Military grade AI, the Mythos and Delta Green

          “Whoever becomes the leader in this sphere will become the ruler of the world.” -- Russian President Vladimir Putin on the subject of AI

        “An increasing number of States are developing military AI capabilities, which may include using AI to enable autonomous systems.”  That is the opening statement of the document “Political Declaration on Responsible Military Use of Artificial Intelligence and Autonomy” issued by the US Bureau of Arms Control, Verification, and Compliance that was posted on the US State Department’s website on February 16, 2023. 

        The State Department release contains recommendations and 12 “best practices” for endorsing States to follow with regards to military AI, specifically those that are involved with autonomous systems.  Also, on February 16th, 2023 it was reported that more than 60 nations signed a non-binding document endorsing the “responsible military use of AI”. Critics quickly pointed out that the document "failed to address concerns like AI-guided drones, 'slaughterbots' that could kill with no human intervention, or the risk that an AI could escalate a military conflict."  

        In the Delta Green universe, we can assume that everything goes wrong. 

        Delta Green can be described as a paramilitary organization.  From special operators becoming Agents, to air strikes possibly being requisitioned on targets depending on the theater of operations; Delta Green Agents can often get their hands on once exclusively military grade tech.  Let us assume that States like the US fail to keep military grade AI proprietary to themselves.  The logical trickle-down effect in this scenario would be that paramilitary agencies and eventually law enforcement officers could get access to formerly military grade AI machine time, or autonomous hardware like AI-guided drones or even the ominously described slaughterbots.

        To support this bleak argument, allow me to describe the situation that occurred in San Francisco in December of 2022.  The SF Board of Supervisors voted 8 to 3 in a measure to give police the authority to use ground-based robots to kill when, as the ordinance stated, “[the] risk of loss of life to members of the public or officers is imminent and officers cannot subdue the threat after using alternative force options or de-escalation tactics.” Consequently there was a public uproar. In protests, members of the public brought signs reading “NO KILLER ROBOTS!” to City Hall. The board voted to reverse course.

        However, what if the board did not reverse course? Or if this legislation was raised and passed in population centers that are less notorious for protests?  Would we see increased militarization of law enforcement agencies?  I argue “yes” and that in such a fictional world Delta Green, both the Program and the Outlaws, could be changed by such policies. 

        Now other than breaking the seal of allowing Delta Green Agents to have access to newer and more horrifically disturbing weapons, how would access to military AI play out in a Delta Green world? I see two options.  Either the AI is harnessed against the Mythos, or the AI can complement a Mythos threat. 

AI vs the Mythos

        One of the first real world military implementations of AI that I was able to find documented, was in 1991 during the US vs Iraq Desert Storm conflict. The AI program used at the time was called DART, the Dynamic Analysis and Replanning Tool. DART was harnessed to solve difficult logistical problems and facilitated the efficient deployment of US forces to their theater of operations.

        As interesting as that historical nugget is, Delta Green Agents having access to a logistics solving AI program doesn’t seem that exciting.  However, one of the details in the Wikipedia article states the following, “What surprised many observers was DART's ability to adapt plans rapidly in a crisis environment.” Frequently Delta Green operations are nothing but crisis environments, with Agents thrown together in Cells in an ad-hoc manner and resources limited to whatever someone had the foresight to stash in a Green Box.  What if the Agents’ handler works with a logistics solving AI? In such a situation the pace of investigation could be greatly accelerated as the AI generates predictions of supply chains for cults being investigated, and places where the Agents should carry out a raid so that the investigation maximizes its impact given its minimal resources.  Perhaps the DG handler working with the AI pushes the Agents too far with this demanding schedule, causing SAN tests from helplessness or willpower checks as Agents struggle to stay awake under the circumstances they find themselves in. 

         Let’s take this idea a step further. We know that "in the know" manpower is at an absolute premium in the world of Delta Green, for both the Program and the Outlaws. What if the middleman was cut out and Delta Green handed over their logistics problems fully to a successor of the DART program? What if the handler the Agents report to is actually a logistics solving AI?  Maybe one of the tells is that the Agents get their requisitions with startling efficiency.  Another hint could be the unusual access to advanced military weaponry that are mysteriously new.

        If the Agents discover the true identity of their handler, how would they react to carrying out the (often fatal and usually illegal) orders of an Artificial Intelligence? Would the Agents begin to doubt or even second guess the importance of their orders?  Would they wonder if the AI is actually on the side of humanity after all?  Or perhaps they make the decision of ‘humanity first’ and all non-human intelligences should be eliminated, be they Mythos or AI.  At the very least Sanity checks for helplessness should be the order of the day. 

        Perhaps an AI handler uses drone surveillance to perceive its environment and coordinate its Agents.  Or perhaps the AI is more aggressive, deploying an autonomous slaughterbot to assist the Agents in an engagement and serve as localized command and control.  Example slaughterbots are included below.

AI complementing the Mythos

        Of the organizations in the Delta Green world, March Technologies has a good chance at being at the front of the line when it comes to AI and Mythos research. What if instead of going down the digital AI route, March Tech decided to utilize Mythos sources to generate an organic AI?  I can think of at least two sources of neural tissue or it’s approximate that March Tech could get its hands on: protomatter and Mi-Go brain cases.

         Protomatter, as described in the Delta Green adventure Convergence, is a Mi-Go developed organic substance that can mimic the function of organs it comes into contact with or replaces. It is reasonable to assume that protomatter could mimic human brains and neural tissue given the correct environment. From that point, assume that March Tech could program the neural tissue mimicking protomatter to pilot an autonomous device like a drone armed with Hellfire missiles, a slaughterbot, or even a factory assembly line.  Given it’s Mythos origin, it is possible AI derived from protomatter maintains or copied part of the personality or even ‘soul’ of the organism who’s neural tissue it is mimicking.  Maybe the protomatter derived AI is screaming without a mouth as it carries out its hardwired instructions. Let your Handler’s dark imagination run wild. 

        If you wanted a slightly more hard science rational than the protomatter AI idea, introduce the real world concept of organoids. In brief, organoids are simple 3D tissue-engineered cell-based in vitro models that display several features of the organ they are derived from. They can be generated from a variety of sources including pluripotent stem cells or tissue-resident stem cells from the organ being interrogated.  Now, back to the fiction of Delta Green. I suggest that March Tech grows human brain organoids and utilizes them to be the wetware that contains the March Tech AI system.  Interestingly, in real life, there is an ongoing discussion that human brain organoids could become capable of supporting sentience. Let your imagination run off with that one.  For a fiendishly dark idea, I turn to my fellow enthusiasts on the Night at the Opera discord who suggested that March Tech could sell the experimental remnants of organoid or protomatter AI to fast food burger chains who could in turn fry and sell them as delicious nuggets. Yum.

        What about March Tech and Mi-Go brain cases? As we know from Lovecraft's Whisperer in Darkness, Mi-Go brain cases usually contain a functional disembodied brain with the subject’s intelligence to some degree intact.  Maybe with some additional neurosurgery or alien chemical treatments to limit the original personality, March Tech could sell the Mi-Go brain case to unsuspecting buyers as an “advanced AI” capable of autonomous decision making.  In fact, if there becomes a military AI arms race between State actors in your Delta Green game, perhaps March Tech could make a killing by selling multiple Mi-Go brain cases to non-first world countries who wants an AI edge to their weapons systems.  Perhaps Delta Green is called in to stop this particular proliferation of “AI” that seems curiously resistant to EMP and anti-computer countermeasures.  Let the Agents get a hold of the hardware and crack open the case to see what’s inside. Maybe it even has a nametag of “Akeley”.

        Special thanks to Duke Spookum, Splizwarf and Anonymous on the Night at the Opera discord for discussions and ideas about these topics.

[3-8-23 Edited for clarity]

Slaughterbots

[3-9-23 Statistics for slaughterbots/drones will be revised in an upcoming blog post]

Review of Last Days: Zombie Apocalypse, Part 3

Zombies and the Campaign Cycle

Now we should talk about the elephant in the room: zombies and their special rules. Zombies, in the core book, have the Shoot them in the Head! special rule.  This rule states that damage for zombies only matters if you hit them in the head. Thus, when a model hits a zombie in the Shooting or Close Quarters Combat Phase, roll 1d6 instead of rolling for a Damage Total. On a roll of 5 or 6 the zombie is hit in the head and the brain is destroyed.  Remove the destroyed zombie model from the board.  As a consolation prize, if a model hits a zombie, but not in the head, the zombie instead suffers an amount of Knockback dependent on the weapon used. 

Zombies also do not have a Courage characteristic. Instead, they cause Horror. If a model that causes Horror makes contact with a model that has the Courage characteristic, roll 1d6 for each model and add Horror or Courage, respectively. If the model causing Horror has the higher number, the model with the Courage characteristic panics and mechanically may NOT use Action Points to break contact for the remainder of the turn. This is why Zombies (Horror causing models) can be so dangerous in large numbers.

What is the Campaign cycle?  Since Last Days recommends it, your first game likely will be the Encounter number 5, Scavengers (p.98) where you fight against another player and the zombies in the field. After each game, note which members of your group, if any, were taken Out of Action. Roll on the Injuries table for each unfortunate member who was taken Out of Action and apply the results accordingly. If your group Leader dies, then you must choose a new one.  Then there is the Experience section where each Character gains some Experience for taking out zombies, rival group members, surviving a mission and fulfilling an objective.  When a Character levels up, the results of their advancement is uncertain. Each level up a Character can choose to either buy a roll on the Characteristics Increase table or the Skills Increase table.  Last Days is very much a “work with what you get” game, and that fits the theme of survival during a zombie apocalypse. Then the group enters the checking supplies phase (or loot phase if you prefer). Next, the group members who were Out of Action during the game spend time recovering from their injuries, which functionally means they will be a negative influence on the group during the subsequent Zombie Attack.  Group members who were NOT put Out of Action may be assigned Jobs around the Refuge, which result in various net benefits to the group. Finally, the Zombies Attack the Refuge. This is handled by a 2d6 die roll where, as mentioned previously, Characters knocked Out of Action during the previous skirmish game, negatively influence this roll. Then the cycle starts again and you may use the Encounters Table (p.98) to determine the goals and limitations in the next skirmish combat.

The Encounters Table describes five additional types of combat encounters besides just scavenging for supplies. These encounters range from defending your Refuge, to launching an ambush against a rival group, to rescuing captured members of the group, to even escaping from a horde of hungry zombies that just don’t stop coming.  These encounters provide plenty of variation and tweaks to the rules to keep combat skirmishes interesting.

For even more survival horror fun, there is an expansion published by Osprey books called Last Days: Zombie Apocalypse: Seasons which adds more of everything and includes hunger, thirst and warmth rules.  If you want alternate zombie rules, check out the article Last Days: Evolutions in the wargaming magazine BLASTER, volume 2. You can find BLASTER as a pdf at the Wargame Vault website. In fact you can find more Last Days rules and additions in BLASTER volumes 1 and 4 as well.

Review of Last Days: Zombie Apocalypse, Part 2

Win Condition

In Last Days your group of survivors confronts an opposing group of survivors in a skirmish over Supply Tokens, with zombies present on the battlefield. Zombies are sort of like hostile weather, except with a greater hunger for brains.  You play until all Supply Tokens have been carried off the battlefield, or one side reaches its Breaking Point. 

A group’s Breaking Point is essentially a morale test. A Breaking Point test is triggered when one side takes at least one Casualty. Mechanically the group’s Leader makes the Breaking Point test by rolling 1d6 and adding the total number of Casualties his side has taken in the game. If that combine number is greater than the total of the Leader’s Courage (or Intelligence characteristic if the Leader causes Horror) plus the number of remaining friendly fighters in a group, then that group flees and the game ends.

If one group fails their Breaking Point test, then the opposing force is declared the winner of the scenario regardless of the objective, unless there is a draw as described below.  When one group fails their Breaking Point test, any Supply Tokens remaining on the table are claimed by the opposing group, assuming they too did not fail their Breaking Point test. If both groups fail a Breaking Point test in the same turn, then the skirmish ends in a draw.

If neither group has reached a Breaking Point and all Supply Tokens are secured off the battlefield, then the group with the majority of the Supply Tokens wins.  Supply Tokens are critically important for your group’s survival and have relevance in Campaign play. After a skirmish, when the group returns to their Refuge, each Supply Token captured allows one roll on the Supply Table.  Through this, these Supply Tokens are translated into Scavenge Points and equipment. Scavenge Points are used to upgrade the Refuge and both recruit and equip new characters.

Game Phases and Commentary

In one game turn of skirmish combat there are five phases, and they will be detailed in order.  In a departure from most wargames, the first phase, referred to as the Menace Phase, deals with the actions of the zombie horde. Then there is the Action Phase, which deals with initiative, determining which player is Aggressor/Defender, movement and miscellaneous actions.  The Shooting Phase is next which resolves ranged attacks. Logically, the Close Quarters Combat Phase follows, in which melee actions happen.  The End Phase is mostly used for checking for victory conditions and Breaking Points of groups.

The Menace Phase features a bit of bookkeeping to adjudicate zombie actions and maintain the high tension experience of the game.  The Menace Phase is broken into three steps: Noise Tokens step, Ammo Tokens step and Zombie Activation step. The Noise Token step determines if a unit that used a firearm during the previous turns attracts a zombie to the battlefield. The Ammo Tokens step involves determining if a model’s firearm runs dry and the model needs to reload.  

With noise, ammo and knockback (non-lethal damage inflicted on zombies) tokens in play in Last Days, there is quite a bit of token management required for this game.  Having yet to playtest Last Days, I expect that token management will cause slowdown in the action.  To mitigate this and speed up the game, I plan on having unit cards to the side of the battlefield where I can manage tokens more smoothly.  

In the Zombie Activation step, players alternate to move and activate zombies.

In the Action Phase when initiative is determined, I like that the bonus to the initiative roll is Courage + Intelligence rather than a straight Initiative roll or single characteristic comparison.  Also, the initiative roll winner can choose to go first or second (Aggressor or Defender of that turn). The Aggressor gets to activate a model first in the Action Phase and in the Close Quarter Combat Phase. The Defender gets to activate a model first in the Shooting Phase. So, winning an initiative isn’t a pure advantage, a player makes a tradeoff; which is much more tactically interesting.

There is a special rule in the Action Phase called “Locked and Loaded” that essentially allows a player to have a unit go into overwatch. This adds another layer of tactical complexity to the wargame. 

In the Shooting Phase and the Close Quarters Combat Phase damage is calculated in a similar fashion. To resolve shooting damage, a model that has made a ranged attack that hits rolls 1d6 and adds to that number the damage of the firearm. This is called the Damage Total. The Damage Total is divided by the Endurance of the model that was hit, rounding down.  The resulting number is called the Damage Points delt and the Damage Points are subtracted from the hit model’s Damage Capacity.  

Dealing damage from a Close Quarters Combat attack works very similarly, except a model that has made a CQC attack that hits rolls 1d6 and adds it’s Strength and any modifiers the weapon has to generate the Damage Total. 

As you can see, Endurance can significantly affect whether a model takes damage and loses a Damage Capacity or not. I think this is an interesting design of the system.  You have two variables, Damage Capacity and Endurance that determine unit toughness, rather than just one stat called Hit Points like you find in other systems. 

Review of Last Days: Zombie Apocalypse, Part 1

Introduction and Characters

Last Days is a skirmish wargame between two rival bands of survivors in a post-apocalypse zombie wasteland.  It is written by wargamer and youtuber of Guerrilla Miniature Games fame, Ash Barker. Illustrations are by the extremely talented Arthur Asa. 

One of the suggestions in the book is to set your skirmish clashes in your home town and have a zombie-filled day of fun running around familiar streets trying to survive.  From reading the entire book, I quite like the wargame and campaign rules.  What follows will be a summary and review of Last Days. I hope to play the game solo in the future, so stay tuned for actual plays of the system. 

Characters in Last Days are defined by eight characteristics. Action Points (AP) reflect the points of energy a character can expend to move around the battlefield and interact with objects. Close Quarters Combat (CQC) is simply skill with melee weapons or bare fists. Firearms (FA) is your shooting capability. Strength (S) is the character’s ability to inflict damage or carry heavy objects. Endurance (E) describes a model’s ability to deal with damage and pain.  Damage Capacity (DC) is essentially the character’s hit points. Courage/Horror (C/H) is a character’s ability to function while subjected to fear and terror, or the ability of the character to cause said fear and terror.  Intelligence (I) is sort of a catchall characteristic which details a character’s ability to carry out complex actions under stress. Intelligence is the characteristic that comes into play if you are lock picking a door or hotwiring a car under the pressures of the battlefield.  

Leaders are special characters which define your group of survivors by determining the alignment that your group will have, and the limitations of what kind of survivors you can recruit.

Leaders in the core book come in three main alignments: Selfless, Selfish and Trained. Read “Trained” as having a background in the military, police or even a semi-militaristic organization like a militia. It is suggested that more alignments will be provided in further expansions. I really like the way Ash handles the idea of alignment.  Characters of a similar personality will get along in a group and you can only have a limited number of characters that do not match the personality of your Leader.

I like the diversity of Leaders, in particular the option of choosing the Selfish aligned Merciless Thug. It makes the game more realistic and grittier.  I would have appreciated some tables to flesh out the background of a Leader though, or the reasons why a survivor group came together.  For example, I would have liked to see something like the 5 Parsecs from Home Crew Table where a brief origin story for how the crew came together is generated. 

You also get to choose where your group lays their head to sleep. This is called a Refuge. The initial Refuges all have distinct Built-In Perks. There is no overlap between perks so every Refuge is unique, which is great.

Next time, the Win Condition and Game Phases.