Artificial intelligence is software that learns from examples to make choices on its own. Here, a neural network teaches itself to play Connect 4 by training against a simulated opponent over many rounds. A genetic algorithm drives the process — it creates many versions of the program and keeps the ones that perform best, gradually sharpening its choices. When a human player with a different strategy steps in, though, the human still wins. The network learned to beat one style of play rather than the game itself, which shows that AI decisions depend heavily on the range of examples it trains against.

A neural network is a program that learns through practice — studying many examples until it can spot patterns on its own. In this experiment, you build one and train it to play Connect 4 using a genetic algorithm: the method creates many versions of the program and keeps the ones that perform best. After many rounds against a simulated opponent, the network learns to beat that opponent reliably. When a human player with a different strategy steps in, though, the human still wins. That result reveals something important: the network learned to recognize one style of play rather than the game itself. Training it against several opponents with different strategies could help it play more broadly.

A genetic algorithm creates many versions of a program and keeps the ones that perform best — mimicking natural selection inside a computer. Here, that process trains a neural network to play Connect 4. The algorithm builds many network players, tests them against a simulated opponent, and selects the strongest performers to produce the next generation. As rounds continue, the network learns to beat that opponent reliably. When a human steps in with a different strategy, though, the human still wins — because the network learned to beat one style of play, not the game itself.