Artificial neural networks
Learning on a molecular scale
5 FACTS ABOUT
AI works in almost the same way as the human brain. Instead of thousands of brain cells, machine learning algorithms have thousands of nodes. Just like a neuron, each node receives incoming signals and has to decide whether to pass them on to the next node. To make this choice, it gives each signal a weight, which determines how important it is. A higher weight means a higher chance the signal will be passed on. To begin with all the weights are set at random, so the algorithm is essentially guessing what to do with each signal. To learn, it makes tiny changes to the weights, and then sees whether its guess is better or worse than before. This trial and error tunes the network, strengthening good connections and weakening bad ones – just like human learning.
Signals pass from one neuron to another across microscopic gaps called synapses. The first neuron releases small packets of chemicals called neurotransmitters, which cross the gap and hit the second neuron. Receptors on the second neuron detect the neurotransmitters, and if the signal is strong enough, they trigger a fresh electrical impulse. Learning increases the neurotransmitters the first neuron releases, and boosts the number of neurotransmitter receptors on the second neuron. These changes strengthen the connection between the two cells, making it easier for them to exchange signals in the future.