It all started with one question: how can we recreate the human brain? That question is still relevant today. Scientists and mathematicians have developed a system that brings us closer to the answer: neural networks.
Neurons, weights, connections, and many optimization methods were created to reach this incredible goal. Naturally, the idea was inspired by the brain. Our brains also contain neurons, connections, and much more that researchers have tried to reproduce. But the brain is not made of electrical impulses alone. Current systems mostly pass an impulse from neuron X to neuron Y, while biology also depends on chemical processes.
But why this long introduction?
As I wrote in AION First, we have a Concept Space containing concepts created by our AI. During development, a problem appeared: the current system could not see all the connections between concepts. That led to an idea: what if we changed a parameter-heavy semantic search into something closer to the behavior of the slime mold Physarum polycephalum, using it to find a path?
You may ask, “Why do you need path finding if you want an intelligent AI?” And you would be partly right — but not completely. I will explain why later :)
This was an experiment. Nobody knew exactly what would come out of it, but the tests confirmed that it was the best way to use Concept Space. I taught a virtual slime mold to work for me: I assigned scores to the correct concept and made the neural systems move toward it through the connections.
That helps solve the AI “black box” problem. I can now track the direction in which the AI is thinking and see how it reaches a specific idea. The process became far more transparent and convenient. Tests confirmed the theory: on one synthetic CS concept-selection test, the result improved from 1/3 to 3/3.