Interview: Ezra Sidran

Right now General Staff’s most effective recruiting sergeant isn’t a smooth-talking influencer, a fancy trailer, or a feature list as long as a Chassepot rifle, it’s a collection of digitised academic papers tucked away in a dusty corner of an obscure website. The pdfs stacked here are the reason I don’t care two hoots* about the game’s frumpy graphics, familiar Nineteenth Century battle line-up, and lack of interest in campaigning. They promise something rarer than rocking horse manure – computer-controlled opponents that think and possibly even learn like real commanders.

* I do, it has to be said, care one hoot.

Before emigrating to academia in the early Noughties, holy grAIl seeker Ezra Sidran made a name for himself crafting innovative computer wargames with unusually broad purviews, analytical enemies, and powerful editors.

Universal Military Simulator (1987), the first (and last?) scrap sim to offer wireframe battlefields, was seven years in the making and spanned two millennia of land warfare.

WeGo in a time when simultaneous order execution was almost unheard of, UMS was also one of the first in its field to embrace delegation. If the unit counts at Arbela, Hastings, Marston Moor, Waterloo, or Gettysburg intimidated or you fancied roleplaying a subordinate commander you could let a friendly AI, either cursorily briefed or left completely to its own devices, handle the lion’s share of your army.

The sequel, UMS II: Nations at War (1990), covered just as much history, but vastly more geography. The bundled Napoleonic, Operation Overlord, and Alexander the Great scenarios were international affairs fought on a mod-friendly global map dotted with economy-influencing cities, and baked, chilled, and dampened by a swirling dynamic weather system.

Ezra returned to pitched battle orchestration in The War College: Universal Military Simulator 3 (1995). Boasting the kind of plump embedded encyclopedia that, sadly, is seldom seen these days, the title offered engrossing treatments of Pharsalus, Austerlitz, Antietam, and Tannenberg.

Turns were banished and unsurprisingly there wasn’t a hex or strength step in sight. PC Gamer’s ‘Desktop General’, William R. Trotter, was bowled over:

“The War College is a mature, deep, thoughtful simulation that embodies a radical departure from the wargaming norm… Obviously, this program is aimed at the serious student of military history rather than the casual gamer. That doesn’t mean, however, that it’s technically intimidating (the interface is a joy to use, and the manual is excellent) or that it shortchanges the fun factor. After getting my butt stomped twice by the computer, I finally managed to win Tannenberg on the Russian side, and it was as joyous a moment as I’ve gotten from any wargame.”

The collapse of publisher GameTek meant royalties from UMS3 never reached Ezra. Lured away from game development by the deep-pocketed US military which was keen to use his skills and creations to train leaders, today’s interviewee might never have returned – never embarked on the project that is Universal Military Simulator 4 in all but name – had it not been for… well, I’ll let him explain.

THC: After a quarter of a century away from commercial game development what persuaded you to return to the fray?

Ezra: In 2003 my wife suggested that if I were to return to academia and earn a doctorate in computer science from a Research 1 university concentrating on artificial intelligence making optimal tactical and strategic decisions within a predefined battlespace I might find support from the U. S. Department of Defense. Actually, what she said was, “A PhD behind your name would give you legitimacy,” And, as usual, she was right. DARPA (Defense Advanced Research Project Agency) funded a part of my grad school which produced TIGER (Tactical Inference GenERator). And, then after receiving my doctorate, DARPA funded MATE (Machine Analysis of Tactical Environments) that produced real-time battlefield analysis and COA (Course of Action).

Then in 2012, due to some extraordinarily weird political machinations that even today I do not fully understand the U. S. Congress decided to cut all budgets 10% across the board and DARPA decided to reflexively chop all 4CI (Command, Control, Computers, Communications, and Intelligence) AI funding. I was attached to a much larger project, for funding purposes, and when it died, I died. So, I returned to academia and taught Computer Game Design at a major university and had a ton of fun.

And then in late 2013 I was diagnosed with a couple of really rare and usually fatal diseases: AL Amyloidosis and Multiple Myeloma. So in 2014 I had an autologous stem cell transplant, followed by a year of chemo. I taught through most of it. But, by the fall of 2015 it was obvious that my voice had been badly damaged (probably by the chemo) and I couldn’t lecture for any length of time without a coughing fit. I also didn’t have the strength to run up and down the lecture hall and teach with the energy that I was used to. So, I had to retire as a Visiting Assistant Professor of Computer Science.

The funny thing is it turns out that I’m an expert in this very narrowly defined area of expertise: computational military reasoning. So, I decided to take all my research in computational military reasoning, all the AI that DARPA paid for and didn’t want (and yes, this is legal, legit, and I own it) and make a wargame AI that learns. Specifically, my doctoral thesis is on unsupervised machine learning of battlefield analysis and this is a subject that I’m really excited to talk about.

THC: Can you explain TIGER and MATE in a little more detail? What part do they play in General Staff?

Ezra: TIGER was my proof of concept program for my doctoral thesis in computer science. In a doctoral thesis you state a hypothesis (or hypotheses) and then thoroughly investigate it. TIGER had two hypotheses:

Hypothesis 1: There is agreement among military experts that tactical situations exhibit certain features (or attributes) and that these features can be used by SMEs to group tactical situations by similarity.

Hypothesis 2: The best match (by TIGER of a new scenario to a scenario from its historical database) predicts what the experts would choose.

This is a just another way of saying that military experts can look at a battlefield map and agree that certain features (such as BLUE has the high ground, RED has an restricted avenue of attack, BLUE has an open or exposed flank, etc.) are present and furthermore the program that I wrote (called TIGER) can also do this and TIGER’s battlefield analyses are statistically indistinguishable from those done by the experts. In other words, TIGER is a SME.

TIGER accomplished this by using unsupervised machine learning. Most people have seen supervised machine learning (even if they don’t know the term). When Netflix suggests a movie or Spotify suggests a song, that is supervised machine learning. You, the user, are the supervisor. By ‘liking’ a movie you are ‘training’ the machine (in computer science programs are called machines for reasons that I won’t bother getting into here). The problem with supervised machine learning is that it usually isn’t that great (because you watched the Wizard of Oz you’ll love Texas Chainsaw Massacre!). Unsupervised machine learning isn’t trained. It uses no human involvement whatsoever. An unsupervised learning machine is ‘fed’ a stream of data ‘objects’ and then makes sense of all the objects by putting them into similar groups. When it ‘sees’ a new object it ‘says’, “I’ve seen something similar in the past and it goes over here in this group.” In TIGER’s case it is given a new tactical situation and says, “This tactical situation is similar to the battle of Gettysburg (for example).”

After receiving my doctorate DARPA funded expanding TIGER into MATE (Machine Analysis of Tactical Environments) and it demonstrated the ability to make real-world tactical analysis and generate Course of Actions.

So the AI for General Staff will employ TIGER’s and MATE’s ability to analyze a battlefield and learn from every time it ‘plays’ a simulation.

THC: Which is easier, coding an AI that defends competently or one that attacks well?

Ezra: Offense is harder. It’s probably apocryphal but I swear I once read that Napoleon would quiz his staff as they rode through the countryside on optimal defensive positions. It’s easy. Position your army between the enemy and their objective. Take the high ground and protect your flanks.

There are five tactical offensive maneuvers all of which can be implemented with algorithms:

THC: From an AI coding perspective, are there forms of warfare or periods in military history that you find particularly intimidating?

Ezra: Yes; modern warfare because it adds more dimensions. Every time you add another dimension you square the complexity. Warfare is always about the high ground. And, with air superiority, you can drop airborne units behind enemy lines. And space is the ultimate high ground.

THC: When I criticise a dev for, say, failing to convey Rommel’s genius am I being unreasonable?

Ezra: Rommel was a brilliant tactician and strategist. I’m not familiar with the AI for SGS: Afrika Korps so I can’t weigh in on the argument. But, I can say I’ve seen some terrible Rommel AI in my time.

THC: Which of General Staff’s 30 historical battlefields prompted the most AI code tweaks?

Ezra: None. Remember that General Staff allows (encourages!) the user to create new scenarios (battles). You can’t write ‘set piece’ AI for situations that haven’t been created yet. The only AI that will work is a general purpose AI that understands concepts like ‘defend the flanks’ ‘don’t attack when you have restricted avenues of approach’. Furthermore, I intend to incorporate TIGER’s unsupervised machine learning which would improve the AI the more you play. I’m reasonably certain that no game has ever done this before.

THC: Wargame AIs often aren’t great at protecting artillery and supply units. Will yours react when it spots me edging cavalry toward a battery or supply wagon?

Ezra: I certainly hope so.

THC: When a player selects “Complete Fog of War” in GS what handicaps are they embracing?

Ezra: I think this screen shot will demonstrate:

This is what General George McClellan could see of the battle of Antietam. This is complete Fog of War. Furthermore, McClellan could only communicate to his subordinate commanders via courier. And his subordinate commanders would have to dispatch their couriers to the actual fighting units introducing delay into the orders:

When playing in ‘Simulation Mode’ (the highest reality level) subordinate commander’s Leadership Value will affect how soon orders are relayed. The Leadership Value of the commander of the unit receiving the orders will affect how and when the unit will respond to the orders (think Burnside at Antietam).

THC: When was the last time you were seriously impressed by a wargame AI?

Ezra: This may be stretching the definition of a ‘wargame’ but I don’t think so. My wife and I like to play ‘couch co-op’ console games. The AI on Divinity 2: Original Sin really impressed me (we played it on the PS4). Tactically, the AI was dealing with melee, ranged weapons and magic attacks and defensive tactics. I was very impressed and more than a few times angry.

THC: The hexagon remains an almost sacred object in digital and board wargaming. How do you feel about them?

Ezra: That’s a great question! Maybe you’ve read my blog post The Problem with Hexagons? Well, the problem with hexagons is that they don’t allow true movement, true line of sight calculations, true, well, anything. But, that said, proposing them for wargames was an extraordinary burst of genius by the legendary John Nash (of A Beautiful Mind, Nash Equilibrium, and Nobel Prize recipient fame) who first proposed it in a paper with R. M. Thrall (Project RAND, 10 September 1952; available as a free download here). But, that said, hexagons are a crutch for bad math. If moving a unit one hex forward (north) costs 1 unit then moving the same unit to the next adjoining hex to the ‘north east’ (45°) should cost 1.41421356 units. But it doesn’t. On a hexagon board it still costs ‘1’. If you want infinitely scalable fidelity a matrix is the only way to go.

THC: Universal Military Simulator broke hectares of new ground. Looking back on it, what do you regard as its most notable achievements and likeable traits?

Ezra: Before UMS game publishers didn’t know that a wargame (much less a computer wargame) could gross over $5 million retail. And that was on a development budget of $15,000. I think it was also the first to introduce add-on content. We had successful Civil War and Vietnam scenario collections even though users could create their own armies, maps and scenarios. Also, this is a theme that runs throughout all my work, I don’t claim to be the definitive expert on anything military. If I have the strength of Doubleday’s 1st Division, I Corps at the battle of Antietam as 2,915 and you think it is 2,223 here’s a utility program that will make it very easy for you to change that value in the scenario.

^ The General Staff: Black Powder army editor.

In fact there are a lot of values that all I can guarantee is that I spent a lot of time researching and here are my best guesses. Doubleday was a pretty good leader. I’ve got his Leadership Value set at 78. Should it be higher? Well, here’s how to change it.

Since the beginning I’ve concentrated on producing very detailed simulations of complex human events. These events are so complex and have so many moving parts that I think they can only be accurately simulated on a computer. More than anything I want to give the user complete access to all the moving parts.

THC: Can you trace your interest in warfare simulation back to a particular incident or person?

Ezra: When I was about nine or ten years old, my friend Carl Hoffman who lived across the street from me in Racine, Wisconsin, introduced me to Avalon Hill board wargames. I think, looking back now fifty-six years later, this may have been one of the most important events in my life; judging by my CV it certainly appears so. It was over the course of a lifetime that I came to realise that very complex events in human life could be accurately modelled. But my first glimpse of this was moving cardboard chits around a map of Gettysburg on a living room floor. Not surprisingly, Carl is now a history teacher.

THC: Why did you choose not to code General Staff’s engine yourself?

Ezra: I wrote TIGER and MATE in the C++ programming language. For a number of reasons I decided to write General Staff in C# using Windows Presentation Foundation (WPF). While asking questions about WPF on the Microsoft Developers Forum I encountered Andy O’Neill who is not only proficient in WPF but an avid wargamer. Andy ended up rewriting my earlier work (the Army Editor, Map Editor and Scenario Editor) and making them a lot better. He really turned them into a suite of interlocking applications like Microsoft’s Office. Andy is writing the Game Engine from my design document. I’m a novice in WPF. Andy is an expert.

THC: Does General Staff have an ETA yet?

Ezra: General Staff: Black Powder is scheduled to ship in 2021. I am confident that at least early backers will be playtesting the game within the next six months or sooner. The real problem, and I’ll be blunt, is that researching all the battles and creating all the maps and OOBs is very time consuming. I’m doing all that work by myself. I do have some help with the artwork from two friends (Ed Kuhrt and Ed Isenberg).

THC: Are we likely to see it on platforms like Steam?

Ezra: Yes. It will definitely be out on Steam.

THC: Thank you for your time