A neural network is a computing model made of many simple units, loosely inspired by brain neurons, arranged in connected layers. Each unit takes inputs, weights them, and passes on a signal. The foundational idea dates to Warren McCulloch and Walter Pitts, whose 1943 paper “A logical calculus of the ideas immanent in nervous activity” in the Bulletin of Mathematical Biophysics modeled neurons as simple logical threshold units.
Modern neural networks stack many such layers and learn by adjusting the connection weights to reduce errors on training data. In their 2015 Nature review “Deep learning,” LeCun, Bengio, and Hinton describe how networks with multiple layers automatically learn useful representations of raw data, driving advances in image recognition, speech, and language.
The power of a neural network comes from depth and training. With enough layers and data, the network learns increasingly abstract features, for example moving from edges to shapes to objects in an image, without a human specifying those features.
Why business readers should care: Neural networks underpin most of today’s AI, from voice assistants to product recommendations to the large language models behind chat assistants. Knowing they are trainable, layered pattern-finders, rather than hand-coded logic, clarifies both their flexibility and their dependence on large datasets.