Feng-hsiung Hsu (aka Crazy Bird) developed a system for computerized chess playing that the team called Deep Thought. Deep Thought, the (albeit fictional) original, was the computer created in Douglas Adam’s The Hitchhiker’s Guide to the Galaxy to compute the answer for Life, the Universe, and Everything. It was successful in determining the answer was “42,” although it remained unknown what the question was. CMU’s Deep Thought, less ambitiously, was a custom designed hardware-and-software solution for solving the problem of optimal chess playing.

Once at IBM, the project was renamed Deep Blue, with the “Blue” being a reference to IBM’s nickname of “Big Blue.”

On February 10th, 1996, Deep Blue won its first game against a chess World Champion, defeating Garry Kasparov. Kasparov would go on to win the match, but the inevitability of AI superiority was established.

Today, computer programs being able to defeat humans is no longer in question. While the game of chess may never be solved (à la checkers), it is understood that the best computer programs are superior players to the best human beings. Within the chess world, computer programs only make news for things like when top players may using programs to gain an unfair advantage in tournament play.

Nevertheless, a chess-playing computer was in the news late last year. Headlines reported that a chess playing algorithm based on neural networks, starting only from the rules of legal chess moves, in four hours created a program that could beat any human and nearly all top-ranked chess programs. The articles spread across the internet through various media outlets, each summary featuring their own set of distortions and simplifications. In particular, writers that had been pushing articles about the impending loss of jobs to AI and robots jumped on this as proof that the end had come. Fortunately, most linked to the original paper rather than trying to decipher the details.

Like most I found this to be pretty intriguing news. Unfortunately, I also happen to know a little (just a little, really) about neural networks, and didn’t even bother to read the whole paper before I started trying to figure out what had happened.

Some more background on this project. It was created at DeepMind, a subsidiary of Alphabet, Inc. This entity, formerly known simply as Google, reformed itself in the summer of 2015 with the new Google being one of many children of the Alphabet parent. Initial information suggested to me an attempt at creating one held company for each letter of the alphabet, but time has shown that isn’t their direction. As of today, while there are many letters still open, several have multiple entries. Oh well, it sounded more fun my way. While naming a company “Alphabet” seems a bit uninspired, there is a certain logic to removing the name Google from the parent entity. No longer does one have to wonder why an internet company is developing self-driving cars.

Google’s self driving car?

 

The last time the world had an Artificial Intelligence craze was in the 1980s into the early 1990s. Neural networks were one of the popular machine intelligence techniques of that time too. At first they seemed to offer the promise of a true intelligence; simply mimicking the structure of a biological brain could produce an ability to generalize intelligence, without people to craft that intelligence in code. It was a program that could essentially teach itself. The applications for such systems seemed boundless.

Unfortunately, the optimism was quickly quashed. Neural networks had a number of flaws. First, they required huge amounts of “training” data. Neural Nets work by finding relationships within data, but that source data has to be voluminous and it has to be suited to teaching the neural network. The inputs had to be properly chosen, so as to work well with the networks’ manipulation of that data and the data themselves had be properly representative of the space being modeled. Furthermore, significant preprocessing was required from the person organizing the training. Additional inputs would result in exponential increases in both the training data requirement and the amount of processing time to run through the training.

It is worthwhile to recall the computer power available to neural net programmers of that time. Even a high-end server of 35 years ago is probably put to shame by the Xbox plugged into your television. Furthermore, the Xbox is better suited to the problem. The mathematics capability of Graphical Processing Units (GPUs) is a more efficient design for solving these kinds of matrix problems. Just like Bitcoin mining, it is the GPU on a computer that is going to best be able to handle neural network training.

To illustrate, let me consider briefly a “typical” neural network application of the previous generation. One use is something called a “soft sensor.” Another innovation of that same time was the rapid expansion in capabilities of higher-level control systems for industrial processes. For example, some kind of factory wide system could collect real-time data (temperatures, pressures, motor speeds – whatever is important) and present them in an organized fashion to give an overview of plant performance and, in some cases, automate overall plant control. For many systems however, the full picture wasn’t always available in real time.

Let’s imagine the production of a product which has a specification limiting the amount of some impurity. Largely, we know what the right operating parameters of the system are (temperatures, pressures, etc) but to actually measure for impurities, we manually draw a sample, send it off to a lab for testing, and wait a day or two for the result. It would stand to reason that, in order to keep your product within spec, you must operate far enough away from the threshold that if it begins to drift, you would usually have time catch it before it goes out of spec. Not only does that mean you’re, most of the time, producing a product that exceeds specification (presumably at extra cost), but if the process ever moves faster than expected, you may have to trash a day’s worth of production created while you were waiting for lab results.

Enter the neural network and that soft sensor. We can create a database of the data that were collected in real time and correlate that data with the matching sample analyses that were available afterward. Then a neural network can be trained using the real-time measurements as input to produce an output predicting sample measurement. Assuming that the lab measurement is deducible from the on-line data, you now have in your automated control system (or even just as a presentation to the operators) a real time “measurement” of data that otherwise won’t be available until much later. Armed with that extra knowledge, you would expect to both cut operating costs (by operating tighter to specification) and prevent waste (by avoiding out-of-spec conditions before they happen).

That sounds very impressive, but I did use the word “assuming.” There were a lot factors that had to come together before determining that a particular problem was solvable with neural networks. Obviously, the result you are trying to predict has to, indeed, be predictable from the data that you have. What this meant in practice is that implementing neural networks was much bigger than just the software project. It often meant redesigning your system to, for example, collect data on aspects of your operation that were never necessary for control, but are necessary for the predictive functioning of the neural net. You also need lots and lots of data. Operations that collected data slowly or inconsistently might not be capable of providing a data set suitable for training. Another gotcha was that collecting data from a system in operation probably meant that said system was already being controlled. Therefore, a neural net could just as easily be learning how your control system works, rather than the underlying fundamentals of your process. In fact, if your control reactions were consistent, that might be a much easier thing for the neural net to learn that the more subtle and variable physical process.

The result was that many applications weren’t suitable for neural networks and others required a lot of prep-work. Projects might begin with redesigning the data collection system to get more and better data. Good data sets in hand, one now was forced into time-intensive data analysis which was necessary to ensure a good training set. For example, it was often useful to pre-analyze the inputs to eliminate any dependent variables. Now, technically, that’s part of what the neural network should be good at – extracting the core dependencies from a complex system. However, the amount of effort – in data collected and training time – increases exponentially when you add inputs and hidden nodes, so simplifying a problem was well worth the effort. While it might seem like you can always just collect more data, remember that the data needed to be representative of the domain space.  For example, if the condition that results in your process wandering off-spec only occurs once every three or four months, then doubling your complexity might mean (depending on your luck) increasing the data collection from a month or two to over a year.

Hopefully you’ll excuse my trip down a neural net memory lane, but I wanted to set your expectations of neural network technology where mine were, because the state of the art is very different than what it was. We’ve probably all seen some of the results with image recognition that seems to be one of the hottest topics in neural networks these days.

So back to when I read the article. My first thought was to think in terms of the neural network technology as I was familiar with it.

My starting point to design my own chess neural net has to be representations of the board layout. If you know chess, you probably have a pretty good idea how to describe a chess board. You can describe each piece using a pretty concise terminology . In this case, I figure it is irrelevant where a piece has been. So whether it started as a king’s knight’s pawn or a queen’s rook’s pawn, that doesn’t effect its performance. So you have 6 possible piece descriptors which need to be placed into the 64 squares that they could possibly reside upon. So, for example, imagine that I’m going to assign an integer to the pieces, and then use positive for white and negative for black:

Pawn Knight Bishop Rook King Queen
1 2 3 4 5 6

My board might look something like this 4,2,3,6,5,3,2,4,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0…-3,-2,-4

If I am still living in the 90s, I’m immediately going to be worried about the amount of data, and might wonder if I can compress the representation of my board based on assumptions about the starting positions. I’ve got all those zeros in the center of my matrix, and as the game progresses, I’m going to be getting fewer data and more zeros. Sixty-four inputs seems like a lot (double that to get current position and post-move position), and I might hope to winnow that down some manageable figure with the kind of efforts that I talked about above.

If I hadn’t realized my problem already, I’d start to figure it out now. Neural networks like inputs to be proportional. Obviously, binary inputs are good – something either affects the prediction or doesn’t. But for variable inputs, the variation must make sense in terms of the problem you are solving. Using the power output of a pump as an input to a neural network makes sense. Using the model number of that pump, as an integer, wouldn’t make sense unless there is a happenstancial relationship between the serial number and some function meaningful to your process. Going back to my board description above, I could theoretically describe the “power” of my piece with a number between 1 and 10 (as an example), but any errors in my ability to accurately rank my pieces contribute to prediction errors. So is a Queen worth six times a pawn or nine? Get that wrong, and my neural net training has an inaccuracy built in right up front. And, by the way, that means “worth” to the neural net, not to me or other human players.

A much better way to represent a chess game to a mathematical “intelligence” is to describe the pieces. So, for example, each piece could be described with two inputs, describing its deviation from that piece’s starting position in the X and Y axes, with perhaps a third node to indicate whether the piece is on the board or captured. My starting board then becomes, by definition, 96 zeros, with numbers being populated (and generally growing) as the pieces move. It’s not terribly bigger (although rather horrifyingly so to my 90s self) than the representation by board, and I could easily get them on par by saying, for example, that pieces captured are moved elsewhere on the board, but well out of the 8X8 grid. Organizing by the pieces, though, is both non-intuitive for we human chess players and, in general, would seem less efficient in generalizing to other games. For example, if I’m modelling a card game (as I talked about in my previous post), describing every card, and each of their possible positions; that is a much bigger data set than just describing what is in each hand and on the table. But, again, it should be clear that the description of the board is going to be considerably less meaningful as a mathematical entity than the description created by working from each game piece.

At this point, it is worth remembering again that this is no longer 1992. I briefly mentioned the advances in computing, both in power and in structure (the GPU architecture as superior for solving matrix math). That, in turn, has advanced the state of the art in neural network design and training. The combination goes a long way in explaining why image recognition is once again eyed as a problem for neural networks to address.

Consider the typical image. It is a huge number of pixels of (usually) highly-compressible data. But compressing the data will, as described above, befuddle the neural network. On the other hand, those huge, sparse matrices need representative training data to evenly cover the huge number of inputs, with that need increasing geometrically. It can quickly become, simply, too much of a problem to solve in a timely manner no matter what kind of computing power you’ve got to throw at it. But with that power, you can do new and interesting things. A solution for image recognition is to use “convolutional” networks.

Not to try to be too technically correct, I’ll try to capture the essence of this technique. The idea is that the input space can be broken up into sub-spaces (in an image, small fractions of the image), that then feed a significantly smaller neural network. Then, one might assume that those small networks are all the same or similar to each other. For an image recognition, we might train 100s or even 1000s of networks operating on 1% of the image (in overlapping segments), creating a (relatively) small output based on the large number of pixels. Then those outputs feed a whole-image network. It is still a massive computational problem, but immensely smaller than the problem of training a network processing the entire image as the input.

Does that make a chess problem solvable? It should help, especially if you have multiple convolutional layers. So there might be a neural network that describes each piece (6 inputs for old/new position (2D) plus on/off board) and reduces it to maybe 3 outputs. A second could map similar pieces.. where are the bishops? where are the pawns? Another sub-network, repeated twice, could try just looking at one player at a time. It is still a huge problem, but I can see that it’s something that is becoming solvable given some time an effort.

Of course, this is Alphabet, Inc we are talking about. They’ve got endless supplies of (computing) time and (employee) effort, so if it is starting to look doable to a mere human like me, it is certainly doable for them.

At this point, I went back to the research paper wherein I discovered that some of my intuition was right, although I didn’t fully appreciate that last point. Just as a simple example, the input layer for the DeepMind system is to represent each piece as a board showing the position of the piece. So 32X a 64-by-64 positional grid. They also use a history of turns, not just current and next turn. It is orders-of-magnitude more data than I anticipated, but in extremely sparse data sets. In fact, it looks very much like image processing, but with much more ordered images (to a computer mind, at least). The paper states they are using Tensor Processing Units, a Google concoction meant to use hardware having similar advantages to the GPU and it’s matrix-math specialization, but further optimized specifically to solve this kind of neural network training problem.

So lets finally go back to the claim that got all those singularity-is-nigh dreams dancing in the heads of internet commentators. The DeepMind team were able to train in a matter of (really) twenty-four hours a superhuman level chess player with no a priori chess knowledge. Further, the paper states that the training set consists of 800 randomly-generate games (constrained only to be made up of legal moves), which seems like an incredibly small data set. Even realizing how big those representations are (with their sparse descriptions of the piece locations as well as per-piece historical information), it all sounds awfully impressive. Of course, that is 800 games per iteration. If I’m reading right, that might be 700k iterations in over 9 hours using hardware nearly inconceivable to we mortals.

And that’s just the end result of a research project that took how long? To get to that point where they could hit the “run” button took certainly months, and probably years.

First you’ve got to come up with the data format, and the ability to generate games in that format. Surprisingly, the paper says that the exact representation wasn’t a significant factor. I suppose that it an advantage of its sparseness. Next, you’ve got to architect that neural net. How many convolutions over what subsets? How many layers? How many nodes? That’s a huge research project, and one that is going to need huge amounts of data – not the 800 randomly generated games you used at the end of it all.

The end result of all this – after a process involving a huge number of PhD hours and petaFLOPS of computational power – you’ve created a brain that can do one thing; learn about chess games. Yes, it is a brain without any knowledge in it – a tabula rasa – but it is a brain that is absolutely useless if provided knowledge about anything other than playing chess.

It’s still a fabulous achievement, no doubt. It is also research that is going to be useful to any number of AI learning projects going forward. But what it isn’t is any kind of demonstration that computers can out-perform people (or even mice, for that matter) in generic learning applications. It isn’t a demonstration that neural nets are being advanced into the area of general learning. This is not an Artificial Intelligence that that could be, essentially, self-teaching and therefore life-like in terms of its capabilities.

And, just to let the press know, it isn’t the end of the world.

The Nature of my Game

I watched with glee while your kings and queens
fought for ten decades for the gods they made.

On October 19th, 1453 the French army entered Bordeaux. The province of Gascony, with Bordeaux as its capital, had been a part of England since the marriage, in 1152, of Eleanor of Aquitaine to the soon-to-be King Henry II of England. Between 1429 and 1450, the Hundred Years War had seen a reversal of English fortunes and a series of French victories which began under the leadership of Jeanne D’Arc and culminated in the French conquest and subsequent control of Normandy.

Following the French victory in Normandy, a 3 year struggle for the control of Gascony began. As the fight went on, it saw dominance shift from the English to the French – and then back again after the arrival of John Talbot, earl of Shrewsbury. Eventually Talbot succumbed to numbers and politics, leading his army against a superior French position at Castillon on July 17th of 1453. He and his son both paid with their lives and, with the defeat of the English army and the death of its leadership, the fall of all of Gascony and Bordeaux inevitably followed before the year’s end.

The Battle of Castillon is generally cited as the end of the Hundred Years War, although the milestone is considerably more obvious in retrospect than it was at the time. The English defeat did not result in a great treaty or the submission of one king to another. England simply no longer had control of her French territories, save for Calais, and in fact never would again. Meanwhile, the world was moving on to other conflicts. The end of the Hundred Years War, in turn, is often cited as a key marker for the end of the (Late) Medieval period and the transition to the (Early) Modern period. Another major event of 1453, the Fall of Constantinople, is also a touchstone for delineating the transition to the modern world. Whereas the Hundred Years War marked a shift from the fragmented control by feudal fiefdoms to ever-more centralized nation states, the Fall of Constantinople buried the remains of the Roman Empire. In doing so, it saw the flight of Byzantine scholars from the now-Ottoman Empire to the west, and particularly to Italy. This produced a symbolic shift of the presumptive heir of the Roman/Greek foundations of Western Civilization to be centered, once again, on Rome.

The term “Renaissance” refers to the revival of classical Greek and Roman thought – particularly in art, but also in scholarship and civics. Renaissance scholarship held as a goal an educated citizenry with the skills of oration and writing sufficient to positively engage in public life. Concurrent to the strides in the areas of art and architecture, which exemplify the period, were revolutions in politics, science, and the economy. The combination of the creation of a middle class, through the availability clerical work, and the emphasis on the value of the individual, helped drive the nail into the coffin of Feudalism.

The designer for the boardgame Pax Renaissance references 1460 as the “start date” for the game, which lasts through that transitional period (roughly 70 years). Inherent in the design of the game, and expounded upon in the manual, is the idea that what drove the advances in art, science, technology, and government was transition to a market economy. That transition shifted power away from the anointed nobility and transferred it to the “middle class.” The game’s players immerse themselves in the huge changes that took place during this time. They take sides in the three-way war of religion, with Islam, Catholicism and the forces of the reformation fighting for men’s souls. The game simulates the transition of Europe from feudalism to modern government; either the nation state empires or the republic. Players also can shift the major trade routes, refocusing wealth and power from the Mediterranean to northwestern Europe.

Technically speaking, I suppose Pax Renaissance is not a boardgame, because there is no board. It is a card game, although it stretches that definition as well. Some of the game’s cards are placed on the table to form a mapboard of Europe and the Mediterranean, while others serve a more straight-forward “card” function – set down from the players’ hand onto the table in front of them. The game also has tokens that are deployed to the cards and then moved from card to card (or perhaps to the space between cards.). While this starts to resemble the more typical board game, if you continue to see the game in terms of the cards, the tokens can be interpreted to indicate additional states beyond the usual up-or-down (and occasionally rotated-sideways) that cards convey.

Thinking about it this way, one might imagine that it drew some of its inspiration a card game like Rummy – at least the way I learned to play Rummy. In that variant, players may draw from the discard pile, with deeper selections into the pile having an increased cost (of holding potentially negative-point cards in your hand). Once collected, cards remain in the hand or are played in front of the player. Of course, this doesn’t really map one-to-one. A unique point in Pax Renaissance is that there are no secret (and necessarily random) draws directly to the players’ hands. Instead, new cards are dealt into the “market,” the openly-visible and available pile, usually to a position that is too expensive to be accessed immediately, giving all players visibility several turns in advance.

Thus the game has no random component, assuming that one allows that the differing deck order (and content – not every card is used in every game) could be thought of as different “boards” as opposed to a random factor. So rather than a checkers or a chess with its square grid, it is a version where there are many 1000s of variations in the board shape and initial setup. Stretching the analogy to its breaking point, that variable board may have also a “fog of war,” as the playing space is slowly revealed over the course of the game.

I don’t actually mean to equate Pax Renaissance with Rummy or chess, but rather to establish some analogies that would be useful when trying to develop a programmed opponent. The game is the third in a “Pax” series from the designer, and can easily be seen as a refinement to that system. Theme-wise, it is a follow-on to his game Lords of the Renaissance from 20 years earlier. That title is a far more traditional map-and-counter game on the same subject, for 12 (!!!) players.

However, I’d like to look at this from an AI standpoint, and so I’ll use the comparison to checkers.

Since the “board” is revealed to all players equally (albeit incrementally) there is no hidden knowledge among players. Aside from strategy, what one player knows they all know. Given that factor, one supposes that victory must go to the player who can think more moves ahead than their opponents can.

I recently read an article about the development of a checkers artificial intelligence. The programmer in this tale took on checkers after his desire to build a chess intelligence was made obsolete by the Deep Blue development efforts in the early 1990s. It was suggested to him that he move to checkers and he quickly developed a top-level player in the form of a computer algorithm. His solution was to attack the problem from both sides. He programed the end-game, storing every possible combination of the remaining pieces and the path to victory from there. He also programmed a more traditional look-ahead algorithm, starting from a full (or nearly so) board and analyzing all the permutations forward to pick the best next move. Ultimately, his two algorithms met in the middle, creating a system that could fully comprehend every possible move in the game of checkers.

Checkers, as a target game for the development of AI, had two great advantages. First, it is a relatively simple game. While competitive, world-class play obviously has great depth, most consider a game of checkers to be fairly easy and casual. The board is small (half the spaces, functionally speaking, as chess) and the rules are very simple. There are typically only a handful of valid moves given any checkers board setup, versus dozens of valid moves in chess. Secondly, the number of players is large (who doesn’t know how to play), and thus the knowledge about what strategies to use is known, even if not quite as well-analyzed as with Chess. Thus, in a checkers game an AI can begin its work by using a “book.” That is, it uses a database of all of the common and winning strategies and their corresponding counter-strategies. If a game begins by following the path of an already-known game, the programmed AI can proceed down that set of moves.

At least until one player decides its fruitful to deviate from that path.

After that, in the middle part of the game, a brute force search can come into play. Note that this applies to a programmed opponent only until the game is “solved”, as described in the article. Once the database has every winning solution from start to end, a search over the combinations of potential moves isn’t necessary. But when it is used, the AI searches all combinations of moves from the current position, selecting its best current turn move based on what the (human) opponent is likely to do. At its most basic, this problem is often considered with a minimax algorithm. This is an algorithm that makes an assumption that, whatever move you (thinking of yourself as the algorithm) make, your opponent will counter with the move least advantageous to you. Therefore, to find the best move for yourself, you alternately search for the best move you can make and then the worst move your opponent can make (the minimum and then maximum ranked choices) to determine the end state for any current turn move.

Wikipedia has a description, with animated example, of how such a search works using a technique to avoid searching fruitless branches of the tree. That inspired me to take a look at Pax Renaissance and do a similar evaluation of the choices one has to make in that game.

 

A smallish example of an animated game tree.

I’m following the color coding of the Wikipedia example, although in the above screenshot it’s not as clear as it should be. First of all, not everything is working correctly. Second, I took a screen shot while it is actively animating. The coloring of the nodes is done along with the calculations and, as the tree is expanded and/or pruned, the existing branches are shifted around to try to make things legible. It looked pretty cool when it was animating. Not quite so cool to watch once I upped the number of nodes by a factor of ten or so from what is displayed in the above diagram.

I’m assuming a two-player game. The actual Pax Renaissance is for 2-4 players, but initially I wanted to try to be as much like the “textbook” example as I could. The coloring is red for a pruned/unused branch and yellow for an active or best branch. The cyan block is the one actively being calculated, and the blue means a block that his been “visited,” but it has not yet completed evaluation. The numbers in each block are the best/worst heuristic at the leaf of each branch, which is four plies down (two computer turns and two opponent turns). Since at each layer the active player is assumed to choose the best move for them, the value in a circle should be the lowest value of any square children and the square’s should the highest value of any circular children.

The value is computed by a heuristic, potentially presenting its own set of problems. On one hand, the heuristic is a comparison between the two players. So if the computer has more money, then the heuristic comes out positive. If the opponent has more money, then the heuristic comes out negative, with that value being the difference between the two players’ bank accounts. In that sense, it is much easier than, say, positional elements on the chess board, because each evaluation is symmetrical. The hard part is comparing the apples to the oranges. A determination is needed much like the “points” assigned to pieces in chess. Beginning chess players learn that a rook with worth 5 pawns. But how much is a “Coronation Card” worth in florins? Perfecting a search algorithm means both getting the algorithm working and implementing that “domain knowledge,” the smarts about the balance among components, within the mathematical formulas of the search.

As I said, this was an early and simple example.  To build this tree, I assumed that both players are going start the game being frugal in their spending, and therefore use their first turn to buy the cheapest two cards. A the turns advance, they look at the combinations of playing those cards and buying more. Even in this simple example, I get something like 4000 possible solutions. In a later attempt (as I said, it starts looking pretty cluttered), I added some more game options and produced a tree of 30,000 different results. Remember, this is still only two turns and, even within those two turns, it is still a subset of moves. Similar to chess and checkers, as the initial moves are complete, the number of possibilities grows as the board develops.

At this point, I need to continue building more complete trees and see how well and efficiently they can be used to determine competitive play for this game. I’ll let you know if I find anything.

Cold War Chess

On May 16th, 1956, the newly constituted Republic of Egypt under the rule of Gamal Abdel Nasser recognized the communist People’s Republic of China.

Egypt had broken from British rule in 1952 with the Free Officers Movement and their coup which ended the Egyptian monarchy. The influence of the military, and particularly Nasser, shifted to more involvement in the political. Nasser and the other officers ruled through a Revolutionary Command Council and, over the next few years, eliminated political opposition. Nasser became chairman of he Revolutionary Command Council and by 1954 was largely himself ruled Egypt.

In the run up to the 1952 coup, Nasser had cultivated contacts with the CIA. His purpose was to provide a counter balance to the British, should they attempt to oppose the Free Officers in their takeover. The U.S. came to see Nasser as an improvement over the deposed King Farouk and looked to his support in the fight against communism. Nasser himself promoted pan-Arab nationalism which concerned itself largely with the perceived threat from the newly-formed State of Israel. Nasser also became a leader of the newly-independent third world countries, helping create the policy of “neutralism,” having the rising powers of the third world remain unaligned in the Cold War.

It was within this context that the recognition of China appeared to be so provocative.

Egypt had begun drifting towards the communist camp due to a frustration with terms of arms sales and military support from the Western powers. A major weapons deal with the USSR to purchase Czechoslovakian weapons in 1955 greatly enhanced Egypt’s profile in the region, and put them on an even military setting with Israel.

When Nasser recognized China, the response from the U.S. was a counter punch; withdrawing financial support for the Aswan Dam project, itself conceived as a mechanism for securing Egypt’s support on the anti-communist side of the Cold War. U.S. officials considered it a win-win. Either they would bend Nasser to their will, and achieve better compliance in the future, or he would be forced to go to the Soviets to complete the Aswan Dam. They figured that such a project was beyond the financial capabilities of the Russians, and the strain would hamper the Soviet economic and military capabilities enough to more than make up for the deteriorated relations with Egypt. In that event, the ultimate failure of the project would likely realign Egypt with the U.S. anyway.

Egypt’s response continued to surprise. Despite having negotiated that the UK turn over control of the Suez Canal to Egypt, on July 26th, 1956, Nasser announced the nationalization of the Suez Canal and used the military to expel the British and seize control over its operation.

Ain’t she a beautiful sight?

There was armored cars, and tanks, and jeeps,
and rigs of every size.

Twenty-eight years after the Jerusalem riots saw the beginning of Operation Nachshon. The Operation was named for the Biblical prince Nachshon, who himself received the name (meaning daring, but it also sounds similar to the word for “stormy sea waves”) during the Israelites exodus from Egypt. According to one text, when the Israelites first reached the Red Sea, the waters did not part before them. As the people argued on the sea’s banks about whom would lead them forward, Nahshon entered the waters. Once he was up to his nose in the water, the sea parted.

Operation Nachshon was conceived to open a path between Tel Aviv and Jerusalem to deliver supplies and ammunition to a besieged Jerusalem, cut off from the coast as the British withdrew from Palestine. The road to Jerusalem led through land surrounded by Arab controlled villages, from which Palestinian militia (under the command of Abd al-Qadir al-Husayni) could ambush Israeli convoys attempting to traverse the route.

The operation started on April 5th with attacks on Arab position and, in the pre-dawn hours on April 6th a convoy arrived in Jerusalem from Tel-Aviv. During the operation, the Israelis successfully captured or reduced more than a dozen villages, and took control of the route. Several more convoys made it into Jerusalem before the end of the operation on April 20th.

Operation Nachshon was also the first time Jewish forces attempted to take and hold territory, as opposed to just conducting raids.

Today also marks a first for A Plague of Frogs. We are delivering, for free download, a PC game depicting the Arab Israeli War of 1948. Click for rules, download link, and other details.

 

They Give Me Five Years. Five Years

I hope you do what you said when you swore you’d make it better.

A great irony is that when a people finally throws of the tyranny of a ruling empire, they so often find that it was their imperial masters that had been keeping them from killing each other.

By the time the Ottoman Empire was broken apart, it had long been seen as a system in decline. After their defeat in the Battle of Vienna in 1683, the Empire no longer threatened Europe with its expansion. After the loss of the Russio-Turkish War in 1774, the European powers saw the ultimate breakup of the Ottoman Empire as an inevitability, and began jockeying for control over the eventual spoils. In the mid-1800s, the term The Sick Man of Europe was coined to describe the Ottoman Empire. Compared to its counterparts in the West, it had lower wealth and a lower quality of life. Non-Muslims were accorded a second-class citizenship status but, even within this system, non-Muslims and particularly Christians were better educated and thus developed an economic gap relative to the Muslim majority.

As the Empire continued to decline, nationalist independence movements caused internal stress. Where armed conflict ensued, one might wonder whether my thesis applies. In the Levant, however, despite a multi-cultural population as well as a rising sense of Arab-nationalism independent from Turkey, there was relative peace. Movements for more autonomy tended to focus their efforts in the political arena rather than through violence. This was the period where the Zionism movement was taking form, but it too expressed itself mostly within the confines of civil government.

The final nail in the Ottoman coffin came from backing the Germans in the First World War. In the Middle East, the British had since 1882 occupied Egypt despite it technically remaining a province of the Ottoman Empire. Egypt became a focus of the British war effort early on, both as a base of operations for the Gallipoli campaign as well as to protect the Suez Canal. Eventually, the British took to the offensive in the Sinai and then Gaza, as a way to provide additional pressure on the Ottomans.

In 1917, the British army captured, from the Turks, Jerusalem and the lands that were to become the modern state of Israel. At the end of the war, occupation of the Levant portion of the Middle East was formalized by the Treaty of Versailles. The rule of London replaced the rule of Constantinople.

While the Arab portions of the Ottoman empire were not immune to nationalistic movements, pre-WWI Arabs under the Turks tended to see themselves as part of a Muslim nation. The advent of WWI and centralization of power in Constantinople, following a January 1913 Ottoman coup d’état, resulted in the Sharif and Emir of Mecca declaring an Arab Revolt in June of 1916. It bears considering that this revolt came after the British were at war with the Ottoman Empire. While many reasons were given for the Revolt, including Arab Nationalism and a lack of Muslim piety on the part of the Committee of Union and Progress (the party of the Young Turks and the Three Pashas installed of the aforementioned coup), Hussein ibn Ali al-Hashimi had made agreements with the British in response to their request for assistance in fighting the Central Powers.

Such understandings contributed to Arab unrest post-WWI, as pre-war promises of Arab Independence differed from the disposition of captured Ottoman territory after the war. It didn’t help with Arab sentiment that Britain, now in possession and control of Palestine, had issued the Balfour Declaration in 1917, which supported the concept of a Jewish Homeland in Palestine. While modern Zionism had been an issue for decades, under Ottoman rule it was largely relegated to the political sphere. With the end of the supremacy of a Muslim power in Palestine, Arabs likely felt a more direct protest was necessary to assert their position in Palestine. Arab nationalism was also reinforced by anti-French sentiment in Syria, brought to a head by the March 7, 1920 declaration of Faisal I (son of Hussein bin Ali and a General in the Arab Revolt of 1916) as King.

Events of early 1920, and a lack of response from the ruling British Authorities, caused Jewish leaders to look to their own defense. By the end of March militia groups had trained something like 600 paramilitaries and had begun stockpiling weapons.

Jerusalem Riots

Sunday morning, April 4th 1920 found Jerusalem in a precarious state. Jewish visitors were in the city for the Passover celebration. Christians were there for Easter Sunday. Additionally, the Muslim festival of Nebi Musa had begun on Good Friday, to last for seven days. In excess of 60,000 Arabs were in the streets for the festival, and by mid-morning there was anti-Jewish violence occurring sporadically throughout the Old City. Arab luminaries delivered speeches to the masses, wherein they advocated for Palestinian independence and the expulsion, by violence if need be, of the Zionists among them. By mid-day, the violence had turned to riots, with homes, businesses, and temples being vandalized and as many as 160 Jews injured.

The British military declared, first a curfew, and then martial law, but the riots continued for four days.  Ze’ev Jabotinsky, a co-founder of the Jewish Legion, along with 200 volunteers tried to work with the British to provide for the defense of the Jewish population. The British ultimately prevented such assistance and, in fact, arrested 19 Jews, including Jabotinsky, for the possession of arms. Jabotinsky was sentenced to 15 years in prison, although his sentence was eventually reduced, along with all of those (Jews and Arabs) convicted as a result of the riots. The total number put on trial was approximately 200, with 39 of them being Jews.

By the time peace was restored to Jerusalem, five Jews and four Arabs were dead. Over 200 Jews were injured, eighteen of them critically and 300 Jews were evacuated from the Old City. Some 23 Arabs were also injured, one critically.

The aftermath of the riots left the British occupiers on everyone’s wrong side.

Among the Arabs, the feeling was that they had been wronged by the lack of independence after being separated from the Ottoman Empire. Furthermore, in the Balfour Declaration they saw that ultimately the British would replace their own rule with a Jewish one. The riots also were the beginnings of a unique Palestinian nationalism, separate from Pan Arabism or the Syrian independence movements.

On the other hand, the Jews suspected British complicity as a cause of the riots in the first place. In addition to some unproven conspiracies, the British had several missteps which allowed the riots to escalate. For example, Arabs arrested during Sunday nights curfew were released on Monday morning, only to see the riots continue through Wednesday. The British halted Jewish immigration to Palestine, punishing the Jews for Arab aggression. The inadequacy of Britain’s defense of the Jewish population lead directly to an organized Jewish defense force called the Haganah (“defense”), which would later become the core of the Israeli military.

The incident surely tipped-off the United Kingdom that she had entered into a situation from which there was no easy way out. Nevertheless, for the next few decades she persevered in bringing enlightened British rule to a difficult region.

It would take more than 19 years before the British partially walked back the Balfour Declaration by halting Jewish Immigration to Palestine. It would be almost 27 years, in February of 1947, before British parliament voted to terminate the Palestinian Mandate and hand the issue over the the United Nations.

 

United We Fruit

Makin’ up a mess of fun,
makin’ up a mess of fun
Lots of fun for everyone
Tra la la, la la la la
Tra la la, la la la la

On March 15th, 1951, Colonel Jacobo Árbenz Guzmán was inaugurated as President of Guatemala. Alas, he found his new socialist policies earned him the ire of the United Fruit Company.

While the CIA had participated in regime change before, and while the U.S. had previously meddled in the Caribbean, this was the first exercise of the Cold War performed in America’s own near abroad. It was the start of decades of Cold War confrontation barely a stone’s throw from American soil. It would continue through, and perhaps culminate in, the early 1980s with the region embroiled in a long term conflict and with ramifications, like the Iran-Contra affair, that seriously shook up the U.S. government.

Fortunately for the world, much of this is quickly becoming ancient history. A 1987 peace agreement began to move the region back towards normalcy and what problems still exist are no where near the level of 3-4 decades ago. If we nonetheless want to relive those wild and crazy times, we might do so through a game called “Latin Intervention.”

Fun for Everyone

Latin Intervention is a one-page, print-and-play game freely available from Board Game Geek (and elsewhere.) As you might expect given that introduction, it is very simple. Players assume the role of the two superpowers and take turns placing pieces on the board. The combination of placed markers and a die roll determines the political alignment for the nations of Central America. A player wins by controlling five out of the seven Central American countries. One catch is that unit placement drives up a “Threat Meter,” representing world tensions. This restricts each players actions, as driving the threat meter over the top will result in losing the game.

Despite the simplicity of the mechanics, the game has real appeal due to its “color.” Pieces are labeled to represent the various means that the superpowers used to meddle in the affairs of third world countries: secret agents, monetary aid, revolutionaries, and the military. All of this was done through proxies, so as to not push the world over the edge into a direct conflict between the U.S. and the U.S.S.R. Another Board Game Geek user has redone the game art, making the print and play product look, actually, quite fetching.

Sadly, though, in playing through, the game is not quite right.

Critiques and Criticisms

There isn’t a whole lot out there on the internet written about this game. It’s simple, it’s free, and it’s probably not for everyone. For those that have expressed an opinion, there are positive comments, but also a few consistent criticisms. Please take a look at the game and the rules if you want to follow along with my narrative.

First off, there is a lot of confusion on the Threat Meter track. In the original game design, the threat track has both green and red steps and it isn’t entirely clear in the rules how they are to interact. The consensus is that they are together a single set of steps and each red step is merely two green steps. The new board has eliminated the “red” altogether, and simply has eight green steps, with counters being worth either one or two steps. One does wonder, given some of the other issues, whether we are missing something here, but I don’t see any other way to interpret this. For example, if the tracks were actually in parallel (that is, the red and green steps were separate), the “red” markers would be effectively free as there are only four of them in the game. That wouldn’t make sense at all.

One realization that I made quickly is that, with the eight step threat track, players must try to put the maximum threats onto the board as quickly as possible. In fact, the order seems pretty much proscribed. The Soviets play 1. Missile Base 2. Revolutionaries and 3. KGB agent. U.S must play 1. Carrier Group 2. CIA agent. At that point, the threat level is at maximum, and no more units can be placed. From this point on, no player can reduce the threat level because the other side would immediate use that to place another piece. So the game must played out with six pieces on the board (the U.S. has Panama Canal to start). This seems like a poor use of the game, as it ignores the bulk of the available pieces.

The other area of agreement is that the Missile Base and Carrier Group, the two +5 units in the game, are overpowered. Because control is gained on a roll of “6 or higher,” these two units are essentially instant wins. I haven’t analyzed too carefully, but I’d think the game would probably see the Russians keeping their Agent and Revolutionaries together (for a +5) while moving the Missile Agreement to capture territory. The U.S. could challenge neither (as both sides would have automatic sixes), and would always have to move to protect his other two pieces if the Russians went after them. Like the Soviets, he probably has to group the two in Panama to prevent being taken out. Maybe I am missing something, but the win would have to come from taking a risk that you could neutralize your opponents +5 piece with a lesser piece by a couple of lucky rolls in a row, allowing you to pick up territory.

Strangely, with all of that, there are several players who talk about what a great game it is to play.

If you’re looking at the Board Game Geek site, there is a video review of the game by YouTuber marcowargamer. He also identified those two major areas of problems within the rules. He explains one workaround that seems to be used, and that is to have separate threat tracks for both players. Thus, you can attempt to, judiciously, lower the threat meter, giving up initiative in the current turn for more power during a subsequent turn. He also proposes a rule for making the +5 makers single-use, to prevent them from completely overpowering the game.

He doesn’t mention it specifically, but he has also made a change where challenged countries are re-rolled every turn, not just after the placement of a new marker.

From the video review, he is not indicating whether his modified rules are play-tested and found to be balanced. He is more interested in the game as a launch point for discussions about history. The changes, and particularly the separate Threat Meters, open up a number of different strategies. However, it seems to me that the common threat track is key to the historical perspective of the game. Like the similar mechanic in Twilight Struggle, it captures the feel of the Cold War arms race. You may not want to escalate yourself, but you can’t let those Russkies develop a missile gap.

I’ve come up with my own variant that addresses the balance issues while preserving the single threat track. Thinking about it, it may just be complicating the rules while achieving the same results. On the other hand, I think these rules fit better with the historical “color,” which may justify the complexity.  I’ve posted my rules, so you can see for yourself.

The Rules

They are summarized on this page. I will note, that I will make changes at the link if I discover problems, so at some point the rules are likely to get out of sync with my commentary.

There are two major changes. I address the Threat Meter issue by making deployments to already-controlled countries “free” in terms of threat. Sending aid to an anti-government faction may be seen as threatening on the international level, but sending aid to a friendly government probably wouldn’t be. This essentially accomplishes the same thing as the separate tracks – a player can either play an existing piece now, or gain a new piece for play in the future.

For the +5 units, I assign a threat penalty for leaving them on the board. A Cuban Missile Base or a Carrier Group hovering off of Nicaragua would be seen as a continuing threat. Thus, you can deploy your (for example) carrier for “free”, but only have so many turns to use it before you have to pull it off map. Furthermore, in doing so you probably lower the threat level, opening up opportunities for your opponent. It it likely that this not only makes these units the equivalent of “one time” plays, but also demands that they be used at the beginning of the scenario, when the threat level can accommodate it.

I made the choice to allocate the threat points from the +5 units at the end, rather than during placement. This means if you are going first in the turn, use of a +5 counter might be an instant loss if your opponent can drive the threat meter up to the last position. It further weakens the play of the most powerful pieces.

The second major change I made was to restrict on-board movement. Movement for some pieces is restricted to adjacent countries. The “Aid” markers, otherwise the weakest of units, can be moved without restriction. This further shakes up the balance, as well as creates some strategic value for the map. The map is no longer just seven bins, into which you can place pieces. The layout of the countries matters, and it creates strategic value to hold some countries over others. It also makes some “real life” sense. It’s easy enough to send suitcases full of money anywhere in the world. But to actually move a couple of brigades of revolutionary armies, that might take controlling the ground that you are required to pass through.

One other change I made was to vary the player order. This helps create some back and forth in that, once you start winning the game, you are now disadvantaged by having to take the first turn. It may also throw the game into imbalance. The Soviets have better pieces, and this might make it so they can’t lose. To balance this out, I’ve tweaked the restrictions on the Aircraft Carrier allowing it to be placed directly into a contested country in response to the Soviet’s use of missiles. In terms of that color, projecting military power must be a lot easier for the Americans, who are a) so close to begin with and b) have the naval assets available. But in terms of game play, it gives the U.S. player a counter-strategy to the Soviet’s ability to grab an early lead. With the finite number of threat steps, it may be that this move remains merely a possibility. It all could use some play-testing to see if things are balanced.

So there it is. I’ve not done much checking for balance, and anyone who has a chance to do so before I do will have their comments welcomed.

They call it The Dance

So you think you know what’s going on inside her head

On June 24th, in 1354, the largest outbreak of Choreomania occurred in Aachen, Germany.It subsequently spread to other cities in Germany, the low countries, and Italy.

This phenomenon has been called, variously, Dancing Mania, Dancing Plague, and St. Vitus’ Dance. At the time, the cause was attributed to a curse sent by St. John the Baptist or St. Vitus, due to correlations between the outbreaks and the June feast days of those saints. Much later, the evolution of medical science diagnosed St. Vitus’ Dance as Sydenham’s chorea, an involuntary jerking of the hands, feet and face.

The mass phenomenon of the middle ages, however, is more often considered a social affliction rather than a medical one. The outbreaks are described as affecting up to tens of thousands of people at a time, making contagions or similar causes (such as spider bites) an improbable source.

The Aachen outbreak and other large outbreaks of the Dancing Plague occurred during times of economic hardship. This has suggested one medical cause, a hallucinogenic effect of a grain fungus that can spread with flooding and damp periods.

The affliction was said to be deadly, with the only cure being the playing of the right music.

Similarly, I have been trying to sooth the violent convulsions in this morning’s financial markets by playing selected songs from less troubled times. Feel free to join me.