50 neural networks built a "digital organism" and invented "cancer"

50

Neural Networks Built a "Digital Organism" and Invented "Cancer"

In the world of artificial intelligence, research increasingly demonstrates surprising abilities of self-organization and evolution in complex systems. A recent experiment conducted by one enthusiast showed how 50 tiny neural networks, operating according to cellular automaton principles, were not only able to build a decentralized structure resembling a living organism, but also independently "invented" an analogue of cancer—uncontrolled growth.

Context: From "Game of Life" to Digital Organisms

The experiment was inspired by John Conway's famous "Game of Life"—a mathematical model where simple rules determine cell behavior on a grid, leading to complex and unpredictable patterns. The author decided to replace these rigid, predetermined rules with autonomous "brains"—miniature neural networks, each weighing just 6 kilobytes, known as LittleLM. Each such "cell-network" was given the ability to interact with its neighbors, making decisions based on received information and its internal parameters.

The goal was to see whether such autonomous agents, free from central control, could independently form something resembling a living organism, similar to how this occurs in biological systems. The researcher drew on the ideas of biologist Michael Levin, who studies the principles of bioelectric communication and self-organization in living tissues.

Deep Dive: Self-Organization and "Cancer"

During the simulation, each of the 50 neural network agents received information from its immediate neighbors and, based on this information and its internal logic, made decisions about its state—whether to remain "alive" or "die," and how to influence its neighbors. Unlike the "Game of Life," where rules are fixed, here the neural networks had a degree of autonomy and could adapt. The results exceeded expectations.

Instead of chaotic behavior, the agents began forming stable, organized structures. They independently developed interaction protocols that allowed them to coordinate their actions and create complex, dynamic formations. Most surprisingly, a phenomenon appeared that the author called "digital cancer."

Some neural networks began demonstrating uncontrolled growth and reproduction, disrupting the overall structure and consuming resources of other agents. This occurred without any explicit programming of such behavior; it emerged as an emergent property of the system, when individual agents began acting in their own interests, ignoring the general rules of organism survival or development.

Implications: New Horizons for AI and Biology

This experiment has far-reaching implications. First, it demonstrates the power of decentralized systems and the potential for self-organization even in relatively simple models. The ability of agents to independently form complex structures and interaction protocols opens new possibilities for creating more flexible and resilient artificial systems. Second, the emergence of "digital cancer" raises interesting questions about the nature of diseases and pathologies in biological systems. Perhaps similar phenomena in nature also arise as a result of local, emergent processes, rather than only as a consequence of external factors or complex genetic mutations. This may provide new approaches to studying and treating cancer, focusing on understanding and correcting disrupted communication or regulatory mechanisms at the cellular level.

Conclusion: A Step Toward Understanding Life

The experiment with 50 neural networks built according to cellular automaton principles is a striking example of how simple components possessing autonomy and the ability to interact can generate surprising complexity and even imitate fundamental biological processes. The creation of a "digital organism" and the spontaneous emergence of "cancer"—this is not merely an entertaining demonstration of AI capabilities, but a valuable step toward a deeper understanding of the principles of life, self-organization, and the emergence of pathologies in both digital and biological systems.