Researchers at North Carolina State University have developed a novel approach to enhancing artificial intelligence (AI) through meta-learning. They addressed the limitations of traditional neural networks, which have fixed structures, by allowing AI to autonomously adjust its neural network composition. This involved diversifying the activation functions of neurons and creating sub-networks with varying neuron types and connection strengths. The AI was given the ability to modify its neural network composition and choose between diverse or homogenous neurons. Interestingly, the AI consistently opted for diversity, which improved its performance. In standard numerical classification tasks, the diverse AI outperformed the homogeneous AI. The researchers plan to further optimize this approach by adjusting hyperparameters and applying it to a broader range of tasks, including regression and classification.