In mobile game Network Wars (Jim Rutt, 2021) you have one job. Conquer the network. But you’re competing against three AI players trying to do the same.

At first, I won a lot of games. And then, I lost some games. As I didn’t think I was developing much of a strategy, I began to view Network Wars as a slot machine rather than a strategy machine. As I couldn’t see how Network Wars executed each skirmish, I couldn’t extrapolate how a player might successfully plot against the AI.

Every game, I would plunge in with “some tactics” and win or lose due to circumstances apparently out of my control. What choice did I have anyway?

I’m afraid, Joel, you had all the choice in the world.

The Network Wars instructions are quite skimpy. Rutt might as well replace the in-game manual with the phrase “Fight Me”. Still, I’m much wiser now, mainly thanks to Frank Lantz.

Alright, each node carries a number; this is the number of armies on that node. You play the red network and always go first. You have a single verb in your vocabulary, and that verb is attack. When you send a node to attack a neighbouring node, the game uses a sequence of coin flips to determine who wins. Whoever loses a coin flip, loses an army. Network Wars keeps coin flipping until the attacker drops to one army or the defender is overrun. If the attacker wins, one army stays behind and the rest occupy the defeated node.

Once you’ve attacked all the nodes you want/are able to, you end your turn. At this point, new armies equal to the size of your largest node network are distributed across the front line of that network.

Then each AI gets their chance to shine with the same rules. The game continues until one faction controls 24 nodes – or you are wiped out.

That’s good, but it’s not enough to really know Network Wars. You need to figure out what impulse drives the AI’s assault. Going back to mobile favourite Slay (Sean O’Connor, 1995), I have a lot of advice for Slay players on how to avoid provoking the opponent AI:

Positioning peasants on your border is the idiot move as the AI cannot resist an easy kill. Unless it’s distracted elsewhere, it will actively send any available meat grinder to mince up the peasant, knowing it is obliterating hard-earned coin. If you can’t defend, then, for the love of dog, don’t give it a reason to attack.

In Network Wars, sometimes the AI left me alone and sometimes I was shanked hard and fast, and initially I had been unable to draw conclusions about why this was. Reading Frank Lantz’s writeup about Network Wars gave me the wisdom I sought. It seems the AI will only attack nodes of lower value; it won’t do something “smart” like a human would, such as using a weak node to soften up a target, then sending another node in to finish the job.

After this revelation, I began to appreciate how seemingly hopeless situations are actually salvageable. Now I take note of the length of my border as a short border intensifies reinforcements. Now I see the danger of lunging into enemy AI territory too early, because it can create a path of low-value nodes for the AI to follow back into the weak meat of your territory. Now I look for opportunities to encourage the AI to attack each other.

For all this talk of strategy and depth, the thing is, a Network Wars match is so short you could treat it like a clicker game. Move armies, you win. Move armies, you lose. If you don’t know what you’re doing, then moving armies is just an alternative to rolling dice. You can be done in a few minutes then start all over again… which was what I was doing until I read Lantz’s post.

Knowing that the AI always goes after low-value fruit is the crucial piece of information which allows you to get into the prediction business. I could probably have figured this out in time but the game executes AI moves quickly and it’s more difficult to unpack what has happened than you’d think. While you can ask the game to execute the AI attack phase more slowly, so many nodes change colour that it can be difficult to remember who did what to whom in an attempt to fathom why.

However, the key takeaway is that attacking is très dangereux. Attacks will leave a node with value 1 behind them and if that node is accessible to any AI opponents, your attack creates a weakness, like a flaw in the network material. And you can’t redistribute armies from node to node – remember, your only verb is attack. This simplicity makes for some difficult decisions. Sometimes the best move is to wait. Sometimes the best move is to bait.

Here I was encouraging the AI to unlock my six node

In developer Jim Rutt’s podcast, Frank Lantz suggested to Rutt that not every level is beatable. Rutt refuted this because, being the developer, he has a power denied his players – to retry a failed level. He’s never encountered an impossible level. But what about all those coin flips? Rutt revealed a fascinating design detail: all players get the same procedurally-generated levels and even the same dice rolls. If Rutt has found a solution, that solution is there for all players to find.

Some of the levels present a hideous starting configuration and, invariably, I lose those games. So I don’t know if the solutions to tricky levels are buried in amber like those painful Slay levels where players have to mine for the correct sequence of bad decisions that establish an unexpected opening in the far future. Digging up the decision tree with a brute force shovel to find this is not interesting and you’re just back to dice rolls again. But Network Wars doesn’t have this problem because it takes away your brute force shovel. You’re never allowed to linger on a level that crumbles beneath your swipes.

What Lantz found exciting about Network Wars was how the AI are simple machines and not really intelligent. For him, the enemy is the game’s system of rules. Looking at it this way, every level is actually a puzzle with an unpredictable element. While the narrative paint tries to make you feel like you’re part of a conflict, what you’re really doing is solving a puzzle.

But Lantz goes further, highlighting that once a player can reduce their play to an algorithm, that’s when a deep game ceases to be. You’re no longer playing but executing the solution to a puzzle. I’d have to agree. The learning bit where you’re assembling strategy is the best bit of cerebral games like this. Figuring out the solution algorithm is the actual victory over your real opponent: a system of rules.

Making progress

Although not exactly the same situation, I remember the euphoria of achieving something monumental after I had figured out how to solve all thirty levels of the Domination mode in Death Crown (CO5MONAUT, 2019). But that moment was simultaneously the death of the game. There are no more highs to be found unless you can boost the challenge. Such as, oh I don’t know, defeating the game in front of a live audience?

Despite the basic truth that mastering a game like Network Wars is about coding an algorithm to defeat an algorithm, I am still enraged whenever one of the AI enemies bores a tunnel through my nodes. Nothing can explain the hatred I have for that AI, which was just following orders. Sure. I’ve heard that one before.

In the algorithm war, I am, first and foremost, human.

Download my FREE eBook on the collapse of indie game prices an accessible and comprehensive explanation of what has happened to the market.

Sign up for the monthly Electron Dance Newsletter and follow on Twitter!

5 thoughts on “Algorithm War

  1. This post arrived at a good time – my girlfriend and I were on holiday, and this was a perfect game to fill a few minutes here and there. I’ve played about ten games; she’s played about twenty. Our win/loss ratios look about the same as yours. We bootstrapped from the knowledge I gleaned from reading this. So, thanks Joel, and thanks Frank Lantz. :g:

    It is interesting how even knowing the trick of the game, you can still slip up if you don’t pay sufficient attention, or if you overreach. My preferred strategy is to consolidate into a corner and control chokepoints, even if that means burning what I leave behind. Turtling, essentially, on the basis that a smaller border concentrates your reinforcements, making it easier to both direct the AI away from you, and build big stacks to push forwards.

  2. Hi Shaun! Yes, retreating into corners and minmizing your border is crucial. Still, some of those initial setups are straight-up nasty, when your nodes are parked next door to a couple of 8s and inaction seems the best course. (Inaction is often the best course on the first turn!)

  3. The game seems a bit like an (even more) abstracted Risk. I propose that retreating into a corner shall henceforth be known as ‘doing an Australia’.

    Risk is a social game and the strategy necessarily entails a lot of game theory. In the absence of other motivation (personal grudges etc) players tend to have a few basic objectives: consolidate a continent, prevent another player from doing the same, or attempt to knock someone out. Advanced strategy tends to be about making yourself look neither strong nor weak enough to invite aggression.

    You note that a predictable AI turns that into a different kind of dynamic. Levels become more like puzzles to be solved. This is sometimes a charge thrown at single player strategy games perjoratively; they are puzzles games in disguise. But a puzzle game with random elements? Inconsistent behaviour? Not to start a war over definitions, but this seems like a contradiction in terms.

    The inability to replay the level to scour for more solutions creates scenarios where the player can only second guess themselves. Was that level really possible? Was it a matter of luck or skill that caused the failure? This seems to be an inevitable source of tension.

    Not that I want to suggest it is a design ‘failure’ or anything. If games are the right length to knock out enough wins in a session and not dwell overmuch on the losses, it might not matter at all.

  4. CA, it’s a curious mix. You can’t predict what’s going to happen but you know it’s clockwork. One of the moves that brings me down is an attempt to weaken a foreign node but not take it down – and sometimes the deterministic random numbers overrun the node. Then I get exposed to much stronger node that was trapped behind it which then blasts through my territory on the next turn.

    This just happened to me – I had Australia’d quite safely but this small unexpected overreach began a decline which seemed impossible to stop. Turn upon turn the blue AI kept digging through my nodes. But I created an opening for another AI to do some damage against blue while I consolidated around five last-ditch nodes.

    I won and it felt like a victory that was only possible because I didn’t give up (lord knows I’ve quit a number of games in anger) and chose to reduce the risk in my actions.

    Most of the time it is about risks even though I’m pretty sure there are difficult scenarios which can only be won because the dice are in your favour at the right sticky moments.

    Bring such a short game means it’s easy to dismiss the tricky levels and try again. There’s always a win around the corner.

    But the point about it becoming a puzzle – the idea that you’ve found the Perfect Strategy – only applies if you’re winning most of the games (even if randomness is involved). If you’re not, you have to ask the question whether the game has “unfair” levels that are impossible to strategise properly or whether you have more to learn.

    Apparently a good Network Wars player has a success rate of around 75%. I hate this statistic because it only brings self-doubt to the table.

    My current 100 game average is 50%.

  5. Very interesting. I understand that strategy games are a broad church, and not always strictly perfect-information or strictly deterministic. Indeed, some designers (Sirlin et al) would insist on some mechanisms of this nature to keep a game from being solvable. Poker is indisputably a strategy game, so is Blood Bowl. I guess so is Solitaire if you play by the more difficult rules (no recycling the deck, nultiple draw etc).

    As regards puzzles, it’s even more interesting. If I follow correctly, we’re talking at one step removed – the meta-puzzle of developing a strategy that gives a consistently better-than-a-coin-flip result, after which you can retire from the field.

    This also reminds me of Sirlin, who in his Playing to Win essays contended that it was possible to be Good at rock-paper-scissors, i.e. devise a strategy that improved success above what most people would be the ceiling of 1-in-3. To this day I find this hard to swallow on an intuitive level, even allowing for the existence of RPS tournaments that have had multiple time winners.

Leave a Reply

Your email address will not be published.