Part VI – Artificial Intelligence

Artificial intelligence from the 1956 Dartmouth workshop and McCarthy’s LISP through expert systems and the AI winters to AlexNet in 2012 and the transformer. It covers the researchers (Minsky, Hinton, LeCun, Bengio, Fei-Fei Li), the companies racing to build large language models, and the debates about bias, safety and superintelligence.

Founders & Early AI

• John McCarthy and LISP — the man who named AI, invented LISP, and pursued symbolic logic for fifty years
• Newell and Simon: The Thinking Machines — the Logic Theorist and the cognitive simulation hypothesis
• Edward Feigenbaum and the Expert System — DENDRAL with Lederberg, “in the knowledge is the power,” the Fifth Generation alarm, and the knowledge-engineering method that did not scale
• Marvin Minsky and the MIT AI Lab — co-founder of the field and simultaneously its greatest obstacle
• Joseph Weizenbaum and ELIZA — the 1966 chatbot that fooled its users, and the founder of computer ethics it horrified
• Chatbots Before ChatGPT — Abelson’s Goldwater machine, Racter, the Loebner Prize, Jabberwacky, A.L.I.C.E., Ask Jeeves, SmarterChild, XiaoIce, and Furby
• McCulloch and Pitts — the 1943 logical neuron, Hebb’s 1949 learning rule, the frog’s eye, and Pitts’s tragedy
• Frank Rosenblatt and the Perceptron — the 1957 Mark I Perceptron, the Minsky-Papert critique, and posthumous vindication
• Expert Systems and the First AI Winter — MYCIN, XCON, Japan’s Fifth Generation Project, and the collapse of knowledge engineering
• Douglas Lenat and Cyc — Eurisko’s war-game victories, forty years of hand-encoding common sense, and a last paper for the LLM era
• Judea Pearl and the Causal Revolution — Bayesian networks, the do-calculus, the ladder of causation, and the 2011 Turing Award
• Karen Spärck Jones and Information Retrieval — inverse document frequency, BM25’s foundations, evaluation as a discipline, and “computing is too important to be left to men”

The Deep Learning Era

• John Hopfield and Neural Networks — memory as energy minimization, Boltzmann machines, Nobel Prize in Physics 2024
• Geoffrey Hinton and Deep Learning — 30 years of unfashionable conviction, AlexNet 2012, Google, Nobel Prize, and a warning
• Yann LeCun and Convolutional Networks — LeNet, check-reading at Bell Labs, and the argument against LLMs
• Alex Krizhevsky — AlexNet trained on two gaming cards in a bedroom, the CIFAR datasets, and the man who started the AI boom and then left the field
• Yoshua Bengio and the Montreal School — word embeddings, attention mechanisms, MILA, and AI safety advocacy
• Schmidhuber, Hochreiter, and LSTM — the 1991 vanishing-gradient thesis, LSTM (1997), Siri and Google Translate, and the credit wars
• Andrew Ng and AI Education — Google Brain, Baidu, Coursera, and democratizing deep learning
• Fei-Fei Li and ImageNet — the 14-million-image dataset that launched the deep learning era
• ImageNet and the Deep Learning Revolution — ILSVRC, AlexNet’s 2012 result, and the competition that changed everything
• The Transformer Architecture — “Attention Is All You Need” (2017) and the architecture behind every major LLM
• Reinforcement Learning — TD-Gammon, DQN, AlphaGo Zero, and RLHF
• Barto and Sutton: Learning from Reward — an Air Force contract to test the “hedonistic neuron,” the 1983 pole-balancer, TD learning, the 1998 textbook, the Bitter Lesson, and the 2024 Turing Award

Modern AI & Frontier Systems

• Demis Hassabis and DeepMind — chess prodigy to AlphaFold; Nobel Prize in Chemistry 2024
• DeepMind and AlphaGo — Move 37, the 2016 Seoul match, and 200 million protein structures
• Sam Altman and OpenAI — ChatGPT’s 100 million users in two months, fired and rehired in four days
• Anthropic and Claude — the OpenAI walkout, Constitutional AI, MCP, Claude Code, and the safety lab that races
• Ilya Sutskever and the GPT Series — AlexNet, GPT-1 through GPT-4, and the board vote against his own CEO
• Andrej Karpathy and Modern AI — CS231n, OpenAI founding member, Tesla Autopilot, “Software 2.0,” Zero to Hero, and coining “vibe coding”
• AlphaFold — a 50-year biology problem solved, 200 million protein structures, and a Nobel Prize in Chemistry
• The LLM Race (2022–) — ChatGPT, Google’s Bard response, Meta’s Llama, and the GPU infrastructure race
• Mistral — Europe’s open-weight champion, Mixtral’s MoE, Le Chat, and AI sovereignty
• The Generative AI Revolution — GANs, diffusion models, DALL-E, and ChatGPT
• Stable Diffusion and AI Image Generation — latent diffusion from a Munich lab, open weights, and the copyright firestorm

AI Subfields

• Computer Vision — from the 1966 Summer Vision Project and SIFT to AlexNet’s 2012 earthquake and Vision Transformers
• Speech Recognition — Audrey, Hidden Markov Models, Jelinek’s “fire a linguist,” and Whisper
• The History of Speech Synthesis — Kempelen’s bellows, the Voder at the 1939 World’s Fair, a singing IBM mainframe, the Speak & Spell chip, Siri’s unwitting voice, WaveNet, and the €220,000 cloned-voice fraud
• Machine Translation — Georgetown-IBM 1954, the ALPAC winter, statistical MT, and the Transformer born from translation
• Lotfi Zadeh — Baku to Tehran to Berkeley, the z-transform and a department chairmanship before the 1965 heresy, twenty years of Kálmán, Kahan and a Golden Fleece nomination, and a state funeral in Baku
• Fuzzy Logic — Zadeh’s 1965 heresy, the Sendai subway, Japan’s 1990 washing-machine boom, and the hype that wore out
• Recommendation Systems — collaborative filtering, the Netflix Prize, and engagement optimization’s rabbit hole
• Robotics — Unimate, Shakey, Brooks’s subsumption, Roomba, Boston Dynamics, Mars rovers, and Moravec’s paradox
• Deep Blue and Computer Chess — Shannon’s 1950 paper, Kasparov 1997, the bug mistaken for genius, and AlphaZero
• Leonardo Torres Quevedo and El Ajedrecista — the 1912 king-and-rook chess automaton, the first machine to decide a game on its own, generations before Deep Blue

AI Themes

• The Rise of Artificial Intelligence — from Turing’s Imitation Game to deep learning and the LLM era
• The Natural Language Processing Revolution — from ELIZA to Word2Vec to Transformer and BERT
• The Voice Assistant Revolution — Siri, Amazon Echo, Google Assistant, and why voice stayed a niche
• Social Robots and AI Companions — Paro the seal, AIBO funerals at a Buddhist temple, Pepper’s 27,000 units, Replika’s overnight personality change, Character.AI’s lawsuits, and California’s SB 243
• The Knowledge Navigator — Apple’s 1987 concept film of a talking agent on a tablet, set in 2011, and the line through Newton and General Magic to Siri and Viv
• AI Ethics and Algorithmic Bias — Gender Shades, COMPAS, the Timnit Gebru firing, and “Stochastic Parrots”
• The Singularity and Superintelligence — Good’s “last invention” (1965), Vinge’s 2005–2030 window, Kurzweil’s 2045, Yudkowsky and Bostrom’s turn to risk, and the extinction statement of 2023
• AI in Clinical Practice — Leeds 1972, mammography CAD that improved nothing, the first autonomous diagnostic device, Epic’s sepsis score, and the radiologists Hinton said to stop training
• Ray Kurzweil — the reading machine for the blind and the K250 synthesizer, then the Law of Accelerating Returns, the 2045 Singularity, and the critics who call it a religion
• Jaan Tallinn — Kazaa and Skype’s Estonian engineer who read Yudkowsky in 2009, seeded DeepMind and Anthropic, and co-founded CSER and the Future of Life Institute to fund the safety of the thing he invests in
• Eliezer Yudkowsky — the self-taught founder of MIRI, the Sequences, LessWrong and a 660,000-word Harry Potter fan fiction, who asked in Time (2023) for a worldwide halt to large AI training runs
• Nick Bostrom — the simulation argument, the paperclip maximizer, Superintelligence (2014), and the Oxford Future of Humanity Institute that closed in 2024