ALGORITHMIC WARFARE HUMAN SYSTEMS
Simulating Civilian Behavior During Military Missions
Defense Dept. photo
LONDON — Conflicts are not constrained to military forces fighting each other. Civilian populations interact with the battlefield as well, and human behavior is a complex variable.
U.K.-based company Hadean is working on simulating how civilians react — both on the ground and online — to military operations to provide greater complexity and reality for training scenarios.
At the DSEI London trade show in September, the company debuted a new spatial computing system called the Hadean Platform for Defence. The platform includes an enhanced version of the company’s Pattern of Life simulation engine to “provide an out-of-the-box city-scale simulation that implements AI-powered civilian behaviors, traffic models and other reality factoring routines,” a Hadean press release stated.
During a demonstration of the platform at the show, Hadean officials displayed a scenario taking place in the southwest of England. The system was able to generate “full pattern of life across the entirety” of the simulated area, said Peter Taylor, a senior solutions architect for the company.
Using the Pattern of Life engine, “we’re actually simulating all of these entities,” Taylor said during the demo. “Each entity represents an individual pedestrian, or traffic, with their own objectives, their own sentiment. So, [their] objectives will be, ‘I need to go to the park, I need to go to work, I need to go to the shops.’” Each entity has “a certain level of anger, allegiance, fear, happiness.”
The system can generate a heat map of these civilian sentiments such as fear, tracking the displacement of the simulated population, he said. Before the adversary forces attack, “the fear level is very low, nothing really is going on.”
Once the adversary forces set off a series of improvised explosive devices, however, the system visualized how “the sentiment has changed” among the civilians, he said.
The civilian population became “very fearful,” running away from the area of the explosions. For military operators training in this scenario, “the next logical step there would be to set up an evacuation point, and in this case, it would be at a local stadium.”
As the evacuation began, the heat map showed “the dispersion of the fear sentiment,” Taylor said. “If we were to watch this over time, we would start to see all that fear in the heat map coalesce on a hotspot within the stadium, and then as everybody starts to feel safe within the stadium, you would see the sentiment over time start to normalize.”
The goal of Hadean’s platform is to take “a very simple … training exercise and [give] it context by putting it in a wider virtual world” that factors in civilian behavior, he said. “It has those various layers of social media and sentiment that will affect the outcome of a particular exercise or operation.”
The company is also using large language models to generate social media posts as if they are civilians observing a military operation, said Chris Arthurs, Hadean’s vice president of innovation.
“Analyzing the sentiment of these [posts] as they are issued, you can determine whether they’re positive or negative, and then based upon that, you can use that positivity or negativity to trigger the civilians that we simulate in the environment,” Arthurs said. If the civilians are reacting negatively to the operation, “you can flick switches in the logic of the civilians … to make them less likely to be helpful,” such as putting up roadblocks to hinder the soldiers. Alternatively, civilians could have a positive reaction to the operation and provide help, such as pointing out an adversary sniper.
“I’m really interested in using this for bridging the physical and the virtual worlds,” Arthurs said. Soldiers could conduct a live training exercise clearing a city street while fellow operators are watching a virtual overlay of the entire city at the same time.
Typically in a live training exercise, actors play local civilians, and the organizer of the exercise tells them whether to be helpful or cause problems for the soldiers, he said. However, if instead the organizer gave the actors playing civilians cell phones with a social media app that showed them the model-generated posts, “then they can automatically see what the sentiment looks like in the environment of their digital brethren, civilians in the virtual world,” and adjust how they interact with soldiers in the live training accordingly.
The soldiers in the live training environment and the operators in the virtual environment would then need to work together to de-escalate the situation and complete their mission, “much like you would in a real-world operation,” he said. “That is one of the many concepts that we have around large language models, and how we hope to use them and hope to deploy them.”
Arthurs said he also wants to explore the possibility of generating “foreign state disinformation” content to train soldiers on how to deal with nefarious actors online.
“To be clear, we’re not exactly sure yet … how that’s going to interface with the system,” he said. “But given that all of the rest of it works, why not throw that into the mix as well, and then it could … help train troops who [have] to deal with that.” ND