Founding the Field

Claude Shannon

Engineer who founded information theory, giving the field the mathematical vocabulary — bits, entropy, channel capacity — that loss functions and compression are still built on.

Claude Shannon (1916–2001) is best known for a 1948 paper, "A Mathematical Theory of Communication," that founded information theory from scratch. It answered a deceptively simple question: how much can a message be compressed, and how reliably can it be sent over a noisy channel? His answer introduced the bit as the fundamental unit of information and defined entropy as a precise measure of uncertainty — concepts borrowed directly from thermodynamics and repurposed for communication.

That vocabulary quietly underlies modern machine learning. Cross-entropy, the loss function used to train almost every classifier and every LLM's next-token prediction, is Shannon's entropy applied to a model's predicted probability distribution. Tokenization schemes, model compression, and even the intuition that a model "learns" by reducing uncertainty about its data all trace back to concepts Shannon formalized before digital computers were common.

Shannon also had a hand in AI's founding directly — he co-organized the 1956 Dartmouth workshop alongside John McCarthy, Marvin Minsky, and Nathaniel Rochester, the meeting that coined the term "artificial intelligence."

See also: Alan Turing, John McCarthy

Learn more: ML Fundamentals · Wikipedia: Claude Shannon