Mechanical Neural Network: Making AI Comprehensible for Everyone

Artificial Intelligence (AI) will become more and more an essential part of our everyday technology, but the function of AI is not yet covered in school education. While STEM education often focuses on the mathematical backbone of AI, i.e., probability calculation, students profit from hands-on activities to understand the function of artificial neural networks (ANNs). Making hands-on activities on AI feasible, this paper introduced the Mechanical Neural Network (MNN), a physical implementation of a multilayer perceptron (MLP) with Rectified Linear Units (ReLUs) as activation functions. With two input neurons, four hidden neurons, and two output neurons, the MNN allows students to experience the effect of changing the parameters of the network on the output. Shifting AI from soft to hardware, each component of an MLP is transferred into the real world. Thus, neurons become small wooden levers, connections between the neurons become strings, and summations become guide rollers of a pulley system. In a game based-learning manner, students can adapt the parts of the MNN for, e.g., modeling logical operators like exclusive or (XOR) and starting to understand the function of an ANN.