100 Years of Artificial Intelligence Explained

The transcript opens in September 2012 with Alex Krizhevsky training a neural network on two gaming GPUs as a side project, framing it as the seed of a technological revolution. The story then rewinds to WWII, where Alan Turing designed the electromechanical Bombe machine to crack the German Enigma code, which had over 100 quintillion possible settings. More than 200 Bombes eventually broke over 4,000 German messages per day, shortening the war by an estimated 2–4 years. Turing later published his famous 'imitation game' paper, proposing that a machine able to fool a human through text could be considered intelligent.

The narrative then covers the 1956 Dartmouth Conference, where John McCarthy — alongside Claude Shannon and others — formally named the field 'artificial intelligence.' The field quickly split into two competing schools: Marvin Minsky's symbolic approach (rule-based logic) and Frank Rosenblatt's neural network approach (brain-inspired adaptive learning). Rosenblatt built the Perceptron in 1958, which the U.S. Navy celebrated as the embryo of a conscious machine. However, Minsky and a colleague mathematically proved in 1969 that neural networks had hard learning limits, effectively killing that research direction for years. Two AI winters followed — one after government-funded research failed to deliver, and another in 1987 when the commercial expert systems boom collapsed as cheaper workstations made specialized Lisp machines obsolete.

The neural network revival began in 1986 when Geoffrey Hinton and colleagues published a paper on backpropagation, solving the problem of tracing errors through multi-layered networks. But the mathematical fix outpaced available hardware. It wasn't until Nvidia's powerful GPUs — built for gaming — became available that training deep networks became practical, cutting training time from months to days. The data problem was solved by Fei-Fei Li's ImageNet project, which assembled 14 million labeled photos by 2010 and launched an annual recognition competition.

In 2012, Krizhevsky's AlexNet entered that competition and achieved 15% error — 11 points better than the previous year's winner — without any hand-coded rules, simply by learning from the data itself. This result instantly made every other approach look obsolete and triggered a mass hiring of deep learning talent by Google, Facebook, and Microsoft. DeepMind's acquisition by Google in 2014 and AlphaGo's historic 2016 victory over 18-time Go world champion Lee Sedol — including a move no human would have conceived — demonstrated that AI could generate genuinely novel strategies.

The modern era of generative AI was launched by the 2017 Google paper 'Attention Is All You Need,' which introduced the transformer architecture. Unlike sequential word-by-word processing, transformers read entire passages in parallel, enabling far better context retention. OpenAI adapted the transformer to predict the next word at scale, releasing GPT-1 (2018), GPT-2 (2019), and GPT-3 (2020), before wrapping the technology into ChatGPT in late 2022. ChatGPT reached 1 million users in 5 days and 100 million in 2 months, triggering a $10 billion Microsoft investment and a Google internal 'code red.'

Anthropicpublicly launched Claude in March 2023, named after Claude Shannon. Google responded with Gemini in December 2023. The three companies diverged in strategy: OpenAI targeted general consumers, Google embedded AI across its ecosystem, and Anthropic focused on developers. Claude 3.5 Sonnet's Artifacts feature and the February 2025 release of Claude Code — a command-line tool capable of reading projects, editing files, and building software autonomously — made it the dominant tool for serious developers. By November 2025, Claude Code was generating over $1 billion annually. Amazon committed $25 billion and Google $40 billion to Anthropic in April 2026, underscoring how central Anthropic had become. The transcript closes by noting that while ChatGPT still leads the consumer market, Claude Code dominates the power-user and developer space, and the AI race continues to accelerate.