A bias is a learnable constant added to the weighted sum in a neuron, allowing the model to fit data that doesn't pass through the origin. Without bias terms, every neuron's output would be constrained to pass through zero when all inputs are zero, severely limiting the functions the network can represent. Each neuron typically has one bias value that is updated alongside its weights during backpropagation. Together, weights and biases form the complete set of learnable parameters in a standard neural network layer.