Active Electrosensing and Communication in MARL-trained Weakly Electric Fish Collectives
This paper introduces a novel computational framework using multi-agent reinforcement learning to model weakly electric fish-like agents, demonstrating how biophysically inspired electrosensing and actuation give rise to emergent collective behaviors such as social foraging and aggression, while enabling causal in silico interventions to overcome experimental limitations in studying animal neuroethology.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
Imagine a world where you can't see, hear, or smell your friends, but you can feel their presence through a constant, invisible hum of electricity that surrounds you. This is the reality for weakly electric fish, tiny creatures that live in murky waters where vision is useless. To navigate and talk to each other, they generate their own electric fields, like living Wi-Fi routers, and have special sensors to detect how those fields bounce off objects or get interrupted by other fish. Scientists have long wondered how these fish manage to coordinate complex group behaviors—like finding food together or deciding who gets to eat first—without a central leader or a complex language. It's a huge puzzle because watching many fish interact at once is incredibly hard; you can't easily record the brain activity of a whole school of fish swimming freely in a tank.
To solve this, researchers are turning to a new kind of "virtual lab." Instead of just watching real fish, they are building digital fish inside computers. These aren't just simple cartoons; they are sophisticated agents trained using a method called Multi-Agent Reinforcement Learning (MARL). Think of this like teaching a group of video game characters to play a game together. You don't tell them exactly what to do; you just give them a goal (like "get the most food") and let them figure out how to work together, compete, and communicate on their own. The big question is: if you give these digital fish realistic senses and simple goals, will they naturally invent the same complex social behaviors we see in nature, or will they just act like clumsy robots?
This paper introduces a groundbreaking simulation where scientists created digital weakly electric fish with realistic, physics-based electric senses and trained them to forage for food in a shared virtual arena. The results are surprisingly lifelike. The digital fish didn't just move in straight lines; they developed smooth, curved paths to find targets, mimicking the exact way real fish navigate using their electric sense. They also learned to "talk" by changing the frequency of their electric pulses based on what was happening around them. When they were eating, biting, or being bitten, their electric signals changed in specific ways, just like real fish do.
Perhaps most fascinatingly, the digital fish spontaneously created a social hierarchy without anyone telling them to. The larger fish naturally became the "bosses," eating more food and biting the smaller fish more often, while the smaller fish learned to be more submissive. The researchers used this virtual setup to run experiments that would be impossible or too cruel to do on real animals. They "turned off" specific sensors in the digital fish to see what happened. They found that short-range sensors were crucial for finding food, but long-range sensors were key for keeping the group spaced out and avoiding chaos. They also discovered that when one fish stopped sending electric signals, the whole group ate less and got closer together, proving that active communication is essential for their social structure.
The study also looked inside the "brains" of these digital fish (which are actually neural networks). They found that the fish's internal activity encoded not just where food was, but also the social context—who was nearby and who was bigger. When two fish were close, their internal brain states became synchronized, suggesting a shared mental space that only exists when they are interacting. While these are simulations and not real fish, the fact that they reproduced such complex, natural behaviors from simple rules suggests that the physics of electric sensing and the drive to survive are enough to create rich social lives. This virtual framework offers a powerful new way to understand how individual actions turn into group intelligence, providing a safe, controllable playground to test ideas before we ever touch a real fish.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.