Replicating natural learning through emergence
Many of the natural systems that drive the rhythm of our daily lives are comprised of large, diverse groups of smaller entities. From cells and their billions of molecules, to brains and their billions of neurons, to economies and their billions of people, each complex whole arises from countless interactions between the autonomous actors within it.
At Cambrya, we are inspired to replicate the interactive behaviors that produce the spectacular cognitive processes that occur in all living creatures.
We know from personal experience that our brains never stop learning and thinking, and we know from practical experience that it is not fully clear how the trillions of interactions between neurons make that possible. Observations by scientists studying neurobiology, psychology, and sociology detail a dizzying array of small interactions that contribute to the coherent outcome of learning and thinking.
“Complex systems and their behaviors emerge naturally and dynamically as a byproduct of many autonomous actors behaving collectively.”
But how do we establish causation, correlation, or coincidence when there are so many variables at play and when so much remains unknown? We believe the answer lies in a new mathematical approach founded on a concept called emergence.
The theory of emergence posits that modeling simple interactions between small entities within a system can allow us to understand and experience the stunning complexity of the system itself. Like crystals spontaneously forming, ants architecting a city, bees building a hive, or flocks of birds and schools of fish moving together, no individual actor creates the complexity of the whole. Instead, complex systems and their behaviors emerge naturally and dynamically as a byproduct of many autonomous actors behaving collectively.
Density Networks seek to replicate learning behaviors following a similar approach. Within Density Networks, autonomous structures continuously interact with each other and update themselves based on sensory information encountered in the environment. These actors and structures each act simply and independently, but when we zoom out we see that they collectively form a complex system that identifies, learns, and eventually predicts the patterns it encounters in its world.
The effectiveness of this approach is neatly illustrated by considering how to simulate the boundless undulating shapes produced by schools of fish.
To mirror these complex 3-dimensional patterns, it might seem sensible to use equations for describing, well, complex 3-dimensional patterns. Perhaps the intricate mix of calculus, geometry, and linear algebra used to bring 3-dimensional characters and scenes to life in virtual worlds? Yes, all of this math may produce a recognizable simulation, but it falls short in producing the breathtaking fluidity, and particularly the diversity, of what we observe in nature.
Emergence offers a different approach. From this perspective, the school’s complex movements are behaviors that emerge from the collective interactions of individual fish. After some trial and error defining each fish’s behavior, we press play on a simulation where each fish has two simple instructions:
1. Move closer to the fish around you
2. But not so close that you might bump into them
The school expands and contracts, bobs and weaves, swirls and dips, fluidly, dynamically, seemingly unpredictably. We’ve done it. These patterns are the collective result of each individual fish continuously adjusting itself to get close–but not too close–to its neighbors.
By modeling entities and interactions that look nothing like the system and behaviors we are ultimately interested in, we achieve a more elegant and accurate model.
Classical machine learning algorithms and large language models perform multiple operations on vast sets of input data to produce simulations of known outputs. Density Networks will dynamically interact with the environment, producing spontaneous outcomes that feel familiar and natural. As a result, they can preserve and interpret context, adjust rapidly to new stimuli, and learn on the fly.
Our school of fish has a lot of room to grow. But the patterns it produces are already breathtaking. Reach out to see the power of emergence applied to learning for yourself.
