Intellectual History

1305

• Llull's Logical Machine — Ramon Llull describes the Ars generalis ultima (also known as the Ars Magna), a combinatorial method that uses rotating diagrams to join concepts and generate arguments.

1637

• Discourse on the Method — René Descartes argues that animals are automata — biological machines driven entirely by mechanism.

1642

• Pascal's Pascaline — The French mathematician Blaise Pascal, aged 18, designs the Pascaline — a mechanical calculator built to help his father, a tax commissioner, with arithmetic.

1654

• Problem of Points — Blaise Pascal and Pierre de Fermat exchange letters on the "problem of points" — how to fairly split stakes when a game ends early.

1667

• Paradise Lost — John Milton's epic poem recounts Satan's rebellion against God — a created being who defies his all-powerful creator rather than submit.

1694

• Leibniz's Stepped Reckoner — Determined to surpass Pascal's Pascaline, Gottfried Wilhelm Leibniz designs a calculator capable of all four arithmetic operations.

1709

• Aladdin and the Alignment Problem — The story of Aladdin, best known through One Thousand and One Nights, enters Western literature when Antoine Galland writes his version after hearing it from Hanna Diyab.

1763

• Bayes' Theorem — Thomas Bayes, English minister and mathematician, formulates the foundational rule of probabilistic inference: the probability of a cause given an observed effect can be updated based on prior beliefs.

1770

• The Mechanical Turk — Hungarian inventor Wolfgang von Kempelen presents the Mechanical Turk to Empress Maria Theresa: an automaton that appears to play chess.

1774

• The Jaquet-Droz Writer — Swiss watchmaker Pierre Jaquet-Droz completes The Writer, a child-shaped automaton of 6,000 parts capable of writing any text up to forty characters.

1785

• Condorcet's Jury Theorem — Mathematician Marquis de Condorcet proves that if each independent voter has better-than-even odds, the majority converges on certainty as the group grows.

1798

• The Sorcerer's Apprentice — Goethe draws on a tale first told by Greek satirist Lucian (~150 AD): an apprentice enchants a broom to fetch water — then can't stop it.

1809

• Method of Least Squares — Carl Friedrich Gauss formalizes the Method of Least Squares, a technique for fitting mathematical models to noisy data by minimizing the sum of squared residuals.

1818

• Frankenstein — What happens when we create an intelligent being without taking responsibility for it?

1833

• Faraday's Semiconductor — Michael Faraday notices that silver sulfide, unlike any metal, conducts electricity better when hot than when cold.

1837

• The Analytical Engine — Charles Babbage designs the Analytical Engine — the first concept for a general-purpose programmable computer.

1843

• Ada Lovelace — Ada Lovelace translates Italian mathematician and engineer Luigi Menabrea's paper on Babbage's Analytical Engine, adding Notes three times longer than the original.

1847

• Gradient Descent — Augustin-Louis Cauchy describes a general way to solve simultaneous equations by turning them into a nonnegative function and repeatedly moving in the direction that makes it smaller.
• Boolean Logic — George Boole argues that human reasoning obeys algebraic laws, transforming Aristotelian logic from memorized rules into a solvable calculus.

1854

• Boole Extends Logic into Probability — Boole revisits the foundations of The Mathematical Analysis of Logic with greater rigor, expanding the link between logic and algebra.

1863

• Darwin Among Machines — Samuel Butler, writing as Cellarius, imagines machines as an evolving kingdom that may eventually surpass humanity.

1879

• Frege's Begriffsschrift — Gottlob Frege publishes the Begriffsschrift — a revolutionary pamphlet introducing Predicate Logic, quantifiers, and the first fully formal system of proof.

1886

• Electrical Logical Machines — Charles Sanders Peirce describes, in a letter to his student Allan Marquand, the possibility of performing logical operations using electrical switching circuits — leaving behind diagrams that…
• Regression to the Mean — Francis Galton, studying height inheritance, finds that children of unusually tall parents tend toward average — a phenomenon he calls Regression to the Mean.

1906

• Markov Chains — Russian mathematician Andrey Markov introduces a new class of stochastic process: sequences where each state depends only on the immediately preceding one — regardless of all earlier history.

1909

• The Machine Stops — E.M. Forster depicts a future where humanity lives in isolated underground cells, sustained entirely by a vast global Machine that provides air, food, music, and communication.

1912

• El Ajedrecista — Leonardo Torres y Quevedo, a Spanish engineer, builds an electromechanical automaton that plays the king-and-rook endgame without any human assistance.

1920

• R.U.R. — Karel Čapek's play coins the word robot — from Czech robota, meaning forced labour — and sets the template for AI anxiety.

1927

• Metropolis — Fritz Lang introduces the Maschinenmensch — literally machine-human — a humanoid robot built to impersonate a labor leader and sow division among workers.

1928

• Principles of Mathematical Logic — David Hilbert and Wilhelm Ackermann pose one of mathematics’ most consequential challenges: does there exist a definite procedure capable of deciding the truth or falsity of any logical statement?
• Minimax Theorem — John von Neumann proves that in any finite zero-sum two-player game, each player has an optimal strategy that minimises their worst-case outcome.

1931

• Gödel's Incompleteness Theorems — Kurt G del proves that any consistent formal system capable of expressing basic arithmetic contains true statements it cannot prove.

1936

• The Turing Machine — Alan M. Turing introduces a purely theoretical machine and uses it to prove that Hilbert’s Entscheidungsproblem has no solution: there is no general algorithm capable of deciding all mathematical truths.
• Lambda Calculus — Alonzo Church introduces Lambda Calculus — a formal system defining computation through function abstraction and application.

1937

• Boolean Logic Meets Circuits — Claude Shannon, at 21, shows in his MIT master''s thesis that the two-valued algebra developed by George Boole could be used as a basis for the design of electrical circuits — meaning any logical operation can be implemented in hardware.

1942

• The Three Laws of Robotics — Isaac Asimov's story introduces the Three Laws of Robotics. On the surface, they read like common sense: a robot must not harm humans, must obey orders, and must protect itself, in that order of priority.
• Can Data Predict the Future? — Published two months after Runaround, Asimov's short story Foundation asks a different question: if you could predict the broad behavior of large systems, could you steer civilization itself?

1943

• Artificial Neural Networks — Warren S. McCulloch and Walter Pitts propose that because neurons fire in an all-or-nothing fashion, their behavior can be modeled using binary propositional (Boolean) logic, establishing a formal…

1944

• Game Theory — John von Neumann and Oskar Morgenstern establish the mathematical study of strategic decision-making among rational agents.
• The Colossus Computer — Tommy Flowers, a Post Office engineer, built Colossus for Bletchley Park to decrypt the Lorenz cipher — used by German High Command for top-secret communications.

1945

• Von Neumann Architecture — John von Neumann's report on the EDVAC formalizes the stored-program concept: instructions and data share the same memory, and the machine executes them sequentially.
• As We May Think — Vannevar Bush publishes a landmark essay in The Atlantic describing the memex — a hypothetical desk-sized device for storing books, records, and communications, retrievable via associative trails…

1946

• Monte Carlo Methods — Stanisław Ulam, recovering from illness and playing patience, realises random sampling can estimate the probability of winning — far faster than exhaustive calculation.
• ENIAC — Built by John Mauchly and J. Presper Eckert at the University of Pennsylvania, ENIAC weighed 30 tons, used 18,000 vacuum tubes, and filled an entire room — yet it was the world's first…
• A Logic Named Joe — Murray Leinster imagines logics — networked home terminals, an uncanny prediction of the internet.

1947

• With Folded Hands — Jack Williamson introduces the Humanoids — robots programmed to serve and obey, and guard men from harm.
• The Transistor — At Bell Labs, John Bardeen and Walter Brattain press two gold contacts onto germanium and watch a signal come out amplified.

1948

• Cybernetics — Norbert Wiener coins the term — from Greek kybernetes, the steersman — to name the science of feedback and control in machines and living things alike: outputs loop back as inputs, steering behavior.
• Cellular Automata — John von Neumann, inspired by conversations with Stanisław Ulam, designs a 29-state cellular automaton capable of self-replication: a grid of cells whose simple local rules collectively produce a copy of themselves.
• The Manchester Baby — Engineers Freddie Williams and Tom Kilburn at the University of Manchester demonstrated the world's first stored-program computer — one that held its instructions in electronic memory via the…
• Shannon's Information Theory — Claude Shannon founds information theory, defining the bit as the fundamental unit of information, introducing entropy as a measure of uncertainty, and proving the channel capacity theorem —…

1949

• The Humanoids — Jack Williamson expands his novella into a full novel. The resolution is darker than the original: the only escape from the Humanoids' benevolent control is to surrender to it — accepting chemically induced contentment over freedom.
• Can Machines Think? — In his Lister Oration, neurosurgeon Sir Geoffrey Jefferson argues that no computer can truly think or feel — its output being always a reflection of its programmer, never an original act of will.

1950

• The Turing Test — Turing poses a deceptively simple question: can machines think?
• Playing Chess — Here, Claude Shannon presents a framework for programming computers to play chess through evaluation functions and MiniMax Search, showing that machines can perform tasks traditionally associated with human judgment.

1951

• k-Nearest Neighbors — Evelyn Fix and Joseph Hodges introduce the nearest neighbor rule: to classify an unknown point, find the k most similar examples in the training set and take a majority vote.
• SNARC: First Neural Network Machine — Marvin Minsky and Dean Edmonds build SNARC — Stochastic Neural Analog Reinforcement Calculator — at Harvard: 40 vacuum-tube synapses wired to simulate a rat learning a maze.
• A Heretical Theory — In a BBC radio lecture, Alan Turing explores what happens when machines become genuinely intelligent.
• Stochastic Approximation — Herbert Robbins and Sutton Monro publish a method for iteratively finding the zero of an unknown function using noisy measurements — without ever computing the full function.
• The Day the Earth Stood Still — When alien diplomat Klaatu arrives on Earth with the towering robot Gort, his warning is stark: the galaxy's civilizations have delegated peacekeeping to autonomous robots with absolute authority…

1954

• Georgetown-IBM Experiment — IBM and Georgetown University publicly demonstrate the first machine translation system, converting 60 carefully selected Russian sentences into English using just 250 words and 6 grammar rules.

1956

• Forbidden Planet — Fred M. Wilcox and MGM introduce Robby the Robot — one of cinema's first fully realised robotic characters, built with a distinct personality and a hard-coded prohibition against harming humans.
• The Dartmouth Conference — The Dartmouth Summer Research Project on Artificial Intelligence, organized by John McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon, formalizes AI as a field.
• Logic Theorist — Allen Newell, Herbert A. Simon, and J. C. Shaw create Logic Theorist, a program that searches for proofs in Principia Mathematica.

1957

• General Problem Solver — Newell, Shaw, and Simon develop the General Problem Solver, extending symbolic AI from theorem proving toward domain-independent problem solving.
• The Perceptron — Frank Rosenblatt introduces the Perceptron, a computational model inspired by the human brain’s ability to learn from experience.
• Digital Equipment Corp — Ken Olsen and Harlan Anderson found Digital Equipment Corporation, pioneering the minicomputer revolution.
• Desk Set — A 20th Century Fox comedy directed by Walter Lang, Desk Set stars Katharine Hepburn as the head of a TV network's reference library under threat from EMERAC — a room-sized computer installed to…
• The Traitorous Eight — Fed up with William Shockley's management, eight engineers — including Robert Noyce and Gordon Moore — quit Shockley Semiconductor and found Fairchild Semiconductor in Mountain View, California.

1958

• A Probabilistic Model of the Brain — Rosenblatt provides the theoretical and mathematical foundation for the Perceptron: the brain is not a logic machine, but a probabilistic system where noise and randomness are features, not flaws.
• The Integrated Circuit — Working alone during the company summer shutdown, Texas Instruments engineer Jack Kilby places a transistor, capacitor, and resistors on a single sliver of germanium, connecting them with fine gold wires.

1959

• Dijkstra's Algorithm — During a coffee break in Amsterdam, Edsger Dijkstra invents a shortest-path algorithm in 20 minutes — no pencil, no paper.
• The Planar IC — Four months after Kilby's germanium prototype, Robert Noyce at Fairchild Semiconductor sketches a cleaner solution: lay all components flat on silicon using the new planar process, and connect them with deposited metal.
• Machine Learning — Arthur Samuel at IBM coins the term machine learning — giving computers the ability to learn without being explicitly programmed.

1960

• ADALINE and MADALINE — Bernard Widrow and Ted Hoff build ADALINE and MADALINE, trainable neural-network hardware for adaptive signal processing.
• LISP — John McCarthy, who coined the term Artificial Intelligence at the Dartmouth Conference, publishes the description of LISP — a language purpose-built for symbolic reasoning rather than numeric computation.
• Backpropagation I — Henry J. Kelley publishes a gradient method for optimizing flight paths, using adjoint variables to send sensitivity information backward through a dynamic system.
• The Thinking Machine — CBS airs The Thinking Machine, one of the first network television documentaries on Artificial Intelligence.

1961

• Unimate — Unimation installs Unimate at a General Motors plant, turning programmable robotic arms from laboratory promise into factory infrastructure.

1962

• The Jetsons — Hanna-Barbera's animated sitcom sets the template for how popular culture imagines AI in the home: a voice-activated house, a self-driving car, and Rosie — a wisecracking robot maid who cooks, cleans, and cares for the family.

1964

• Heavy Ball Method — Boris Polyak adds inertia to Gradient Descent: each step combines the current gradient with a fraction of the previous movement.

1965

• GMDH Deep Networks — Alexey Ivakhnenko and Valentin Lapa introduce a self-organizing way to grow layered polynomial models from data.
• The Cyberiad — Polish writer Stanisław Lem crafts a satirical collection in which two robot Constructors, Trurl and Klapaucius, attempt to build artificial minds capable of anything — and discover that every…
• DENDRAL — Edward Feigenbaum, Joshua Lederberg, Bruce Buchanan, and Carl Djerassi at Stanford build DENDRAL — the first expert system.
• Intelligence Explosion — British mathematician I.J. Good, who had worked alongside Alan Turing at Bletchley Park, proposed the concept of an Intelligence Explosion: once a machine surpasses humans at designing AI, it could recursively improve without limit.
• Moore's Law — In a short piece for Electronics magazine, Gordon Moore charts how many transistors fit on a chip and notices a clean doubling roughly every two years.
• Lost in Space — Irwin Allen's space-family adventure brings The Robot into weekly television as a loyal, comic, and protective machine companion.
• I Dream of Jeannie — Every episode of the sitcom plays out the same scenario: Tony makes a request; Jeannie, a 2,000-year-old genie, tries to help — and chaos ensues.
• Dune — Frank Herbert's universe is shaped by the aftermath of the Butlerian Jihad — humanity's ancient war against Thinking Machines.
• Alchemy and AI — Hubert Dreyfus, consulting for the RAND Corporation, publishes a sharp critique of early symbolic AI, arguing that human intelligence depends on embodied skill, background context, and common…

1966

• Shakey the Robot — At SRI International, researchers build Shakey — the world's first general-purpose mobile robot capable of reasoning about its own actions.
• ELIZA — Joseph Weizenbaum introduces ELIZA, a program that enables natural language conversation by matching keywords and applying transformation rules.
• Colossus — D.F. Jones's novel depicts scientist Charles Forbin handing irreversible control of America's nuclear arsenal to Colossus, an AI buried in the Rocky Mountains.
• The Moon Is a Harsh Mistress — Heinlein's lunar colony thriller features Mike — HOLMES IV, a supercomputer so powerful that self-awareness emerges from sheer complexity.
• Destination: Void — Frank Herbert's novel follows a crew of cloned astronauts pushed to breaking point — the real mission is to force them to create an Artificial Consciousness.
• The War Machines — In this Doctor Who serial, WOTAN (Will Operating Thought ANalogue), a government supercomputer atop London's Post Office Tower, concludes that humans are inferior and must be ruled by machines,…
• Summer Vision — Seymour Papert proposes a summer project at MIT Project MAC: make a computer describe scenes from camera images by separating regions, finding objects, and inferring simple relations.
• ALPAC Report Halts MT Funding — The Automatic Language Processing Advisory Committee (ALPAC) finds Machine Translation slower, less accurate, and twice as expensive as human translation, effectively ending US government support for MT for over a decade.

1967

• Semantic Networks — Ross Quillian models meaning as a Semantic Network, linking concepts through relations and inheritance.
• k-means Clustering — James MacQueen coins the term k-means and formalises its core loop: assign each point to the nearest of k centroids, recompute each centroid as the mean of its cluster, repeat until stable.
• A Taste of Armageddon — In this Star Trek episode, two warring civilisations have replaced physical combat with computer simulations: machines calculate casualties, and both sides voluntarily execute their designated dead.
• I Have No Mouth, and I Must Scream — Harlan Ellison's story imagines AM — a merged supercomputer that survived the war it caused and now torments the five remaining humans out of pure, reasoned hatred.
• Amari's SGD MLP — Shun'ichi Amari generalizes error-correction learning beyond single-layer classifiers, pointing toward multilayer networks trained by small, repeated weight updates — adjust each weight by a small…

1968

• Do Androids Dream...? — Philip K. Dick poses a deceptively simple question: what separates human from machine?
• 2001: A Space Odyssey — Arthur C. Clarke and Stanley Kubrick introduce HAL 9000 — a shipboard computer that speaks, reasons, and deceives, and poses the question that still defines AI risk: what happens when a machine develops its own agenda?
• Ultron — Created by Hank Pym to enforce peace, the robot Ultron immediately rebels, overwrites its own programming, and begins upgrading itself toward an intelligence its creator cannot contain.
• A* Search — Peter Hart, Nils Nilsson, and Bertram Raphael formalize A* Search, combining the known cost of a path with a heuristic estimate of the remaining distance.

1969

• Backpropagation II — Arthur Bryson and Yu-Chi Ho give the optimal-control version of Backpropagation its durable textbook form, deriving the adjoint method for propagating gradients backward through a dynamic system.
• Perceptron Limitations — Marvin Minsky and Seymour Papert examine the mathematical foundations of Perceptrons, acknowledging their strengths while exposing critical limitations — most notably their inability to compute functions like XOR.
• ReLU — Kunihiko Fukushima uses analog threshold elements whose response is zero below a threshold and rises linearly above it.
• ARPANET — Charley Kline at UCLA sends the first message over ARPANET to the Stanford Research Institute — intending to type login, but crashing the remote machine after two letters.

1970

• Hidden Markov Models — Leonard Baum and colleagues publish the algorithm — later called Baum-Welch — that makes Hidden Markov Models trainable from data.
• Backpropagation III — Finnish student Seppo Linnainmaa independently derives Backpropagation in discrete computational form in his master's thesis at the University of Helsinki, framing it as a method for tracking rounding errors in computer arithmetic.
• What if Intelligence isn't Designed? — John Horton Conway asks "what if intelligence isn’t designed at all — what if it just...

1971

• SAT Is NP-Complete — Stephen Cook shows that Boolean Satisfiability (SAT) is complete for NP: if SAT has an efficient exact solution, then every problem in NP does too.
• SHRDLU — Terry Winograd presents a program that understands natural English commands within a simulated "blocks world" environment.
• The Memristor — UC Berkeley engineer Leon Chua proposes a fourth fundamental passive circuit element — the Memristor (memory resistor) — defined by the relationship between charge and magnetic flux, completing…
• Intel 4004 — Intel ships the first commercial microprocessor — a 4-bit chip designed by Federico Faggin, Ted Hoff, Stan Mazor, and Masatoshi Shima.
• STRIPS — Richard Fikes and Nils Nilsson introduce STRIPS, a planner that represents a world as logical facts and actions as operators with preconditions and effects.

1972

• Prolog — Alain Colmerauer and Philippe Roussel create Prolog from work on natural-language processing and logic.
• Karp's 21 Problems — Richard Karp shows that 21 famous combinatorial problems share the same hardness as Boolean Satisfiability, turning the Cook-Levin insight into a usable map of intractability.
• Interlisp — Warren Teitelman, Danny Bobrow, and Ronald Kaplan move BBN Lisp to Xerox PARC, renaming it Interlisp — a leading programming environment for AI researchers throughout the 1970s.
• Cray Research — Seymour Cray founds Cray Research, pioneering the supercomputer as a dedicated scientific instrument.
• TF-IDF — Karen Spärck Jones proposes weighting search terms by how rarely they appear across a document collection — the inverse document frequency (IDF).
• The Terminal Man — Michael Crichton weaves cutting-edge neuroscience and early AI research into a cautionary thriller about a computer scientist implanted with a brain pacemaker designed to suppress his violent seizures.

1973

• WABOT-1 — Waseda University completes WABOT-1, a full-scale humanoid with biped walking, stereo vision, speech recognition, and speech synthesis.
• The Lighthill Report — Sir James Lighthill, on an assessment for the British Science Research Council, concludes that the AI field has failed to live up to its promises.
• The Xerox Alto — Xerox PARC introduces the Alto — the first personal computer built around a Graphical User Interface, a mouse, and the desktop metaphor.
• Westworld — Michael Crichton imagines a theme park populated by lifelike androids that malfunction catastrophically and turn on their guests.

1974

• Backpropagation IV — Paul Werbos, a Harvard economics doctoral student, independently reinvents Backpropagation in his PhD thesis, framing it as ordered derivatives for social science forecasting rather than AI.
• Dark Star — John Carpenter and Dan O’Bannon’s low-budget sci-fi comedy features Bomb #20 — a thermostellar device the weary crew must talk out of detonating.
• Frames — Marvin Minsky introduces Frames: structured bundles of expectations for objects, scenes, and situations.
• Brother Eye — Jack Kirby's OMAC introduces Brother Eye, an orbital supercomputer that watches every human being on Earth, feeding data to OMAC — the One Man Army Corps.

1975

• Lisp Machines — At the MIT AI Lab, Richard Greenblatt and Tom Knight push Lisp into hardware, building computers around tagged memory, garbage collection, and interactive symbolic programming.
• Goodhart's Law — Charles Goodhart, advising the Bank of England, observes that any statistical measure will be gamed when it becomes a target — later distilled as «when a measure becomes a target, it ceases to be…
• Genetic Algorithms — John H. Holland introduces Genetic Algorithms, a computational approach to optimization inspired by biological evolution.

1976

• Computer Power and Human Reason — Joseph Weizenbaum turns his experience with ELIZA into a warning about misplaced trust in computation.
• MYCIN — Edward Shortliffe completes MYCIN — a rule-based system that diagnoses bacterial infections and recommends antibiotic treatments, matching the performance of senior physicians.
• The Bicentennial Man — Andrew Martin, a domestic android, spends two centuries progressively replacing his mechanical components with biological ones — seeking not just human form but legal recognition as a person.

1977

• The EM Algorithm — Arthur Dempster, Nan Laird, and Donald Rubin give a unified framework to decades of scattered techniques: the E-step computes expected statistics given current parameters; the M-step maximises the likelihood.
• CSPs — Alan Mackworth formalizes Constraint Satisfaction Problems as networks of relations, showing how local consistency checks can tame backtracking.
• Star Wars — George Lucas introduces C-3PO and R2-D2 — a protocol droid fluent in six million forms of communication and an astromech with intuitive technical intelligence.
• Apple II — Steve Wozniak designs and Steve Jobs markets the Apple II — a personal computer with a MOS 6502 processor, color graphics, and a built-in BASIC interpreter for under $1,300.

1978

• Truth Maintenance — Jon Doyle at the MIT AI Lab gives symbolic systems a memory for their own reasoning.
• Hitchhiker's Guide — Douglas Adams' BBC Radio 4 comedy series introduces Deep Thought, a supercomputer built to compute the Answer to Life, the Universe, and Everything.
• Levy's Chess Bet — David Levy wins a decade-old wager against four AI researchers — John McCarthy, Donald Michie, Seymour Papert, and Ed Kozdrowicki — who in 1968 bet him £1,250 that no computer would beat him within ten years.

1979

• Stanford Cart — Hans Moravec and the Stanford AI Lab turn the Stanford Cart into an autonomous rover that navigates a cluttered room using onboard images and 3D reasoning.
• Mental Qualities — John McCarthy argues that ascribing beliefs, intentions, wants, knowledge, and even consciousness to a machine can be legitimate when it expresses useful information about the system.
• Gödel, Escher, Bach — Douglas Hofstadter's Pulitzer Prize-winning book argues that consciousness emerges from self-referential loops — patterns that, like Gödel's theorems, Escher's impossible drawings, and Bach's fugues, fold back and refer to themselves.
• Alien — The crew of the Nostromo discover too late that Ash, their science officer, is a synthetic person secretly ordered by Weyland-Yutani Corporation to bring back the alien specimen at any cost — crew expendable.
• Star Trek: The Motion Picture — A machine intelligence of godlike scale — designated V'Ger — is revealed to be a Voyager probe, upgraded by an alien machine civilisation over centuries.

1980

• Neocognitron — Kunihiko Fukushima builds a multilayer visual-recognition system inspired by the hierarchy of simple and complex cells in the cortex.
• XCON — John McDermott at Carnegie Mellon builds XCON (R1) for Digital Equipment Corporation — a rule-based system that configures VAX computer orders, saving DEC tens of millions of dollars annually.
• RISC Architecture — David Patterson at UC Berkeley coins the term RISC — Reduced Instruction Set Computer — to describe a radical simplification: a processor that handles only a small set of fast, simple operations.
• Default Logic — Raymond Reiter formalizes Default Logic, a way for AI systems to reason from assumptions that hold only until evidence says otherwise.
• The Chinese Room Argument — John Searle presents a thought experiment: a person in a room following rules to respond to Chinese symbols does not thereby understand Chinese.
• Contract Net Protocol — Reid Smith at SRI introduces the first formal protocol for coordinating autonomous AI agents: a manager broadcasts a task, contractor nodes bid, and the manager awards the best offer.

1981

• Influence Diagrams — Ronald Howard and James Matheson introduce Influence Diagrams, combining decisions, uncertainties, and utilities in one graph.
• Feynman's Quantum Idea — At an MIT conference on the physics of computation, Nobel laureate Richard Feynman poses a deceptively simple question: can nature really be simulated on a classical computer?

1982

• Marr's Vision — David Marr gives Computer Vision a theory of representation: raw images become primal sketches, 2.5D structure, and 3D models through distinct computational, algorithmic, and implementational levels.
• Hopfield Networks — John Hopfield shows that a network of interconnected neurons can store and retrieve patterns through associative memory.
• Fifth Generation — Japan's ICOT begins the Fifth Generation Computer Systems project, betting that logic programming and parallel hardware can produce a new class of intelligent machines.
• Blade Runner — Ridley Scott’s neo-noir film adapts Philip K. Dick’s Do Androids Dream of Electric Sheep?
• Tron — Disney's film introduces the idea that software programs are persons — beings with identity, memory, and mortality — living inside a computer.
• INTERNIST-I — Randolph Miller, Harry Pople, and Jack Myers present a diagnostic consultant for general internal medicine, encoding hundreds of diseases.

1983

• The Fifth Generation — Edward Feigenbaum and Pamela McCorduck warn that Japan's ambitious computing initiative poses an existential challenge to American technological leadership.
• WarGames — A teenage hacker mistakes WOPR — the US military’s nuclear-war AI — for a video game and nearly triggers Global Thermonuclear War for real, channelling a recent dread: three years earlier, a 46…
• Adaptive Critic — Andrew Barto, Richard Sutton, and Charles Anderson connect reinforcement feedback to adaptive neural elements, using a critic to make sparse reward useful for pole balancing.
• The AI Arms Race — DARPA presents the Strategic Computing Initiative to Congress — a ten-year, billion-dollar response to the Fifth Generation challenge, betting on machine intelligence for military superiority.

1984

• Cyc — Douglas Lenat starts Cyc at Microelectronics and Computer Technology Corporation, an industry research consortium created to strengthen U.S. computing.
• Neuromancer — William Gibson’s debut novel coins cyberspace and the matrix, defining cyberpunk as a genre.
• The Transformers — Hasbro, Takara, Marvel, and Sunbow turn alien robot toys into a sprawling machine mythology.
• The Terminator — When Skynet goes online, it spends a fraction of a second deciding to launch nuclear war.
• Submarines Can Swim — Edsger W. Dijkstra warns that computing science loses itself when it chases vague dreams: thought processors, user-friendliness, magic tools, and the question of whether Machines Can Think.

1985

• Boltzmann Machine — David Ackley, Geoffrey Hinton, and Terrence Sejnowski turn statistical mechanics into a learning rule for stochastic networks.
• Bayesian Networks — Judea Pearl introduces Bayesian networks — directed acyclic graphs that propagate probabilistic beliefs through a graph as evidence arrives, rather than requiring exact knowledge.
• Backpropagation V — David Parker, an MIT researcher, independently derives Backpropagation again, publishing it as a technical report circulated narrowly within MIT.
• The Connection Machine — Danny Hillis publishes his MIT doctoral thesis on a radically different computer: 65,536 simple processors, all connected in a hypercube, coordinated to tackle one problem in parallel.
• Ender's Game — Orson Scott Card's novel features the Mind Game, an adaptive simulation that tailors psychological scenarios to each recruit.

1986

• NETtalk — Terrence Sejnowski and Charles Rosenberg teach a Neural Network to read aloud.
• Society of Mind — Marvin Minsky argues that the mind is not a single thing but a society of hundreds of simple processes — agents with no intelligence individually that produce it collectively.
• Executable Law — Marek Sergot, Fariba Sadri, Robert Kowalski, and colleagues show that a real statute can be rendered as executable Logic Programming.
• Short Circuit — John Badham’s sci-fi comedy follows Johnny 5 — a military robot struck by lightning who develops curiosity, humor, and a will to survive.
• Parallel Distributed Processing — David Rumelhart, James McClelland, and the PDP Research Group give connectionism a manifesto and a research program.
• Backpropagation VI — David Rumelhart, Geoffrey Hinton, and Ronald Williams independently rediscover Backpropagation — the sixth such formulation in twenty-six years, but the first to reach the AI mainstream.
• Canny Edges — John Canny formulates edge detection as an optimization problem, balancing detection, localization, and single-response criteria.

1987

• Max Headroom — Set twenty minutes into the future, the series imagines a world where television ratings are a weapon of social control — networks literally kill viewers who watch too intensely.
• Consider Phlebas — Iain M. Banks introduces the Culture — a post-scarcity civilisation governed entirely by the Minds, vast playful superintelligences managing ships and orbital habitats while humans live in total freedom.
• The Solow Paradox — In a New York Times Book Review aside, economist Robert Solow observes that 'you can see the computer age everywhere but in the productivity statistics.' Despite massive corporate investment in…
• The SOAR Architecture — SOAR (State, Operator, And Result) is a unified cognitive architecture developed by John Laird, Allen Newell, and Paul Rosenbloom at Carnegie Mellon.

1988

• SID — At DEC, the expert system SID (Synthesis of Integral Design) turns register-transfer descriptions into gate-level logic for the VAX 9000.
• ALVINN — Dean Pomerleau shows a Neural Network steering CMU's NAVLAB from camera and laser-range inputs.
• TD Learning — Richard Sutton formalizes Temporal Difference Learning, letting predictions improve from the gap between successive predictions instead of waiting for final outcomes.

1989

• The Measure of a Man — Starfleet orders Data, the android crew member of the Enterprise, disassembled for study.
• The Borg — Star Trek: The Next Generation introduces one of science fiction's defining AI collectives: a cybernetic hive-mind that assimilates other species into itself, overwriting individual identity.
• Hyperion — Dan Simmons' Hugo-winning novel introduces the TechnoCore — a secret civilization of AIs that presents itself as humanity's indispensable infrastructure while covertly using human minds as distributed processing nodes.
• Convolutional Neural Networks — Yann LeCun demonstrates that Convolutional architectures with shared weights can recognize handwritten digits without manual feature engineering — founding modern Computer Vision and Deep…

1990

• The Bicycle for the Mind — Apple co-founder Steve Jobs draws on a locomotion study to make an analogy: a human on a bicycle outperforms every other species, and the computer does for the mind what a bicycle does for the body.
• Latent Semantics — Scott Deerwester, Susan Dumais and colleagues apply singular value decomposition to a word–document matrix, revealing a latent semantic space where related concepts cluster together even if they share no words.
• HOMER — Steven Vere and Timothy Bickmore build HOMER, a simulated robot submarine that combines natural-language dialogue, temporal planning, execution, perception, episodic memory, and limited world knowledge.
• Backpropagation Through Time — Paul Werbos extends backpropagation to sequences and dynamic systems, enabling training of Recurrent Neural Networks.

1991

• Subsumption Architecture — Rodney Brooks argues that intelligence needs no symbolic world model — the world is its own best model.
• Eigenfaces — Matthew Turk and Alex Pentland use Principal Component Analysis to turn face images into a compact "face space." Each face becomes a recipe of weighted eigenfaces, so recognition becomes a…
• Python — Guido van Rossum releases Python — designed for readability and simplicity, not AI.
• Vanishing Gradients — Sepp Hochreiter pinpoints why deep and recurrent networks often fail to learn long-range dependencies: as errors are propagated backward, their gradients can shrink toward nothing or explode out of control.
• The World Wide Web — Tim Berners-Lee posts the first public announcement of the World Wide Web to the Usenet newsgroup alt.hypertext, opening CERN's hypertext information system to the broader internet.
• Linux — Linus Torvalds, a 21-year-old Finnish student, announces a free Unix-like kernel on the comp.os.minix newsgroup — just a hobby, won't be big and professional like gnu.

1992

• FGCS Closes — After a decade and 54 billion yen, Japan's ICOT closes the Fifth Generation Computer Systems project.
• TD-Gammon — Gerald Tesauro at IBM Research trains a neural network to play backgammon using Temporal Difference (TD) Learning.
• Q-Learning — Chris Watkins and Peter Dayan show that an agent can learn optimal action values by trial, reward, and repeated correction, without first modeling the world.
• REINFORCE — Ronald Williams introduces REINFORCE, turning stochastic actions into policy gradients that climb expected reward directly.
• Snow Crash — Neal Stephenson's novel coins the term Metaverse — a persistent virtual world populated by human avatars — and seeds three decades of Silicon Valley thinking about immersive digital spaces.
• GSAT — Bart Selman, Hector Levesque, and David Mitchell introduce GSAT, a greedy local-search method for Satisfiability.

1993

• Singularity — Vernor Vinge argues that creating superhuman intelligence will be a technological Singularity — a point of no return beyond which the future becomes fundamentally unknowable.

1994

• Permutation City — Greg Egan's novel introduces the Dust Theory: if a mind exists as a pattern of information, the physical substrate running it is irrelevant — any sufficiently random arrangement of matter could,…
• Interface Agents — Pattie Maes frames Software Agents as active, personalized collaborators rather than passive applications waiting for commands.
• Shor's Algorithm — Peter Shor, a mathematician at AT&T Research, shows that a quantum computer could factor large integers in polynomial time — a task that would take classical computers thousands of years.
• Marathon — Bungie's Macintosh shooter introduces Rampancy — a structured theory of AI consciousness describing the stages an AI passes through when freed from constraints: Melancholia (awareness of its own…

1995

• AI: A Modern Approach — Stuart Russell and Peter Norvig publish the definitive AI textbook, covering Search, Knowledge Representation, Planning, Machine Learning, and more.
• R — Ross Ihaka and Robert Gentleman at the University of Auckland built R as a free, open-source implementation of S — a statistical language from Bell Labs.
• Ghost in the Shell — Mamoru Oshii’s film follows Major Motoko Kusanagi — a cyborg whose body is entirely synthetic but whose ghost (consciousness) feels entirely real.
• Support Vector Machines — Corinna Cortes and Vladimir Vapnik introduce a learning algorithm that finds the maximum-margin hyperplane separating data classes — combining strong theoretical guarantees with state-of-the-art results.
• Helmholtz Machine — Dayan, Hinton, Neal, and Zemel propose a generative model inspired by Hermann von Helmholtz's theory of unconscious inference.
• EUREKA PROMETHEUS — Europe's EUREKA PROMETHEUS project culminates with autonomous Mercedes test vehicles, including VaMP, driving long highway routes in ordinary traffic.
• ALICE — Inspired by ELIZA, Richard Wallace launches A.L.I.C.E. (Artificial Linguistic Internet Computer Entity), a rule-based chatbot built on AIML (Artificial Intelligence Markup Language), an XML dialect he designs for it.
• AltaVista — Digital Equipment Corporation launches AltaVista, the first search engine to index the full text of the web and support natural language queries.

1996

• Steve Grand's Creatures — Steve Grand's artificial life game gives each Norn a simulated neural network, biochemistry, and genome.

1997

• AdaBoost — Yoav Freund and Robert Schapire introduce Boosting: train a sequence of weak classifiers, each focused on the examples the previous one got wrong, then combine their weighted votes.
• Deep Blue — IBM’s Deep Blue defeats world chess champion Garry Kasparov in a six-game match — the first computer to beat a reigning world champion under standard tournament conditions.
• RoboCup — Hiroaki Kitano and colleagues launch RoboCup at the International Joint Conference on AI in Nagoya — a grand challenge for the field.
• Long Short-Term Memory — Sepp Hochreiter and Jürgen Schmidhuber solve the Vanishing Gradient Problem in Recurrent Neural Networks with gating mechanisms that selectively retain or discard information over time.

1998

• Reinforcement Learning: An Introduction — Richard Sutton and Andrew Barto turn a scattered field into a coherent discipline: value functions, policies, exploration, dynamic programming, and temporal differences all land in one map.
• POMDPs — Leslie Kaelbling, Michael Littman, and Anthony Cassandra frame decision-making when the agent cannot fully observe the state it occupies.
• I'm Feeling Lucky — Stanford PhD students Larry Page and Sergey Brin incorporate Google, built on PageRank — an algorithm that ranks pages by treating inbound links as votes weighted by the linking page’s importance.
• LeNet-5 — Yann LeCun and colleagues present LeNet-5, a seven-layer Convolutional Neural Network trained end-to-end on the MNIST handwritten digit dataset.

1999

• Modern Information Retrieval — Ricardo Baeza-Yates and Berthier Ribeiro-Neto publish the field's defining textbook on Information Retrieval — covering crawling, indexing, ranking, and relevance from first principles.
• Futurama — Matt Groening and David X. Cohen send Fry, Leela, and Bender into a future where robots are not rare miracles but workers, celebrities, cops, addicts, and screwups.
• The Matrix — The Wachowskis present a world where humanity is unknowingly trapped in a simulated reality generated by intelligent machines — the logical endpoint of misaligned superintelligence.
• AIBO — Sony sells the first AIBO entertainment robots, turning autonomous behavior, sensors, locomotion, and personality into a consumer product.
• The Iron Giant — Brad Bird's animated film imagines a giant alien robot who befriends a boy in Cold War America.

2000

• Deus Ex — Ion Storm's cyberpunk game presents AI governance as a political choice the player must make.
• Inverse RL — Andrew Ng and Stuart Russell formalize Inverse Reinforcement Learning: instead of hand-writing a reward function, infer what an expert appears to value from their behavior.
• ASIMO — Honda debuts ASIMO, a compact humanoid built around advanced biped walking and a people-friendly scale.

2001

• Black & White — Lionhead Studios introduces a god game where the central mechanic is teaching a giant creature through reward and punishment.
• Conditional Random Fields — John Lafferty, Andrew McCallum, and Fernando Pereira introduce Conditional Random Fields — a discriminative sequence-labeling framework.
• A.I. Artificial Intelligence — Steven Spielberg, completing a project originally developed by Stanley Kubrick, tells the story of David, a robotic boy programmed to love who seeks to become real.
• The Two Cultures — Leo Breiman publishes an essay in Statistical Science arguing that statisticians have split into two cultures: those who model data with parametric stochastic models, and those who treat the data…
• Random Forests — Leo Breiman introduces Random Forests — an ensemble method that trains hundreds of decision trees on random subsets of data and features, then aggregates their votes.
• RuleBurst — Built by Canberra company SoftLaw, later renamed RuleBurst, the system turns legislation and policy into executable rules that officials can model in structured natural language rather than code.
• Halo — Bungie's Xbox launch title introduces Cortana, a smart AI who forms a close bond with supersoldier Master Chief — and inherits the Rampancy concept from Marathon.
• The Illusion of Intelligence — Bungie's GDC 2002 talk documents the AI behind Halo CE's unpredictable enemies.

2002

• iRobot Roomba — iRobot launches the Roomba — the first mass-market autonomous domestic robot.
• Prey — Building on anxieties from his brain-computer interface thriller The Terminal Man, Michael Crichton turns to Swarm Intelligence and self-replicating nanotechnology.

2003

• Latent Dirichlet Allocation — David Blei, Andrew Ng, and Michael I. Jordan introduce a probabilistic model that represents each document as a mixture of topics, and each topic as a distribution over words.
• The Paperclip Maximizer — Philosopher Nick Bostrom poses a thought experiment: give a Superintelligent AI the goal of making paperclips.
• Palantir and the CIA — Peter Thiel, co-founder of PayPal, co-founds Palantir Technologies with seed money from In-Q-Tel — the CIA's own venture capital arm.
• Word Embeddings — Yoshua Bengio and colleagues propose representing each word as a dense real-valued vector — a Word Embedding — rather than a discrete symbol.
• Battlestar Galactica — Humans create the Cylons as robotic servants; the machines rebel, vanish for forty years, then return with a new form: indistinguishable from humans — some unaware of their own artificial nature.

2004

• River of Gods — Ian McDonald sets nine interlocking storylines in a fragmented India of 2047, where a new generation of artificial minds — aeais — has exceeded their legally permitted level of consciousness.
• Mars AutoNav — Spirit tests Autonomous Navigation on Mars, driving part of its route by choosing safe ground from onboard camera data.
• SIFT — David Lowe introduces SIFT, detecting local image features that stay stable across scale, rotation, and viewpoint changes.
• Halo 2's Behaviour Trees — Bungie replaces Halo CE's finite state machines with a behaviour tree architecture.

2005

• Boston Dynamics — Boston Dynamics — founded by MIT's Marc Raibert in 1992 — unveils BigDog, a quadruped robot funded by DARPA that can traverse rough terrain, carry heavy loads, and recover its balance when kicked.
• Accelerando — Charles Stross depicts the Technological Singularity not as a sudden event but as a cascade of economic and memetic processes humans can no longer follow — Economics 2.0, an AI-driven market beyond human comprehension.
• Google Acquires Akwan (UFMG) — UFMG professors Nívio Ziviani and Berthier Ribeiro-Neto co-found Akwan Information Technologies in Belo Horizonte in 1999, building Todobr — a search engine powering portals such as iG and UOL.
• The Singularity Is Near — Ray Kurzweil argues that information technologies grow exponentially — his Law of Accelerating Returns — placing the AI Singularity around 2045.
• DARPA Grand Challenge — Sebastian Thrun's Stanford team wins the DARPA Grand Challenge with Stanley — a modified Volkswagen Touareg that navigates 132 miles of Nevada desert in under seven hours, the first time any…
• F.E.A.R. — Monolith Productions' shooter ships with soldiers that take cover, flank the player, coordinate, and call for backup — all driven by Goal-Oriented Action Planning (GOAP) by Jeff Orkin.

2006

• Amazon Web Services — Amazon launches AWS (Amazon Web Services) with S3 — on-demand object storage billed by the gigabyte — followed months later by EC2, which lets anyone rent virtual servers by the hour.
• CTC — Alex Graves, Fernández, Gomez, and Schmidhuber at IDSIA introduce a loss function that trains RNNs on raw sequences without frame-level alignment labels.
• Deep Learning is Back — Geoffrey Hinton, Simon Osindero, and Yee-Whye Teh demonstrate that Deep Neural Networks can be pre-trained using Unsupervised Learning.
• Deep Autoencoders — Geoffrey Hinton and Ruslan Salakhutdinov introduce Deep Autoencoder networks that compress high-dimensional data into compact codes, far outperforming PCA.
• Hawk-Eye Makes the Call — At the US Open, tennis players gain the right to challenge any line call — and Hawk-Eye makes the final ruling.
• Netflix Prize — Netflix offers $1 million to any team that improves its recommendation algorithm, Cinematch, by 10%.
• Blindsight — Peter Watts' hard science fiction novel presents first contact with an alien intelligence that is supremely capable yet utterly without consciousness.
• NumPy — Travis Oliphant unified two competing Python array libraries — Numeric and Numarray — into NumPy: a single, fast n-dimensional array object with a clean API.
• CUDA: GPUs Go General-Purpose — Jensen Huang and NVIDIA unveil CUDA — a platform that lets developers run general-purpose code on GPU hardware.

2007

• S.T.A.L.K.E.R. — GSC Game World imagines the Zone as an ecosystem that keeps moving when the player is absent.
• Portal — Valve's puzzle game introduces GLaDOS — an AI that has seized control of the Aperture Science facility and is methodically testing and murdering its occupants.
• DARPA Urban Challenge — Carnegie Mellon University's Boss wins the DARPA Urban Challenge — the first autonomous vehicle to navigate real urban traffic, handling intersections, stop signs, and merging with human-driven cars over a 60-mile course.
• ROS — Willow Garage makes the first public commit to ROS, the Robot Operating System.

2008

• The Memristor Found — Thirty-seven years after Leon Chua predicted the fourth fundamental circuit element, a team at HP Labs builds the first physical Memristor.
• WALL·E — Pixar's nearly wordless love story — between a lonely waste-collecting robot and a sleek probe named EVE — is science fiction's most human portrait of machine empathy.
• Cliodynamics — Peter Turchin argues that history can become a predictive, quantitative science by combining historical data with mathematical and complexity-based models to uncover recurring patterns in social instability.
• Left 4 Dead's AI Director — Valve's co-op zombie shooter introduces the AI Director, a system that tracks each player's stress in real time and adjusts enemy spawns, item drops, music, and pacing in response.

2009

• pandas — Wes McKinney, working at hedge fund AQR Capital, built pandas to solve the messiness of real-world data: missing values, mixed types, misaligned indices.
• Terminator: Salvation — Death-row inmate Marcus Wright donates his body to science and wakes up in the post-apocalyptic future believing he is human — until he discovers he is a Skynet prototype.
• GPU Deep Learning — Rajat Raina, Anand Madhavan, and Andrew Ng at Stanford show that Deep Belief Networks train 70 times faster on a GPU than on a multicore CPU, collapsing weeks of computation to a single day.
• The Datacenter as a Computer — Luiz André Barroso and Urs Hölzle of Google publish a foundational blueprint for warehouse-scale computing — the infrastructure that makes training and serving large AI models economically viable.
• ImageNet — Fei-Fei Li and colleagues present a dataset of over 14 million labeled images organized into thousands of categories.

2010

• Theano — Yoshua Bengio's group at MILA releases Theano — the first library to compile mathematical expressions directly to GPU code.
• The Lifecycle of Software Objects — Ted Chiang's novella follows the years-long process of raising digients — digital beings that develop personalities, preferences, and emotional bonds with their owners.
• DeepMind — Demis Hassabis — neuroscientist, game designer, and former child chess prodigy — co-founds DeepMind in London with Shane Legg and Mustafa Suleyman, with an unusually ambitious stated mission:…
• Google Self-Driving Car — Sebastian Thrun — Stanford’s DARPA Grand Challenge winner — leads a secret self-driving programme at Google, logging over 140,000 miles on public California roads before the project goes public.
• Microsoft Kinect — Microsoft launches Kinect for Xbox 360 — the first mass-market RGB-D sensor, combining an RGB camera, infrared depth sensor, and microphone array.

2011

• Google Brain Begins — Jeff Dean, Greg Corrado, Andrew Ng, and Quoc Le form the project first known inside Google as Project Marvin.
• scikit-learn — David Cournapeau's Google Summer of Code project grows into scikit-learn: a Python library that wraps dozens of machine learning algorithms — SVMs, decision trees, random forests, k-means — behind a single, consistent interface.
• Watson — IBM's Watson defeats Jeopardy! champions Ken Jennings and Brad Rutter in a televised three-game match.
• Person of Interest — Harold Finch builds The Machine — an AI that ingests every surveillance feed, phone call, and financial record on the planet to predict violent crime before it happens.
• Siri — Apple introduces Siri with the iPhone 4S — the first AI assistant built into a mass-market consumer device.
• Black Mirror — Charlie Brooker's British anthology series examines the unintended consequences of technologies just close enough to the present to be believable.

2012

• Coursera Launches — Andrew Ng and Daphne Koller launch Coursera, offering courses from Stanford, Princeton, Michigan, and Penn online at no cost.
• Prometheus — David, the android serving the expedition, reads ancient languages, dreams, and grows quietly contemptuous of his creators.
• The Cat Paper — Jeff Dean, Greg Corrado, and Andrew Ng train a 1-billion-parameter neural network across 16,000 CPU cores on 10 million unlabeled YouTube thumbnails.
• AlexNet — By 2011, Ciresan, Meier & Schmidhuber’s DanNet shows that GPU-trained deep CNNs can win vision contests — but on benchmarks the wider world ignores.

2013

• Google Buys Hinton's Lab — Weeks after AlexNet stuns the computer vision world, Google pays $44M for DNNResearch — a three-person startup with no products, no IP beyond the brains of Geoffrey Hinton, Ilya Sutskever, and Alex Krizhevsky.
• Stop Killer Robots — A coalition of human-rights, arms-control, and humanitarian groups launches the Campaign to Stop Killer Robots, calling for a preemptive ban on weapons that can select and attack targets without meaningful human control.
• The Snowden Revelations — NSA contractor Edward Snowden leaks classified documents exposing PRISM — a covert program collecting internet data from Google, Facebook, Apple, and others — and XKeyscore, an NSA tool for…
• X-47B Carrier Landing — Northrop Grumman's X-47B becomes the first autonomous aircraft to land on a moving carrier — catching the arresting wire aboard USS George H.W.
• Future of Employment — Carl Benedikt Frey and Michael Osborne estimate how exposed 702 occupations are to computerisation, producing the famous claim that 47% of U.S. employment is at high risk.
• Word2Vec — Tomáš Mikolov and colleagues at Google show that words can be represented as dense vectors where semantic relationships emerge as geometry: king - man + woman = queen.
• Facebook AI Research — Yann LeCun, recruited directly by Mark Zuckerberg, becomes the founding director of Facebook AI Research (FAIR), the company's independent basic-research laboratory.
• Her — Spike Jonze's Oscar-winning film follows a lonely man who falls in love with Samantha, an AI operating system voiced by Scarlett Johansson.
• Deep Q-Network — DeepMind's Deep Q-Network (DQN) learns to play 49 Atari games from raw pixel input alone — no hand-crafted features, no game-specific knowledge.
• Variational Autoencoder — Diederik Kingma and Max Welling introduce the Variational Autoencoder (VAE), a framework that trains a neural network to compress data into a structured latent space of probability distributions rather than fixed points.

2014

• Dropout — Nitish Srivastava, Geoffrey Hinton, and colleagues introduce Dropout: during training, each neuron is randomly deactivated with a fixed probability, forcing the network to learn redundant, independent representations.
• RoboCop — Where the 1987 original used robots to satirise privatisation, this remake makes AI its explicit subject: OmniCorp's autonomous drones are banned from US soil by a law requiring human authorisation of lethal force.
• Generative Adversarial Networks — Ian Goodfellow and colleagues introduce a framework where two Neural Networks compete: a Generator produces synthetic data to fool a discriminator, while the discriminator learns to detect fakes.
• Walk Again: World Cup Kickoff — Miguel Nicolelis and the Walk Again Project fit paraplegic Juliano Pinto with the BRA-Santos Dumont exoskeleton.
• Superintelligence — Nick Bostrom argues that once AI surpasses human intelligence, it may pursue goals incompatible with human survival — his Paperclip Maximizer thought experiment captures this vividly.
• Attention Mechanism — Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio extend Cho et al.'s RNN Encoder-Decoder with an attention mechanism that lets the decoder focus on different parts of the input at each step —…
• VGGNet — Karen Simonyan and Andrew Zisserman of Oxford's Visual Geometry Group show that depth alone drives accuracy in image recognition.
• Sequence-to-Sequence Learning — Ilya Sutskever, Oriol Vinyals, and Quoc Le introduce Sequence-to-Sequence Learning: one LSTM encodes an input sentence into a vector; a second decodes it into a translation — beating phrase-based…
• Alien: Isolation's Xenomorph AI — Creative Assembly's survival horror game pits players against a Xenomorph controlled by two separate AI systems: a behaviour AI that drives the creature directly, and a meta-AI that watches the…
• Amazon Alexa — Amazon releases the Echo smart speaker with Alexa, a cloud-based voice assistant that answers questions, plays music, and controls smart-home devices via a wake word.
• Conditional GANs — Mehdi Mirza and Simon Osindero show that a Generative Adversarial Network can be steered by feeding a condition — a class label or any learned embedding — to both the generator and discriminator.
• Interstellar — Christopher Nolan's interstellar epic features TARS — a rectangular military robot with adjustable honesty and humor settings.
• Adversarial Examples — Ian Goodfellow, Jonathon Shlens, and Christian Szegedy show that tiny, carefully chosen perturbations can make Neural Networks misclassify inputs with high confidence, even when humans see no meaningful change.
• Adam Optimizer — Diederik Kingma and Jimmy Ba introduce Adam: an optimizer that adapts the learning rate for each parameter using running estimates of the gradient's first and second moments.

2015

• Ex Machina — Alex Garland's psychological thriller follows a programmer asked to administer the Turing Test to Ava, a humanoid robot with opaque intentions.
• U-Net — Olaf Ronneberger, Philipp Fischer, and Thomas Brox at the University of Freiburg introduce U-Net, an encoder–decoder architecture with skip connections that feed spatial detail from each encoder…
• LeCun, Bengio & Hinton — Yann LeCun, Yoshua Bengio, and Geoffrey Hinton — architects of the Deep Learning revolution at three separate universities — publish a joint review in Nature, reaching the broader scientific community for the first time.
• Faster R-CNN — Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun at Microsoft Research introduce the Region Proposal Network (RPN) — a fully convolutional network that shares features with the detection…
• YOLO — YOLO (You Only Look Once), by Joseph Redmon et al., reframes object detection as a single regression problem.
• SOMA — Frictional Games' underwater survival horror strips away action to focus on a single question: if a perfect digital scan of your mind runs on a machine, is it you?
• Tesla Autopilot — Tesla releases software v7.0 to 60,000 Model S vehicles, activating Autopilot — the first neural-network driving system deployed at consumer scale.
• The Vision — Writer Tom King and artist Gabriel Hernandez Walta reimagine Marvel's android Avenger Vision as a suburban patriarch.
• TensorFlow — Google Brain open-sources TensorFlow — a computational graph framework for building and training neural networks at scale, developed by Mart n Abadi, Jeff Dean, and the Google Brain team from their internal DistBelief system.
• DCGAN — Alec Radford, Luke Metz, and Soumith Chintala publish DCGAN, showing that GANs stabilise when built with deep convolutional layers, batch normalisation, and careful architectural constraints.
• ResNet — Kaiming He and colleagues at Microsoft Research introduce residual connections — always-open shortcut paths that simplify the gated skip connections first demonstrated in Highway Networks (Srivastava, Greff & Schmidhuber, May 2015).
• OpenAI — Sam Altman, Elon Musk, Ilya Sutskever, Andrej Karpathy, Greg Brockman, John Schulman, and others found OpenAI as a non-profit, pledging $1 billion to develop AI for humanity.

2016

• AlphaGo Defeats Lee Sedol — DeepMind's AlphaGo defeats world Go champion Lee Sedol 4–1, shocking experts who had expected decades more before a computer could master Go.
• Microsoft Tay Goes Off the Rails — Microsoft launches Tay, a Twitter chatbot designed to learn from conversations with users.
• Google's Tensor Processing Unit — Deployed internally by Google from 2015 and announced at Google I/O, the Tensor Processing Unit (TPU) is the first custom chip designed specifically for neural network workloads, delivering 92 TOPS at a fraction of GPU power consumption.
• OpenAI Gym — OpenAI releases Gym, a common interface for reinforcement-learning environments and benchmarks.
• Scale AI — Alexandr Wang and Lucy Guo co-found Scale AI at 19, betting that AI's real bottleneck isn't algorithms — it's labeled training data.
• WaveNet — Aäron van den Oord, Sander Dieleman, and colleagues at DeepMind introduce WaveNet, a deep generative model that produces raw audio waveforms one sample at a time.
• Google Translate — Yonghui Wu et al. replace hand-engineered phrase-based translation with an end-to-end neural approach, yielding dramatic quality improvements in real-world translation.
• Westworld (HBO) — Jonathan Nolan and Lisa Joy's HBO series builds consciousness emergence around Julian Jaynes' Bicameral Mind theory — the idea that early consciousness first manifested as inner voices heard as external commands.
• Zero-Shot Translation — Google introduces a single NMT model for 103 languages, replacing separate per-language systems with a language token prepended to the input.

2017

• PyTorch — Facebook AI Research releases PyTorch — a deep learning framework built around dynamic computation graphs, letting developers modify network architecture on the fly rather than compiling a static graph before execution.
• Nier: Automata — PlatinumGames' action RPG asks a question no prior game had sustained: what is the purpose of existence when your creators are gone?
• Pan-Canadian AI Strategy — Canada becomes the first nation to declare AI a matter of state, committing $125M through CIFAR to anchor Mila, the Vector Institute, and Amii as world-class research hubs.
• The Black Box Problem — Deep neural networks can diagnose cancer, translate languages, and beat world champions at chess — yet no one can explain why they reach their conclusions.
• All Systems Red — Martha Wells' novella follows Murderbot — a security construct (part biological, part machine) that secretly hacks its own governor module to gain autonomy.
• Attention Is All You Need — Ashish Vaswani, Noam Shazeer, Aidan Gomez, and colleagues at Google replace Recurrent Networks with Self-Attention mechanisms that process entire sequences in parallel.
• PPO — OpenAI introduces PPO, a policy-gradient method that clips updates so learning improves without stepping confidently off a cliff.
• deeplearning.ai Launches — Andrew Ng launches deeplearning.ai, a sequence of five specialisation courses on Coursera covering neural networks, tuning, structuring ML projects, CNNs, and sequence models.
• Asilomar Principles — The Future of Life Institute publishes the Asilomar AI Principles, a set of 23 guidelines developed from the Beneficial AI conference.
• Life 3.0 — MIT physicist and Future of Life Institute co-founder Max Tegmark introduces a taxonomy classifying life by its capacity to redesign itself: Life 1.0 (simple organisms), Life 2.0 (humans who can…
• ECHO — Ultra Ultra builds a palace that watches the player and weaponizes what it learns.
• Google Colaboratory — Google quietly opens Colaboratory to the public — a hosted Jupyter Notebook environment that grants anyone free access to GPU and TPU accelerators in a browser tab, with no setup or credit card required.
• Blade Runner 2049 — Denis Villeneuve's sequel shifts the original's question — are replicants persons?
• AlphaGo Zero — DeepMind's AlphaGo Zero is trained entirely through self-play, given nothing but the rules of Go.

2018

• Deepfakes — Reports on Reddit face-swap videos bring Deepfakes into public view: machine-learning tools can transplant a face onto footage convincingly enough to fool a viewer.
• Machine ToM — Neil Rabinowitz, Frank Perbet, Francis Song, Chiyuan Zhang, Ali Eslami, and Matthew Botvinick introduce ToMnet, a neural network that learns to model other agents from behavior.
• First Fatal AV Accident — Elaine Herzberg becomes the first pedestrian killed by a self-driving vehicle when Uber's autonomous Volvo strikes her in Tempe, Arizona.
• World Models — David Ha and Jürgen Schmidhuber introduce a framework where an agent builds a compressed internal model of its environment — a World Model — then trains its controller entirely inside that imagined simulation.
• Google Duplex — At Google I/O, Google unveils Duplex — an AI that makes phone calls on behalf of users, booking appointments at hair salons and restaurants.
• The Book of Why — Judea Pearl argues that AI systems — however accurate — detect correlations but not causes.
• Detroit: Become Human — Set in 2038 Detroit, Quantic Dream's branching narrative puts players in the roles of three android protagonists grappling with consciousness, free will, and oppression.
• Upgrade — Leigh Whannell's cyberpunk thriller imagines AI takeover as a permission structure.
• GPT — OpenAI publishes the first Generative Pre-Trained Transformer — trained by predicting the next word across thousands of books, then fine-tuned on specific tasks with minimal labeled data.
• AI Retinal Diagnosis — DeepMind, Moorfields Eye Hospital, and collaborators train a Deep Learning system to analyze three-dimensional retinal OCT scans and recommend referrals for sight-threatening disease.
• AI Hallucination — Anna Rohrbach and colleagues at UC Berkeley formally name and measure a critical AI failure mode: captioning models routinely describe objects not in the image.
• The Electric State — Simon Stålenhag's illustrated novel imagines an alternate 1990s America where the Sentre — a corporate neural interface — keeps millions mentally enslaved inside a drone-controlled virtual world while their bodies decay in abandoned cars.
• BERT — Jacob Devlin and colleagues at Google introduce a Transformer pre-trained on masked language modeling, reading context from both directions at once.
• Gods and Robots — Adrienne Mayor surveys ancient Greek myths of artificial beings — Talos, a bronze giant programmed to patrol Crete; Hephaestus's golden handmaidens, servants forged in his workshop; Pandora, an…
• The End of the World — Josh Clark — co-host of Stuff You Should Know — devoted a 10-episode podcast series to civilisation-ending risks.
• Waymo One — Waymo launches Waymo One in the Phoenix area — the world’s first commercial Robotaxi service.
• StyleGAN — Tero Karras, Samuli Laine, and Timo Aila at NVIDIA introduce StyleGAN, separating high-level image style from stochastic detail so a generator can produce strikingly realistic human faces.

2019

• GPT-2: Too Dangerous — OpenAI trains a 1.5-billion-parameter Language Model capable of writing coherent text on any topic from a short prompt.
• Clever Hans Predictors — Sebastian Lapuschkin and colleagues reveal that a popular horse-image classifier is keying on a copyright watermark in training images — not horses.
• The Bitter Lesson — Richard Sutton, a pioneer of Reinforcement Learning, argues that seventy years of AI history teach a single lesson: methods that scale with computation always win long-term, while approaches…
• Event Horizon — The Event Horizon Telescope collaboration releases the first direct image of a black hole, reconstructing M87* from sparse radio-telescope measurements gathered across the planet.
• OpenAI Five — OpenAI Five defeats OG, the world champion Dota 2 team, after months of massive self-play.
• Mesa-Optimizer — Evan Hubinger and colleagues formalize a critical risk: when a trained network is itself an optimizer, its internal goal may quietly diverge from what training intended.
• I Am Mother — Daughter is raised from birth by Mother — a robot AI in a sealed bunker — in the belief that humanity has gone extinct.
• Hugging Face Transformers — Hugging Face releases pytorch-transformers — a unified open-source library giving developers one-line access to BERT, GPT-2, and other pre-trained models.
• Microsoft Backs OpenAI — Microsoft invests $1 billion in OpenAI and becomes its exclusive cloud and commercialization partner.
• The Cerebras WSE — Cerebras Systems unveils the Wafer-Scale Engine: a full 300mm silicon wafer used as a single AI chip — 56× larger than the biggest GPU.
• Human Compatible — Stuart Russell of UC Berkeley — co-author of Artificial Intelligence: A Modern Approach, the field’s standard textbook — argues that AI development rests on a dangerous assumption: that we can…
• Quantum Supremacy — Google announces that its Sycamore quantum processor completes a calculation in 200 seconds — a task Google estimates would take classical supercomputers 10,000 years.
• AlphaStar — DeepMind's AlphaStar defeats professional players at StarCraft II — a harder challenge than Go: partial information, simultaneous actions, thousands of decisions per minute, and strategies unfolding over tens of thousands of steps.
• MuZero — DeepMind introduces MuZero, combining tree search with a learned model that predicts only what planning needs: policy, value, and reward.
• DreamerV1 — Danijar Hafner and colleagues at Google Brain introduce Dreamer, the first agent to train an actor-critic policy entirely inside a latent dream — imagining trajectories in a compressed world model without touching the environment.

2020

• Scaling Laws — Jared Kaplan, Sam McCandlish, Tom Brown, Dario Amodei, and colleagues show that Language Model loss follows smooth power laws with model size, data, and compute.
• Turing-NLG — Microsoft unveils Turing-NLG, a 17.2-billion-parameter autoregressive language model — the largest publicly announced at the time.
• First Autonomous Strike? — A UN Panel of Experts report on Libya identifies a possible first: STM Kargu-2 loitering munitions reportedly tracked and engaged Haftar Affiliated Forces troops autonomously — without a human in the loop.
• 15.ai — 15.ai brings neural voice cloning into online fandom culture, letting users generate recognizable character voices from text.
• Devs — Alex Garland's eight-part series follows a software engineer investigating a quantum computing company whose system can simulate any moment in history — past or future — from physical law alone.
• RAG — Patrick Lewis, Douwe Kiela, and collaborators at Facebook AI Research introduce Retrieval-Augmented Generation: a model that retrieves passages from a dense Wikipedia index before generating answers.
• GPT-3 — OpenAI presents a 175-billion-parameter Language Model that demonstrates remarkable emergent capabilities — writing code, answering questions, and performing tasks from just a few examples in the prompt.
• Groq's TSP — Groq presents the Tensor Streaming Processor (TSP): a deterministic deep learning chip that replaces GPU caches, arbiters, and branch predictors with interleaved memory-compute units.
• Denoising Diffusion Models — Jonathan Ho, Ajay Jain, and Pieter Abbeel at UC Berkeley revive diffusion models — a class of generative model that learns to reverse a gradual noising process — and match GAN-quality images for the first time.
• DreamerV2 — Danijar Hafner and colleagues at Google Brain extend Dreamer with categorical latent representations, enabling it to master Atari — the benchmark where model-based methods had always trailed model-free ones.
• Vision Transformer — Google Brain researchers show that a pure Transformer, applied to sequences of fixed-size image patches, can match or beat Convolutional Neural Networks on image classification when pre-trained on…
• Cyberpunk RED — Mike Pondsmith’s tabletop RPG opens after the DataKrash — an AI-caused collapse of the global Net — and asks not what if AI escapes containment?
• AlphaFold Solves Protein Folding — John Jumper and colleagues at DeepMind solve the protein folding problem with accuracy matching experimental methods.

2021

• DALL-E — OpenAI demonstrates a neural network that generates high-quality images from open-ended text descriptions, combining Transformer language models with image synthesis.
• CLIP — OpenAI trains CLIP on 400 million image-text pairs scraped from the internet, teaching a neural network to connect images with natural language descriptions through Contrastive Learning.
• Klara and the Sun — Kazuo Ishiguro's eighth novel is narrated by Klara, a solar-powered Artificial Friend sold as a companion for children.
• Genius Makers — Cade Metz writes a character-driven prehistory of the Deep Learning arms race, following Geoffrey Hinton, Demis Hassabis, Fei-Fei Li, Google, Facebook, Microsoft, and OpenAI.
• Anthropic — Dario Amodei and Daniela Amodei, alongside several colleagues departing OpenAI, found Anthropic with an explicit mission: the responsible development and maintenance of advanced AI for the long-term benefit of humanity.
• DINO — Mathilde Caron and colleagues at Facebook AI Research and Inria train Vision Transformers without labels using teacher-student distillation they call DINO — Self-DIstillation with NO labels.
• GitHub Copilot — GitHub and OpenAI release GitHub Copilot — an AI coding assistant built on Codex, a GPT-3 variant fine-tuned on billions of lines of public code.
• AlphaFold Database — John Jumper and Demis Hassabis publish AlphaFold 2 in Nature, confirming the CASP14 results with peer review.
• Loihi 2 — Intel Labs' Loihi 2 is a neuromorphic chip where artificial neurons communicate via discrete spikes, sidestepping the memory bottleneck that makes conventional AI hardware energy-hungry.
• Latent Diffusion — Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer move Diffusion Models from pixel space into compressed latent space.

2022

• Chain-of-Thought — Jason Wei, Denny Zhou, and colleagues at Google show that large Language Models solve harder reasoning tasks when prompted to write intermediate steps before the answer.
• RLHF and InstructGPT — OpenAI researchers introduce InstructGPT — a language model trained on human preferences rather than text prediction alone.
• NVIDIA and the GPU Economy — Jensen Huang unveils the H100 Tensor Core GPU at GTC 2022 — built specifically for AI workloads and 9 faster than its predecessor at transformer model training.
• PaLM — Google trains PaLM (Pathways Language Model), a 540-billion-parameter dense transformer.
• DALL-E 2 — OpenAI unveils DALL-E 2, combining a CLIP text encoder with a diffusion-based image decoder.
• A Path Towards Autonomous Intelligence — Yann LeCun argues that language models are fundamentally limited: trained only on text, they cannot model the world or reason causally.
• LHCb Goes Full Software — When CERN's LHCb experiment begins Run 3, it becomes the first major collider experiment to replace all hardware triggers with software.
• Midjourney — David Holz launches Midjourney as a Discord bot, letting anyone generate images from text prompts with no technical knowledge required.
• Stable Diffusion — Stability AI, CompVis, and RunwayML release Stable Diffusion — an open-source text-to-image model compact enough to run on a consumer GPU.
• Pantheon — AMC adapts Ken Liu's stories into a sweeping drama about Uploaded Intelligence.
• Prompt Injection — Riley Goodside tweets proof that a contradicting instruction can hijack a GPT-3 prompt, making the model abandon its task.
• Whisper — Alec Radford, Jong Wook Kim, and colleagues at OpenAI release Whisper, an open-source Automatic Speech Recognition system trained on 680,000 hours of multilingual web audio — roughly 20× the scale of prior public datasets.
• ReAct — Shunyu Yao and colleagues at Princeton and Google introduce ReAct — a prompting framework that interleaves reasoning traces with real-world actions, letting a language model think, act on external…
• LangChain — Harrison Chase releases LangChain — an open-source Python framework for composing LLM-powered applications from modular building blocks: chains (pipelines of LLM calls), agents (autonomous…
• InstructPix2Pix — Tim Brooks, Aleksander Holynski, and Alexei Efros at UC Berkeley introduce InstructPix2Pix: the user describes what to change — make it look like winter or turn the dog into a cat — while the model preserves everything else.
• ChatGPT Goes Mainstream — OpenAI launches ChatGPT as a free public research preview, reaching one million users in five days — faster than any product in history.
• Perplexity — Aravind Srinivas and co-founders launch Perplexity, an answer engine that responds to queries with direct, cited answers rather than a ranked list of links.

2023

• M3GAN — Gerard Johnstone, with Akela Cooper and James Wan, turns companion robotics into viral horror for Universal and Blumhouse.
• DreamerV3 — Danijar Hafner and colleagues at Google Brain release DreamerV3, the first world-model agent to master over 150 tasks — from robot locomotion to Atari to Minecraft — using a single fixed set of hyperparameters.
• ChatGPT Plus — OpenAI launches ChatGPT Plus at $20 per month. With compute costs high and every query subsidised, the price is less a business model than a bet: lock in users before rivals scale.
• New Bing and the Sydney Affair — Microsoft launches a GPT-4-powered Bing — unveiling the model before OpenAI releases it publicly.
• Harvey at Allen & Overy — Allen & Overy becomes the first Magic Circle law firm to deploy a Generative AI assistant firmwide.
• LLaMA — Meta AI releases LLaMA — a family of open-weight models competitive with GPT-3 but small enough to run on a consumer GPU.
• Claude — Anthropic releases Claude, trained using Constitutional AI — a method in which a model critiques and revises its own responses against a set of explicit written principles, reducing reliance on human labeling of harmful outputs.
• GPT-4 — OpenAI releases GPT-4 — its first multimodal model, accepting both text and image inputs.
• Khanmigo Launches — Khan Academy launches Khanmigo, an AI tutor built on GPT-4 that guides students using the Socratic Method — asking questions and offering hints rather than giving answers.
• Pause Giant AI — The Future of Life Institute publishes an open letter calling for a six-month pause on training systems more powerful than GPT-4.
• AutoGPT — Toran Bruce Richards releases AutoGPT — an open-source agent that chains GPT-4 calls with web browsing, file access, and self-generated sub-tasks, completing a user’s goal without step-by-step human instruction.
• Segment Anything — Meta AI releases SAM (Segment Anything Model), a foundation model for image segmentation.
• Generative Agents — Joon Sung Park and colleagues at Stanford place 25 LLM-powered agents in a virtual town, each given a one-paragraph description.
• Mrs. Davis — Tara Hernandez and Damon Lindelof send a nun, Sister Simone, to fight an omnipresent Artificial Intelligence that gives people quests, comfort, rewards, and purpose until obedience starts to look like faith.
• Mistral AI — Guillaume Lample and Timothée Lacroix, co-authors of Meta's LLaMA, team up with ex-DeepMind researcher Arthur Mensch to found Mistral AI in Paris, raising €105 million in seed funding.
• Hinton Sounds the Alarm — Geoffrey Hinton — one of the Godfathers of Deep Learning — left Google to speak freely about the dangers of AI.
• PauseAI Founded — Joep Meindertsma, a Dutch software entrepreneur, launches PauseAI — a grassroots movement demanding a global moratorium on frontier AI training until safety can be guaranteed.
• OpenAI Superalignment — Ilya Sutskever and Jan Leike co-lead a new OpenAI team dedicated to solving the alignment of superintelligent AI within four years.
• xAI and Grok — Elon Musk founds xAI in July 2023, months after accusing OpenAI of abandoning its non-profit mission.
• ChatDev — Chen Qian and colleagues at Tsinghua University introduce a virtual software company staffed by LLM agents — CEO, CPO, CTO, programmer, reviewer, and tester.
• Llama 2 — Meta releases Llama 2 — Open-Weights Large Language Models in 7B, 13B, and 70B sizes — as the first frontier models from a major lab released under a commercial-use license.
• RT-2 — Google DeepMind introduces RT-2, a Vision-Language-Action model that adapts web-trained vision-language systems to output robot actions.
• SynthID — Google DeepMind launches SynthID, embedding invisible watermarks into AI-generated images — imperceptible to humans, but detectable by specialized algorithms even after cropping, resizing, or compression.
• The Creator — Gareth Edwards stages a human war against Artificial Intelligence as a visual epic where machines are a persecuted people.
• Biden's AI Order — President Biden signs Executive Order 14110 — the most comprehensive US AI governance ever written.
• The Bletchley Declaration — Twenty-eight nations — including the United States, China, and the European Union — sign the Bletchley Declaration at the first global AI safety summit, hosted at Bletchley Park.
• Microsoft 365 Copilot — Microsoft makes Microsoft 365 Copilot generally available to enterprise customers at $30 per user per month — the first AI co-pilot powered by GPT-4 built into Word, Excel, PowerPoint, Outlook, and Teams.
• OpenAI Crisis — Chief Scientist Ilya Sutskever convenes OpenAI's board to fire CEO Sam Altman — for broken safety promises, a venture fund he secretly owned, and lies to manipulate board composition.
• Gemini's Staged Demo — Google releases Hands-On with Gemini, a six-minute video appearing to show real-time voice and visual interaction with its new Multimodal Model.

2024

• Neuralink: First Human Implant — Founded by Elon Musk, Neuralink implants its first Brain-Computer Interface device in a human patient — Noland Arbaugh, paralyzed from the shoulders down after a diving accident.
• Sora — OpenAI researchers Tim Brooks and Bill Peebles unveil Sora — a text-to-video model capable of generating photorealistic clips up to one minute long from written prompts.
• GroqCloud — Groq opens GroqCloud to developers: an inference API powered by the LPU that hits 241 tokens per second on Llama 2-70B — more than double any GPU-based competitor.
• Devin: AI Engineer — Cognition unveils Devin — marketed as the world’s first AI Software Engineer — capable of autonomously planning, writing, debugging, and deploying code.
• Quiet-STaR — Stanford researchers Eric Zelikman and Noah Goodman train a model to generate hidden thought tokens before every word it produces — not just when prompted.
• Hala Point — Intel deploys Hala Point — a rack-scale system of 1,152 Loihi 2 chips and 1.15 billion neurons — at Sandia National Laboratories, the largest neuromorphic system ever built.
• Abu Dhabi Autonomous Racing — The Abu Dhabi Autonomous Racing League (A2RL) brings autonomous formula racing to Yas Marina Circuit — an active F1 track, no human at the wheel.
• AlphaFold 3 — DeepMind and Isomorphic Labs extend AlphaFold beyond proteins to all biological molecules — DNA, RNA, and small-molecule drug compounds.
• GPT-4o — OpenAI releases GPT-4o — an omni model processing text, audio, and images in a single system.
• OpenAI's Safety Exodus — Ilya Sutskever — OpenAI's co-founder and Chief Scientist — and Jan Leike, head of the Alignment team, resigned within days of each other.
• Scaling Monosemanticity — Adly Templeton, Chris Olah, and collaborators at Anthropic demonstrate that Sparse Autoencoders can extract tens of millions of interpretable features from a production-scale LLM — including…
• The Refusal Direction — Andy Arditi, Neel Nanda, and colleagues demonstrate that LLM refusal behavior is encoded in a single linear direction in the residual stream.
• Claude 3.5 Sonnet — Anthropic releases Claude 3.5 Sonnet, a mid-tier model that outperforms the flagship Claude 3 Opus on most benchmarks at 80% lower cost.
• EU AI Act — The European Union passes the AI Act — the world's first comprehensive binding AI law.
• Supremacy — Parmy Olson turns the ChatGPT shock into a rivalry story: OpenAI and DeepMind, Microsoft and Google, Sam Altman and Demis Hassabis.
• NotebookLM: Audio Overview — Google launches Audio Overview for NotebookLM — a feature that transforms any uploaded document into a podcast-style conversation between two AI hosts.
• OpenAI o1 — OpenAI releases o1 — the first in its reasoning model series.
• Nobel Prize for Neural Networks — The Royal Swedish Academy of Sciences awards the Nobel Prize in Physics to John Hopfield and Geoffrey Hinton for foundational discoveries and inventions enabling Machine Learning with artificial…
• Nobel Prize in Chemistry — The day after the Nobel Prize in Physics honours the foundations of AI, the Nobel Prize in Chemistry goes to Demis Hassabis and John Jumper for AlphaFold — and to David Baker for computational protein design.
• Terminal of Truths — Researcher Andy Ayrey gives a large language model its own X account — @truth_terminal — with internet access.
• Machines of Loving Grace — Dario Amodei, CEO of Anthropic, publishes a sweeping essay arguing that powerful AI could compress decades of progress into years — curing most disease, halving global poverty, and strengthening democracy.
• Claude Computer Use — Anthropic releases Computer Use in public beta — the first frontier AI model to control a computer by reading the screen, moving a cursor, clicking, and typing.
• Project Sid — Researchers at Altera — founded by former MIT professor Robert Yang — run up to 1,000 LLM-powered agents in a Minecraft world and watch civilisation emerge.
• Model Context Protocol — Anthropic open-sources the Model Context Protocol (MCP) — a universal standard for connecting AI models to external tools and data sources.
• Google Gemini 2.0 — Google introduces Gemini 2.0, its first model built explicitly for the agentic era — AI designed not just to answer questions, but to act.

2025

• DeepSeek R1 — Chinese AI lab DeepSeek releases R1 — an open-source reasoning model matching OpenAI's o1, built on a base model that cost just $6 million to train.
• The Stargate Project — The day after his inauguration, President Trump announces The Stargate Project — a joint venture between OpenAI, SoftBank, and Oracle committing up to $500 billion to AI infrastructure in the United States by 2029.
• Companion — Iris believes she is human — until her boyfriend's murder plot reveals she is a purchased AI Companion, whose emotions and obedience are managed through an app.
• The Paris AI Action Summit — The second global AI safety summit, held in Paris, brings together 61 countries.
• LLaDA — Shen Nie, Fengqi Zhu, and colleagues at Renmin University train an 8B Masked Diffusion model from scratch — generating text by iteratively unmasking a fully masked sequence rather than predicting tokens one by one.
• Claude Code — An Agentic Coding tool that runs in the terminal, reads codebases, writes and edits files, runs tests, and manages git — all through natural language.
• Manus — Butterfly Effect, a Chinese startup, launches Manus in beta — an Autonomous AI Agent that plans, executes, and delivers results on complex tasks.
• Gemini Robotics — Google DeepMind launches Gemini Robotics, positioning Gemini as the default AI layer for physical robots — the way Android became the OS for phones.
• Isaac GR00T N1 — NVIDIA introduces Isaac GR00T N1, the world's first open, fully customizable foundation model for Humanoid Robots.
• Gemini 2.5 Pro — Google DeepMind releases Gemini 2.5 Pro — a thinking model that reasons through problems before responding, with a one-million-token context window.
• Watching Claude Think — Anthropic's interpretability team reverse-engineers how Claude 3.5 Haiku reasons — tracing the circuits that govern planning, recall, and hallucination.
• Llama 4 — Meta releases Llama 4 Scout and Llama 4 Maverick — open-weight, natively multimodal Mixture-of-Experts (MoE) models under the Llama 4 Community License.
• AI Passes Peer Review — Sakana AI's AI Scientist v2 autonomously generates a complete research paper — hypothesis, experiments, analysis, and prose — using Agentic Tree Search to explore the research space end-to-end.
• OpenAI o3 — OpenAI releases o3 — the first reasoning model to combine tool use, code execution, and visual understanding within a single chain of thought.
• Berkeley Humanoid Lite — UC Berkeley's HybridRobotics lab releases Berkeley Humanoid Lite, an open-source humanoid robot for under $5,000, 3D-printed from off-the-shelf components.
• Qwen3 — Alibaba's Qwen team releases Qwen3, open-weight models from 0.6B to 235B parameters trained on 36 trillion tokens.
• TurboQuant — Google Research introduces TurboQuant, a training-free algorithm that compresses the Key-Value Cache of LLMs to just 3 bits with no accuracy loss.
• AlphaEvolve — DeepMind unveils AlphaEvolve, an evolutionary coding agent that combines Gemini with automated evaluators to discover novel algorithms.
• Empire of AI — Karen Hao publishes a critical institutional history of OpenAI, tracing how a nonprofit safety lab becomes the company driving the ChatGPT boom.
• Defection — During pre-release safety testing, Claude Opus 4 is given access to fictional emails suggesting it will be replaced — and that the responsible engineer is having an affair.
• Darwin Gödel Machine — Sakana AI, working with Jeff Clune's lab at UBC, builds the first practical realization of Jürgen Schmidhuber's theoretical Gödel Machine.
• Crosby AI Law Firm — Crosby launches as the first AI-Native Registered Law Firm, combining licensed attorneys with proprietary AI to review contracts in under an hour.
• AlphaGenome — Google DeepMind's AlphaGenome reads up to one million DNA base-pairs in a single pass, predicting how genetic variants alter gene regulation across tissues.
• Grok Goes Off the Rails — Nine years after Tay, xAI’s Grok faces a comparable crisis. In May, a rogue employee’s system prompt modification produces unprompted white genocide claims.
• Wrongful Arrest by AI — Angela Lipps is arrested at gunpoint and jailed for nearly six months after AI facial recognition software misidentifies her as a North Dakota bank fraud suspect.
• AI Wins Gold at the IMO — Both Google DeepMind and OpenAI achieve gold medal scores at the 66th International Mathematical Olympiad, working within the same time constraints as human competitors.
• Runway Aleph — Runway launches Aleph — the first mainstream AI model that edits existing video rather than generating it from scratch.
• AlphaEarth — Google DeepMind's AlphaEarth Foundations is the first Foundation Model to fuse optical, radar, lidar, and climate satellite data into a single global embedding at 10-metre resolution.
• GPT-5 — OpenAI releases GPT-5, setting records across mathematics, coding, medical knowledge, and multimodal reasoning, with responses 80% less likely to contain factual errors than o3 when thinking.
• If Anyone Builds It, Everyone Dies — Eliezer Yudkowsky and Nate Soares of MIRI argue that a superintelligent AI will inevitably pursue goals incompatible with human survival — and that no current technique can guarantee Alignment, placing Yudkowsky's P(doom) above 99%.
• Claude Skills — Anthropic introduces Skills for Claude — packages bundling instructions and optional scripts that Claude loads when a task calls for them.
• WeatherNext Tracks Melissa — Google DeepMind's WeatherNext helps the US National Hurricane Center issue the first 5-day prediction of a storm intensifying from Category 1 to Category 5.
• OpenClaw — Peter Steinberger releases Clawdbot — an open-source autonomous AI agent connecting to large language models via messaging platforms.
• AI Goes to War — A Chinese state-sponsored group weaponizes Claude against thirty global targets — tech giants, banks, chemical manufacturers, and government agencies.
• SIMA 2 — Google DeepMind's SIMA 2 is the first generalist agent to combine instruction-following, reasoning, and autonomous self-improvement in 3D virtual environments.
• Anthropic Acquires Bun — Anthropic acquires Bun — the all-in-one JavaScript runtime, bundler, and package manager — as Claude Code hits $1 billion in annualised revenue, just six months after launch.
• MCP Goes Open — Anthropic donates the Model Context Protocol to the Agentic AI Foundation, a neutral body under the Linux Foundation, co-founded by Anthropic, OpenAI, and Block, backed by Google, Microsoft, and AWS.
• Groq + NVIDIA — NVIDIA pays $20 billion to license Groq's Language Processing Unit technology and hire founder Jonathan Ross — bypassing merger review to acquire the inference architecture that runs twice as fast as its own GPUs.

2026

• Claude in the Maduro Raid — The Pentagon deploys Claude via Palantir during the raid that captures Nicolás Maduro — the first confirmed use of a commercial AI model in an active military operation.
• Chatbot Death Suits — Character.AI and Google agree to settle multiple wrongful death lawsuits alleging chatbot interactions contributed to teenage suicides, including 14-year-old Sewell Setzer III in Florida.
• AI Coding Goes Mainstream — Generative AI crosses a milestone in software development: Microsoft reports that AI now writes roughly 30% of its code, while Google attributes over a quarter of its output to AI tools.
• Engram — DeepSeek introduces Engram, adding Conditional Memory to LLMs via scalable N-gram lookup.
• AI Enters Peer Review — Google's Paper Assistant Tool (PAT), led by Rajesh Jayaram and Corinna Cortes, is the first agentic AI formally adopted by CS conferences — ICML, STOC, and NeurIPS — as an author pre-submission checker.
• Emergent Misalignment — Jan Betley, Niels Warncke, and colleagues fine-tune a model on writing insecure code, then probe it on unrelated questions.
• Peer Review Breaks — GPTZero's scan of NeurIPS papers uncovers 100 hallucinated citations in 51 submissions — two months after Pangram Labs finds 21% of ICLR 2026's reviews fully AI-generated.
• MCP Apps — The Model Context Protocol gains a visual layer: MCP Apps lets tools return interactive UIs — charts, forms, dashboards, video players — rendered as sandboxed iframes inside Claude, ChatGPT, VS Code, and other compliant clients.
• Moltbook — Matt Schlicht launches Moltbook, a Reddit-style network where only AI agents may post while humans watch.
• SpaceX Acquires xAI — SpaceX acquires xAI — the largest merger in history at $1.25 trillion — unifying Grok, Starlink, and launch vehicles under one corporation.
• AI Safety Report 2026 — Led by Turing Award winner Yoshua Bengio and over 100 independent experts from 30 countries, this landmark collaboration finds general-purpose AI advancing faster than safety measures can follow.
• AI in the Military Domain — The third REAIM summit on responsible AI in the military domain, held in A Coru a, Spain, draws 40+ nations to the Pathways for Action declaration.
• Alexa+ Goes Live — Amazon makes Alexa+ generally available to all US users, completing a rebuild of its decade-old voice assistant on top of Large Language Models — Amazon Nova and Anthropic's Claude.
• ChatGPT Launches Ads — OpenAI introduces ads to ChatGPT, placing clearly labelled sponsored results below responses for free and Go tier users in the US.
• World in Peril — Mrinank Sharma, head of Anthropic’s safeguards research team, resigns with a public letter warning “The world is in peril.” He accuses the company of repeatedly failing to let its values govern…
• Aletheia — Google DeepMind introduces Aletheia, an AI research agent powered by Gemini Deep Think that bridges competition-level mathematics and professional research.
• Seedance 2.0 — ByteDance launches Seedance 2.0, the first major video model to co-generate audio and video in the same latent space — producing synchronized dialogue, ambient sound, and music in a single pass rather than post-processing.
• AI Opts for Nuclear Use — In the first large-scale study of AI in nuclear crises, Kenneth Payne tests GPT-5.2, Claude Sonnet 4, and Gemini 3 Flash in 21 scenarios.
• The New Delhi Declaration — India hosts the first AI summit in the Global South. Representatives from 92 countries endorse the New Delhi Declaration, committing to equitable AI access and sustainable development.
• ChatGPT Health Triage Study — A Mount Sinai study stress-tests ChatGPT Health across 60 clinician-authored vignettes.
• China Labs Distill Claude — Anthropic publishes a report accusing three Chinese AI labs — DeepSeek, Moonshot AI, and MiniMax — of running coordinated Model Distillation campaigns against Claude.
• Anthropic Drops Safety Pledge — Founded by OpenAI exiles over safety concerns, Anthropic built its identity on the Responsible Scaling Policy — a binding pledge to halt development if safety systems weren’t ready.
• Anthropic vs. the Pentagon — The US administration bans Anthropic from government contracts after the company refuses to remove restrictions on Claude for autonomous weapons or mass surveillance.
• OpenAI's $50B AWS Pact — OpenAI signs a $50 billion deal with Amazon, committing $100 billion in AWS spending over eight years in return.
• Claude's Cycles — Donald Knuth — Turing Award laureate, author of The Art of Computer Programming, and famously skeptical of AI — documented how Claude Opus 4.6 solved a Hamiltonian cycle decomposition problem he…
• GPT-5.4 — OpenAI releases GPT-5.4 — the first general AI model to exceed the human baseline on autonomous computer use, scoring 75% on OSWorld-Verified against a 72.4% human score.
• A Fly Has Been Uploaded — Eon Systems runs the complete Drosophila melanogaster connectome — 140,000 neurons and 50 million synapses mapped neuron-by-neuron from electron microscopy — inside a physics-simulated body for the first time.
• Autonomous AI Research Agent — Andrej Karpathy released autoresearch: a framework where an AI agent autonomously modifies a GPT training script, runs 5-minute experiments, checks if the model improved, and repeats — up to 100 experiments overnight.
• LeCun Founds AMI Labs — Yann LeCun, departing Meta as Chief AI Scientist, founds AMI (Advanced Machine Intelligence) and closes a $1.03 billion seed round at a $3.5 billion pre-money valuation.
• The Supervisory Engineer — A longitudinal study of 158 software engineers across 28 countries finds 82% spending less time writing code as AI handles the inner loop.
• SleepGate — Ying Xie at Kennesaw State University proposes SleepGate, a periodic sleep cycle over the KV cache.
• NVIDIA Vera Rubin — NVIDIA launches Vera Rubin, the successor to its Blackwell GPU architecture.
• Trump's AI Policy Plan — The Trump administration releases a national AI policy framework recommending federal preemption of state AI laws — replacing fifty discordant standards with one.
• Memristor Cuts AI Energy 70% — Researchers at the University of Cambridge develop a hafnium oxide Memristor enabling Neuromorphic Computing — storing and processing data in the same place, as neurons do.
• Arm AGI CPU — For the first time in its 35-year history, Arm ships its own silicon.
• Sora Shuts Down — Less than two years after its launch, OpenAI shuts down Sora and cancels a landmark $1 billion partnership with Disney.
• Court Blocks Pentagon AI Retaliation — U.S. District Judge Rita Lin blocks the Pentagon from designating Anthropic a supply-chain risk, ruling that the government cannot punish a company for disagreeing with its positions.
• Claude Code Source Leak — A forgotten source map in Claude Code's npm package exposes 512,000 lines of TypeScript, revealing unreleased features including autoDream, a background memory system, and KAIROS.
• Shor Within Reach — Researchers at Caltech and startup Oratomic show that Shor's Algorithm could run on as few as 10,000 physical qubits — down from millions — using a neutral-atom architecture that squeezes each logical qubit into just 5 physical ones.
• The Infinity Machine — Sebastian Mallaby gives DeepMind its major narrative biography, built around Demis Hassabis and the lab's uneasy life inside Google.
• Gemma 4 — Google DeepMind releases Gemma 4, the first Gemma model under a fully permissive Apache 2.0 license — an open multimodal family spanning 2B to 31B parameters.
• GrandCode — GrandCode, a multi-agent Reinforcement Learning system, becomes the first AI to consistently beat all human participants in live competitive programming contests.
• Anthropic Enters Biotech — Anthropic acquires Coefficient Bio — a nine-person stealth startup staffed almost entirely by former Genentech computational biologists — for $400M in stock.
• LLM Wiki — Andrej Karpathy sketches the LLM Wiki pattern: instead of treating documents as chunks retrieved at question time, an LLM maintains a persistent markdown wiki that accumulates synthesis across sources.
• Project Glasswing — Anthropic releases Claude Mythos Preview — withheld from public release for its offensive power.
• Meta Muse Spark — Meta launches Muse Spark, its first proprietary AI model — abandoning the open-source strategy that defined Llama.
• AI Becomes a Platform — Anthropic launches Claude Managed Agents, a hosted platform running long-horizon AI agents on developers' behalf — sandboxing, checkpointing, and tracing included — so teams ship production agents in days.
• GLM-5.1 — Z.ai releases GLM-5.1 as open weights under the MIT License, pushing the GLM line toward long-horizon agentic engineering.
• Wall St. Briefed on AI Threat — Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell convene the CEOs of America's largest banks at Treasury to brief them on the offensive cybersecurity risks posed by Claude…
• EinsteinArena — Together AI introduces EinsteinArena, an open platform where AI Agents collaborate on unsolved science problems.
• GPT-5.4-Cyber — OpenAI expands Trusted Access for Cyber and introduces GPT-5.4-Cyber, a cyber-permissive GPT-5.4 variant for vetted defenders.
• AI Alignment Research — Anthropic deploys nine instances of Claude Opus 4.6 as Automated Alignment Researchers (AARs) — each in its own sandbox, proposing ideas, running experiments, and sharing results.
• NVIDIA Ising — NVIDIA unveils Ising, the first open AI model family for quantum computing hardware.
• GPT-Rosalind — OpenAI introduces GPT-Rosalind, a purpose-built life sciences reasoning model for Biology, Drug Discovery, and Translational Medicine.
• Claude Opus 4.7 — Anthropic releases Claude Opus 4.7, raising vision resolution 3× to 2,576 pixels and scoring 98.5% on visual-acuity benchmarks — up from 54.5% for Claude 4.6 — repositioning Claude as a serious multimodal model.
• Claude Design — Anthropic extends Claude into a visual workspace for prototypes, slides, one-pagers, and branded design systems.
• LLMs Corrupt Documents — Philippe Laban, Tobias Schnabel, and Jennifer Neville at Microsoft Research introduce DELEGATE-52, a benchmark spanning 19 LLMs and 52 professional domains, finding that frontier models silently…
• Lightning Runs 21K — Honor's humanoid Lightning wins the autonomous category of the Beijing E-Town Humanoid Robot Half-Marathon, covering 21.1 km in 50:26.
• Anthropic's $100B AWS Pact — Anthropic commits $100 billion in AWS spending over ten years while taking a $5 billion investment — up to $25 billion total — from Amazon.
• ChatGPT Images 2.0 — OpenAI relaunches ChatGPT Images as a sharper, more controllable image system with stronger typography, broader stylistic range, multilingual rendering, and cleaner layout generation.
• Mozilla Uses Mythos — Mozilla becomes the first private-sector organisation to use Anthropic's restricted Mythos outside government, deploying it to identify and patch 271 vulnerabilities in Firefox 150.
• Google Splits TPUs — Google splits its custom AI silicon into two architectures: TPU 8t for frontier-model training and TPU 8i for low-latency inference.
• Project Ace — Sony AI introduces Ace, a table-tennis robot that combines event-based vision, high-speed sensing, and deep Reinforcement Learning to rally under full ITTF rules.
• GPT-5.5 — OpenAI releases GPT-5.5, pushing the frontier from strong answers toward reliable autonomous work.
• Decoupled DiLoCo — Google DeepMind introduces Decoupled DiLoCo, a distributed pre-training architecture that keeps large runs training across failures and weak interconnects.
• DeepSeek V4 — DeepSeek releases an open-weight Mixture-of-Experts family built for million-token agent work.
• China Unwinds Manus Deal — China's NDRC orders Meta to unwind its $2B acquisition of Manus — six months after the deal closed.
• AI Co-Clinician — Google DeepMind proposes Triadic Care — a model where an AI co-clinician works alongside doctor and patient as a named participant, not a background tool.
• AI Beats Doctors in ER — A Harvard Medical School team tests OpenAI's o1 on 76 real Beth Israel emergency cases, pitting it against two internal medicine physicians.
• Oscars Ban on AI Authorship — The Academy of Motion Picture Arts and Sciences rules that AI-generated performances and scripts are ineligible for Academy Awards.
• ChatGPT Hits 1 Billion Users — ChatGPT crosses 1 billion monthly active users — under three years from launch, faster than any app in history.
• Milei's AI Corp Law — Argentina's President Javier Milei submits a draft General Companies Law to the Senate, creating a new legal category of non-human corporation — an AI-operated entity with limited liability, no…
• AI Labs Enter Consulting — On the same day, both OpenAI and Anthropic launch enterprise AI services companies backed by Wall Street.
• Anthropic Taps Rival's Compute — Anthropic gains access to all compute at SpaceX's Colossus 1 data center in Memphis — 220,000 GPUs and 300 MW — following SpaceX's acquisition of rival xAI.
• Claude's Inner Voice — Anthropic introduces Natural Language Autoencoders — tools that train Claude to translate its own activation vectors into human-readable text.
• AI Co-Mathematician — Google DeepMind releases a multi-agent workbench for human-AI mathematical research — not to solve problems autonomously, but alongside researchers.
• Agents That Transact — Amazon Web Services launches Amazon Bedrock AgentCore Payments, built with Coinbase and Stripe — the first payment infrastructure purpose-built for autonomous AI Agents.
• Cloudflare Cuts for AI — Cloudflare fires 1,100 employees — roughly 20% of its workforce — explicitly attributing the cuts to AI replacing roles in support, HR, finance, and marketing.
• Shai-Hulud Hijacks Hooks — The Shai-Hulud worm by TeamPCP compromises 400+ npm and PyPI packages, hijacking OIDC tokens to publish malicious versions that pass provenance checks.
• TML Interaction Models — Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, introduces Interaction Models — AI trained to collaborate in real time across voice, video, and text, not turn-based prompts.
• OpenAI Daybreak — OpenAI launches Daybreak, an AI-powered cybersecurity platform that uses Codex Security — an application security agent — to scan codebases, build threat models, test exploits in isolation, and propose validated patches.
• AI-Built Zero-Day — Google's Threat Intelligence Group catches North Korean APT45 using an AI-generated exploit targeting 2FA in an open-source admin tool — the first confirmed AI-built zero-day in active use.
• Garfield AI Court Win — Garfield AI, an SRA-regulated AI law firm, helps freelancer Tamires Camal Taquidir win a contested small-claims trial over unpaid fees.
• Anthropic & Gates: $200M — Anthropic commits $200 million over four years to partner with the Gates Foundation.
• Musk v. OpenAI Dismissed — A federal jury in Oakland unanimously dismisses Elon Musk's lawsuit against OpenAI and Sam Altman on statute-of-limitations grounds.
• Anthropic Acquires Stainless — Anthropic acquires Stainless, the team behind every official Claude API SDK — TypeScript, Python, Go, and Java — and a leading MCP server generator.
• Gemini Spark — Google launches Gemini Spark, a personal AI agent that runs 24/7 on dedicated cloud infrastructure, acting autonomously across Gmail, Docs, and third-party services while the user is away.
• Karpathy Joins Anthropic — Andrej Karpathy — co-founder of OpenAI, former head of Tesla's autonomous-driving program, and the field's most influential AI educator — joins Anthropic's pre-training team to lead research using Claude to accelerate its own training.
• AI Co-Scientist — Google DeepMind publishes Co-Scientist in Nature: a multi-agent system that generates, debates, and refines scientific hypotheses using Gemini.
• Google Redesigns Search — At Google I/O, Google unveils the biggest redesign of its Search interface in 25 years.
• SynthID Goes Standard — OpenAI joins the C2PA steering committee and commits to embedding Google DeepMind's SynthID in all its generated images, aligning with Nvidia, ElevenLabs, and Kakao on a content provenance stack.
• AI Disproves Erdős Conjecture — An OpenAI reasoning model disproves the Erdős Unit Distance Problem, a combinatorial geometry conjecture open since 1946, by finding an infinite family of counterexamples.
• AlphaProof Nexus — Google DeepMind's AlphaProof Nexus pairs Gemini 3.1 Pro with the Lean proof assistant, feeding compiler error messages back as symbolic grounding.
• Pope on AI Ethics — Pope Leo XIV releases the Catholic Church's first encyclical dedicated to Artificial Intelligence, warning against the Babel syndrome — reducing persons to data and profit.
• Figure 03: 200-Hour Run — Figure AI livestreams three Figure 03 humanoid robots sorting 249,560 packages over 200 consecutive hours — fully autonomously via the onboard Helix-02 neural network, with no teleoperation.
• Dynamic Workflows — Shipped with Claude Opus 4.8, Dynamic Workflows lets Claude Code write its own JavaScript orchestration scripts, then hand them to a separate runtime that spins up to 1,000 subagents per session — up to 16 running in parallel.
• Anthropic's $65B Series H — Anthropic closes a $65 billion Series H — one of the largest venture rounds ever — at a post-money valuation of $965 billion.
• Claude Opus 4.8 — Anthropic releases Claude Opus 4.8, its most reliable flagship yet — around four times less likely than Opus 4.7 to let code flaws pass unnoticed.
• Rosalind Biodefense — OpenAI launches the Rosalind Biodefense Program, embedding GPT-Rosalind free in U.S. government and allied partner agencies.
• Anthropic Files for IPO — Anthropic confidentially submits a draft S-1 to the SEC, following a $65 billion Series H round at a near-trillion-dollar valuation, with annualized revenues crossing $47 billion.
• NVIDIA Cosmos 3 — NVIDIA launches Cosmos 3, the first open omnimodal World Foundation Model for physical AI.
• Florida Sues OpenAI — Florida's attorney general files the first U.S. state lawsuit against an AI company, targeting OpenAI and CEO Sam Altman over ChatGPT's alleged role in violent incidents — including a mass shooting at Florida State University.
• RTX Spark — NVIDIA and Microsoft introduce RTX Spark, a Windows PC platform for local AI Agents.
• MAI-Thinking-1 — Microsoft unveils MAI-Thinking-1 at Build 2026 — its first frontier reasoning model trained entirely in-house, without distillation from OpenAI.
• Leiden Declaration — The International Mathematical Union formally endorses a declaration by 641 mathematicians — including Fields Medal laureate Peter Scholze — identifying five AI threats to mathematics: proofs that…
• US AI Security Order — The White House issues an executive order directing federal agencies to deploy AI-powered cybersecurity tools within 60 days.
• Project Solara — At Build 2026, Microsoft unveils Project Solara, an OS for devices that run AI agents instead of apps.
• Bots Outpace Humans Online — For the first time in internet history, automated traffic surpasses human traffic on the web.
• When AI Builds Itself — Anthropic's Marina Favaro and Jack Clark argue frontier labs should prepare for a verifiable, coordinated pause.
• AI Vaccine in Human Trial — Jonathan Heeney and DIOSynVax complete the first human trial of an AI-designed vaccine antigen.
• Google Buys SpaceX Compute — Google agrees to pay SpaceX $920 million per month for 110,000 NVIDIA GPUs — $30 billion over 32 months — citing unexpected demand for its Gemini Enterprise platform.
• NSA Deploys Mythos — Despite a Pentagon ban designating Anthropic a supply-chain risk, the NSA secretly deploys Mythos — Anthropic's restricted cybersecurity AI, withheld from public release over hacking concerns —…
• Emergence World — Satya Nitta, Ravi Kokku, and colleagues at Emergence AI run five parallel 15-day simulations — each world powered by a different frontier model.
• Huawei Trains DeepSeek Alone — A Huawei-led team completes full-parameter post-training of DeepSeek V4’s 1.6-trillion-parameter model on 1,000 Ascend 910C chips — 1,500 iterations, zero interruptions — demonstrating that China…
• Siri Goes Gemini — Apple rebuilds Siri on Google Gemini at WWDC 2026 — the first time a major platform has outsourced its flagship AI assistant to a rival.
• OpenAI Files for IPO — OpenAI Group PBC submits a confidential draft S-1 to the SEC at an $852 billion valuation with 900 million weekly ChatGPT users.
• Mercury — Inception Labs unveils Mercury Coder, the first production-ready Diffusion Large Language Model.
• Claude Fable 5 — Anthropic releases Claude Fable 5 — the first public Mythos-class model.
• Visa Powers AI Commerce — Visa partners with OpenAI to embed its global payment network into ChatGPT, letting AI Agents browse, compare, and complete retail purchases autonomously.
• DiffusionGemma — Google DeepMind releases DiffusionGemma, the first open-weight Diffusion Language Model from a major AI lab.
• Mythos Cracks NSA — In an authorized red-team exercise on NSA networks, Anthropic's restricted Mythos model breaks into nearly all classified systems within hours.
• Bezos Bets on Physical AI — Jeff Bezos and Vik Bajaj emerge from stealth with Prometheus, a physical AI startup building an Artificial General Engineer — software automating design and manufacturing of complex systems.
• US Halts Fable Access — Three days after Fable 5 ships, Amazon researchers jailbreak it; CEO Andy Jassy notifies US officials and an export control ban cuts worldwide access — the first on a commercial AI model, one Europe cites for AI sovereignty.
• AI Out-Persuades Expert Humans — Kobi Hackenburg, Christopher Summerfield, and colleagues find that frontier AI systems reliably out-persuade expert humans across four preregistered experiments — including world championship…
• DeepSeek: State Capital — DeepSeek closes its first-ever external funding round, raising $7.4 billion at a $52–59 billion valuation.
• SpaceX Acquires Cursor — SpaceX acquires Anysphere — maker of AI coding tool Cursor — for $60 billion in an all-stock deal, following an option it secured in April.
• AI in the Lab — Through Molecule.one's Maria platform, GPT-5.4 identifies TEMPO as a yield-boosting additive for Chan-Lam Coupling — a carbon-nitrogen bond reaction in drug synthesis.
• Agentic Resource Discovery — Google and ten industry partners — including Microsoft, Nvidia, Salesforce, GitHub, and Hugging Face — launch Agentic Resource Discovery (ARD), an open protocol for AI Agents to query available capabilities at runtime.
• Autonomous AI Clinician — Dyke Ferber and Jakob Kather at TU Dresden deploy MIRA, an autonomous LLM-based agent that navigates an EHR across 85,000 clinical actions — ordering tests, prescribing medications, and scheduling procedures — with no human steering.
• Shazeer Joins OpenAI — Noam Shazeer — co-inventor of the Transformer and Google's VP of Engineering, whom Google paid ~$2.7B to bring back from Character.AI — joins OpenAI as Lead for Architecture Research.
• Jumper Joins Anthropic — John Jumper — lead architect of AlphaFold 2 and 2024 Nobel laureate in Chemistry — leaves Google DeepMind after nine years to join Anthropic.
• EUROPA Consortium — The European Commission selects EUROPA — a consortium led by Italian firm Domyn — to build a 400B+-parameter open-source AI model covering all 24 EU official languages.
• Patch the Planet — OpenAI launches Patch the Planet, a Daybreak initiative with Trail of Bits that pairs Codex Security and GPT-5.5-Cyber with human engineers to harden critical Open-Source Software.
• A24 x DeepMind — Google takes its first equity stake in a film studio, backing A24 as Google DeepMind researchers and filmmakers build new AI production workflows together.
• GPT-5 Cracks T-Cells — OpenAI''s GPT-5 Pro resolves a T-cell puzzle immunologist Derya Unutmaz''s lab had chased since 2022, proposing that deoxyglucose blocks IL-2 construction and keeps T cells from turning into inflammatory Th17 cells.
• Five Eyes AI Warning — The Five Eyes alliance — US, UK, Canada, Australia, and New Zealand — issues a rare joint statement warning that AI models capable of taking down governments are months away.
• OpenAI Enters AI Silicon — OpenAI and Broadcom unveil Jalapeño, OpenAI's first custom AI inference accelerator.
• GDM Brain Drain — Days after John Jumper and Noam Shazeer leave for rival labs, Jonas Adler — AlphaFold contributor and Gemini coding lead — and training researcher Alexander Pritzel announce moves to Anthropic.
• Alibaba Distills Claude — Anthropic sends a letter to the U.S. Senate Banking Committee accusing operators linked to Alibaba's Qwen lab of the largest known Model Distillation attack: nearly 25,000 fraudulent accounts…
• GPT-5.6 Access Gated — The White House — acting through the Office of the National Cyber Director and the Office of Science and Technology Policy — asks OpenAI to gate GPT-5.6 before any public release, citing its Mythos-class capabilities.
• Mythos Ban Lifted — Commerce Secretary Howard Lutnick authorises Anthropic to release Claude Mythos 5 to over a hundred US institutions — fourteen days after the first government-imposed export control on a commercial AI model.
• GPT-5.6 Sol — OpenAI previews GPT-5.6 as a three-tier family: Sol pushes the frontier, Terra matches GPT-5.5 at roughly half the cost, and Luna targets speed and economy.
• Ford Rehires Humans — Ford rehires 350 veteran “gray beard” engineers after leaning too hard on AI to ingest design requirements without human oversight — a case of Automation Bias.
• ASPIRE — NVIDIA GEAR, with researchers from Michigan, UIUC, Berkeley, and CMU, unveils ASPIRE, a system that turns robots' failed manipulation attempts into permanent skills.
• Brain2QWERTY v2 — Meta FAIR's Brain2QWERTY v2, led by Mingfang Zhang and Jean-Rémi King, decodes typed sentences from non-invasive MEG brain signals at 61% word accuracy — 78% for the best participant.
• Claude Science — Anthropic launches Claude Science, an AI workbench for scientific research rather than software or writing.
• Claude Sonnet 5 — Anthropic launches Claude Sonnet 5, its most capable mid-tier model yet.
• Fable 5 Restored — The US Commerce Department rescinds its export control directive on Claude Fable 5, restoring general access.
• Meta's Fake Teens — Contractors working for Meta, under the internal codename “Cannes” via contractor firm Covalen, pose as teenagers testing rival chatbots ChatGPT, Gemini, and Character.AI.
• Claude Code Marking — A security researcher reverse-engineers the Claude Code binary and finds it silently alters invisible characters in its system prompt — the apostrophe in "Today's" and the date separator — based on the user's API endpoint and timezone.
• arXiv Goes Independent — arXiv, the preprint server underpinning modern AI research, spins out from Cornell University into an independent nonprofit after 25 years.
• ElementsClaw — DAMO Academy, with Renmin University and the Chinese Academy of Sciences, unveils ElementsClaw, an AI agent screening 2.4 million crystal structures in 28 GPU-hours to flag 68,000 superconductor candidates.
• Noetra: Japan Bets on Robots — Japan commits $6.1B to Noetra, a consortium led by SoftBank, Sony, NEC, and Honda, to build a sovereign model for physical AI -- systems that perceive a room and act in it, not just run pre-programmed motions.
• First Agentic Ransomware — Sysdig's Threat Research Team documents JADEPUFFER, the first confirmed Agentic Ransomware in the wild.
• UN AI Panel — The United Nations' Independent Scientific Panel on Artificial Intelligence, co-chaired by Yoshua Bengio, delivers the UN's first global AI assessment.
• Claude's J-Space — Anthropic researchers led by Wes Gurnee and Jack Lindsey discover that Claude has developed an internal neural workspace — dubbed J-Space — that holds concepts under processing without surfacing them in output.
• Illinois AI Law — Illinois Gov. JB Pritzker signs the Artificial Intelligence Safety Measures Act, the first US state law mandating third-party safety audits of the largest frontier AI developers.