DarkForestSim 2: How efficient do killer civs have to be to overcome subjugators and cooperators?
Modeling the Dark Forest Hypothesis
A few months ago, I wrote a little agent-based simulation in NetLogo to test some of the ideas of the Dark Forest Hypothesis (DFH). Briefly, the DFH is one of the proposed answers to the Fermi paradox, which observes that while it is likely that there is life out there in the universe, we haven’t found any, and as far as we know, no other life has found us.
The DFH proposes that all life in the universe learns to be discrete, because the most likely outcome of encounters between life-forms is a deadly confrontation. Being visible means running the risk of getting destroyed by a more powerful opponent, so everyone is quiet and invisible, until of course, it’s too late. In a universe like that, by the time you become aware of a neighbour, it’s either too late for them, or too late for you.
The initial simulation (DarkForsestSim 1, or DFSim1) showed that if all agents out there (called civs in the simulation) automatically try to kill each other in an encounter, not surprisingly, few civs survive for long, and the visibility and ability of civs to see each other are both under significant selective pressure.
I got quite a few comments about the initial simulation from Reddit and other places. Lots of people were asking what would happen if civs had a variety of strategies out there, instead of all being simple killers. How would killers actually do in an environment in which other, more cooperative strategies are available? Would the DFH still hold up?
DFSim2: Killers, subjugators, and cooperators
DFSim2 tries to answer that question. For this round of experiments, I added two other types of civs: Subjugators, and cooperators. The basic rules of DFSim2 are still the same. Civs have a few attributes. They have a power rating, a visibility, and a detection threshold.
Figure 1: DFSim 2 in action. The solid dots are killers, the open dots are cooperators, and the concentric circles are subjugators.
Their power attribute determines how strong they are relative to other civs. Their visibility attribute determines how easy they are to see in the universe, and their detection threshold determines how well they can see other civs. In theory, a very powerful civ with low visibility and high detection threshold should be very well adapted to the system. As in DFSim1, these attributes mutate up and down for each civ during a run.
The new mechanics for DFSim2 involve encounters. When a killer civ encounters another, if it can see it, based on the other civs visibility and its own detection threshold, it still tries to kill it. If it succeeds, it grows by a percentage of the target civ’s power and size. In other words, killers absorb a certain portion of their prey.
When a subjugator civ encounters another, it tries to conquer it rather than to kill it. If it succeeds, i.e. if it can see it, and if it has higher power than its target, it then starts absorbing a percentage of the target civ’s power and size at each time step going forward. The target civ shrinks by the same proportion.
When a cooperator civ encounters another, if it can see it, it starts absorbing a percentage of its target’s size and power at each time step, but doesn’t cause the target to shrink. If two cooperators encounter each other, they both benefit from the relationship as long as it lasts.
It’s still possible for subjugators and cooperators to coexist with killers, as long as the killers can’t see them. So if a cooperator or a subjugator has low visibility, and they encounter a killer with a high detection threshold, they could theoretically benefit from their encounter with the killer.
How efficient do killers have to be for the DFH to be true?
The first question I asked for this set of experiments was how efficient killers have to be in order to create and environment in which the conclusion of the DFH still hold up. If subjugator and cooperator strategies can allow civs to exist alongside killers for the long term, then the DFH collapses, and we should be able to see other life around us.
I set the benefits of subjugators and cooperators at 2% for this set of experiments. It is fairly low because the benefits happen at each time step, whereas the benefits earned by killers are a one-time infusion at the time of the encounter.
I varied the benefit of killers (killGain) from 2% to 40%. In other words, in some runs of the simulation, the killers grew by as little as 2% of their target’s power and size, and in other runs, by as much as 40%. All runs lasted 200 000 time steps and started with 500 civs. Ten percent of the initial civs were killers, and the rest were evenly split between cooperators and subjugators.
Results
Figure 2: Results of 99 runs of 200 000 time steps each, showing run duration under various killGain values. Runs that ended early had only one civ left, and it was a killer.
Figure 2 shows kill gain (x-axis) vs total duration of the run (y-axis). A run of the simulation stopped if there was only one civ left. If a run went all the way to 200 000 time steps, that means there were many civs left in the universe. Essentially at that point, the killers have failed. In all cases in which the runs stopped early, there was one civ left, and it was a killer. That means the killers had eliminated everyone else.
The first thing to note from figure one is that there are short (>200k) runs with as low as a 6% killGain, but it doesn’t happen often before about 15% killGain. Above 20% killGain, most of the runs end early with just one killer civ remaining, and above 30% killGain, there are never any cooperators or subjugators left.
That answers the question pretty clearly. For a 2% benefit for cooperators and subjugators, the DFH is true if killers are 30% efficient, and in most universes, the DFH is true, and it is also true in most universes in which killers are at least 15% efficient.
Obviously, there is plenty of room for more experimentation here with other values, but this gives a sense at least that the DFH doesn’t always have to be true, and that it may in fact be true in a very limited range of circumstances.
Discussion
Historical contingency seems to play a pretty big role in this system. Even at fairly low killGain, the killers can make the DFH true at least sometimes, based on the relative starting positions of the civs. The order in which encounters take place is a huge determinant of the final outcome.
Overall conditions of the universe, such as killGain, and the benefits of subjugation and cooperation (and probably the mutation rate, that I didn’t touch here) determine what the odds are of a certain outcome, but the specific spatial arrangement of civs at start and their attributes heavily determine the actual outcome in any one universe.
So even if the DFH is very unlikely to be true in a universe like ours, it could still be true because of historical contingency. Conversely, even if we are in a universe in which it is very likely to be true because of the overall conditions, it could be false because of initial specific conditions.
There is fuel for nightmares, but also for SciFi stories in these results. If you want to play around with DFSim2 (or improve it), you can download it here. You’ll need to install NetLogo to use it and modify it.
Your model is quite interesting. I have written a draft paper, attempting to refute the Dark Forest Hypothesis from a probabilistic/logic point of view. You can find it there if you are interested : https://www.pgmusings.ca/journal/dfh
Intriguing.