Typically, algorithms learn to act in predetermined conditions — like water flowing through a ready-made channel. The new method turns this logic on its head: it allows the system to simultaneously master actions and change the environment itself. Like a stream that doesn't just flow, but carves a path through soft soil, making the route shorter over time.
The key is splitting the environment's tuning into two phases: one, like a momentary current, affects the current step, while the other, like erosion, shapes future events. This avoids chaos and precisely calculates how each change affects the outcome. The algorithm doesn't need a ready-made map — it learns from its own trial and error.
The effectiveness was tested on a swarm of drones: they simultaneously transmit signals and adjust their positions, much like water that naturally finds the shortest path downhill. Surprisingly, trying to optimize everything at once can worsen the result — the system is forced to balance between immediate gain and long-term improvements. This approach will make technology truly adaptive.
Ants don't just walk along trails — they alter the environment with pheromones so their kin can find better routes.
Like in the movie 'Inception,' where the architect alters the dream space, the new algorithm lets a system flexibly reshape its environment to reach a goal.