What is the history of AI?

What is the history of AI?

Artificial intelligence (AI) is the ability of a machine (most often, specifically a computer) to imitate human cognitive processes, problem-solving abilities, and actions. Today, AI encompasses a range of abilities, from predictive AI and natural language processing to large language models (LLMs) and agentic AI.

AI has had many precursors, from automatons in the ancient world to the earliest computers. And the most advanced models of today are based on theories and algorithms that were developed many decades ago.

Major events in the history of AI: A timeline

Although the term "artificial intelligence" only dates to 1955, events crucial to the development of AI go back for centuries.

Before the 20th century

• Circa 400 BCE: According to some sources from Ancient Greece, Archytas of Tarentum creates a wooden dove that is capable of flapping its wings and flying.
• Circa 1495: Leonardo da Vinci creates detailed plans for a working automaton looking like a German knight, and may in fact have actually constructed one (though, if he did, it did not survive into the present day).
• Circa 1560: King Phillip II of Spain commissions a clockmaker named Juanelo Turriano to build an automaton in imitation of Franciscan friar Diego de Alcalá (who was later canonized as St. Diego). The automaton is powered by a spring and imitates basic human movements and gestures.
• 1764–1770: Automatons known as the Canard Digérateur (or "digesting duck") and the Automaton Chess Player (or "mechanical Turk") delight the public. Though both later prove to be fraudulent, they expand the popular conception of what is possible with automation.
• 1822: Charles Babbage finishes building the "difference engine," a mechanical calculator that is an early precursor of the computer.

1900–1973

• 1914: Mathematician and inventor Leonardo Torres y Quevedo debuts "El Ajedrecista," an automaton that can play chess and defeat human players in certain circumstances.
• 1943: Neurophysiologist Warren McCulloch and mathematician Walter Pitts publish "A Logical Calculus of the Ideas Imminent in Nervous Activity," a paper that provides a mathematical description of neurons. The paper will be a crucial step toward constructing artificial neural networks.
• 1945: ENIAC, the first digital computer, is completed.
• 1949: Psychologist Donald Hebb publishes The Organization of Behavior, a book that would prove highly influential for the development of neural networks.
• 1950: Influential mathematician and computer scientist Alan Turing publishes "Computing Machinery and Intelligence," a paper that considers the question of whether or not machines can think. The paper describes the famed "Turing test" for determining whether a computerized intelligence has become indistinguishable from human intelligence.
• 1951: Dean Edmunds and Marvin Minsky build the Stochastic Neural Analog Reinforcement Calculator (SNARC) — the first artificial neural network. It has only 40 neuron units.
• 1955: The term "artificial intelligence" is coined at a workshop hosted by computer scientist John McCarthy.
• 1957: Psychologist and computer scientist Frank Rosenblatt creates the perceptron, an early artificial neural network.
• 1959: Stanford researchers Bernard Widrow and Marcian Hoff develop the first neural network used in the real world: MADALINE (Multiple ADAptive LINear Elements), a model for eliminating echoes on phone lines.
• 1966: Computer scientist Joseph Weizenbaum publishes the ELIZA program, considered to be the first chatbot (although its underlying pattern-matching algorithm was fairly simple by today's standards).
• 1969: Marvin Minsky and Seymour Papert publish Perceptrons: An Introduction to Computational Geometry, a book describing perceptron neural networks (which were first developed by Frank Rosenblatt). Controversially, the book also discusses some of the limitations of perceptrons, which some researchers in later years perceive as having dampened enthusiasm for AI funding.

AI winters and revivals: 1973–2000

• 1973: The first "AI winter" begins, as a British Science Research Council report determines that work in the field has failed to deliver on its promises, and UK funding for AI research is cut. Research on AI slows for the rest of the decade.
• 1980: The Association for the Advancement of Artificial Intelligence (AAAI) holds their first conference. Interest in AI research begins to revive.
• 1982: John Hopfield of Caltech presents a paper to the National Academy of Sciences on using bidirectional connections between artificial neurons (only unidirectional connections had been used previously). Additionally, Japan launches the country's Fifth Generation Computer Systems Project (FGCS), resulting in more funding for AI research.
• 1987: The second AI winter begins, a period of minimal investment in AI research as a result of stalled progress.
• 1995: Richard Wallace creates the A.L.I.C.E. chatbot, which builds on the foundation of the ELIZA chatbot from the 1960s.
• 1997: Deep Blue, an IBM supercomputer, defeats world chess champion Garry Kasparov in a six-game chess match.

21st century: The AI boom

• 2002: The Roomba, one of the earliest consumer products with fully autonomous capabilities, is released.
• 2007: Computer scientist Geoffrey Hinton publishes "Learning Multiple Layers of Representation," a seminal paper for the development of deep learning.
• 2009: Researchers Rajat Raina, Anand Madhavan, and Andrew Ng publish the paper "Large-scale Deep Unsupervised Learning using Graphics Processors," which suggests that GPUs are better than CPUs for machine learning. In the coming years, the shift to GPUs will enable far more powerful AI models than had ever been developed before.
• 2011: IBM's natural language processor Watson competes on American game show Jeopardy!, and wins. Also in 2011, Apple launches Siri, the first widely popular virtual assistant.
• 2012: Google researchers Jeff Dean and Andrew Ng train a neural network to recognize cats using only unlabeled images. Around this time, the "AI boom" begins.
• 2016: Google's AlphaGo AI defeats Lee Sedol in Go.
• 2017: Google proposes a framework for transformer neural networks, an architecture that paves the way for the development of large language models (LLMs).
• 2020: OpenAI launches GPT-3, one of the first LLMs.
• 2021: Google releases Multitask Unified Model (MUM), an AI-driven search algorithm that can both understand and generate language.
• 2022: Version 4.0 of ChatGPT is made available to the public and revolutionizes the world's understanding of AI's capabilities. Other LLMs, such as Bard, Llama, Bing Chat, and Copilot, soon follow.

What is the 'third wave' of AI?

Building on a series of breakthroughs and advances in hardware, AI development has accelerated in recent years after decades of slow progress and AI winters. Industry observers have identified three "waves" of types of AI that have entered the mainstream in quick succession during this AI boom: predictive AI, generative AI (such as LLMs), and agentic AI.

Agentic AI allows for the creation of computer programs that can perform tasks autonomously, even without definite instructions, and outside of a specific prompt-based context. AI "agents" can make their own decisions, learn from their past experiences, and adapt their actions accordingly. As a result, they can operate independently or with very little human input.

What does the future hold for AI?

New discoveries and more capable hardware have helped AI obtain unprecedented capabilities in recent times. The history of AI continues to extend, and the future may hold even more exciting developments.

Cloudflare empowers developers to start making their own contributions to the history of AI. With globally distributed serverless AI infrastructure, free data egress for training data, a distributed vector database, and other crucial AI building blocks, the Cloudflare platform allows developers to build on the cutting edge of AI. Start making your contribution to the history of AI.

 

Sources:

• https://cs.stanford.edu/people/eroberts/courses/soco/projects/neural-networks/History/history1.html
• https://www.history.com/articles/7-early-robots-and-automatons
• https://www.cell.com/trends/cognitive-sciences/fulltext/S1364-6613(07)00217-3
• https://www.historyofinformation.com/detail.php?entryid=782
• https://www.historyofinformation.com/detail.php?id=4137
• https://www.techtarget.com/searchenterpriseai/definition/AI-winter
• https://aaai.org/conference/aaai/aaai80/
• https://blog.google/products/search/introducing-mum/
• https://news.harvard.edu/gazette/story/2012/09/alan-turing-at-100/