The History of AI | A Brief Timeline of Development

Today, it’s hard to go five minutes without encountering artificial intelligence (AI) in some shape or form—this technology powers our phones, shapes how we work, and underlies much of the content we consume. Given its galactic rise in popularity in recent years, it’s easy to consider AI a modern phenomenon; however, its roots can actually be traced back to the mid-1900s.

Read on to learn more about the key innovators and innovations that have brought about AI as we know it today. And if you find yourself wondering if what you’re reading online is written by a human these days, try QuillBot’s AI Detector!

A brief history of AI

With a technology as diverse as AI, it’s easy to get bogged down in the details. To keep things simple, the table below highlights key eras, people, and breakthroughs contributing to the past and present of AI.

History of AI timeline

The sections below summarize major advancements in the field of AI.

AI in myth and fiction

The notion of intelligent machines has captivated people for millennia.

Ancient Greek mythology references automatons, or intelligent machines with agency and willpower, such as the bronze giant Talos. In literature, Jonathan Swift’s Gulliver’s Travels (1726) described a fictional machine called The Engine that can write books. The term “robot” originates from the 1920s play Rossum’s Universal Robots, written by Czech writer Karel Čapek.

While these are works of fiction, they captured people’s evolving ideas about machines and, alongside influential works in philosophy and mathematics, set the stage for concrete developments in AI in the 20th century.

1950s and 1960s: The golden age of early AI

1950: The Turing Test. Alan Turing, who is considered the father of modern computer science, wrote a paper in 1950 called “Computing Machinery and Intelligence.” In it, he addressed the question of whether machines think. Rather than answer this question directly, Turing proposed a thought experiment—he asked whether a machine can behave in a way that is indistinguishable from a human. He called this the “Imitation Game,” but it is now known as the Turing test. His paper hugely influenced how people think about AI.

1951: SNARC, the first neural network. Developed by cognitive scientist Marvin Minsky, SNARC (or Stochastic Neural Analog Reinforcement Calculator) was a physical realization of theoretical artificial neurons. Roughly the size of a grand piano, the machine used reinforcement learning to solve mazes, marking the first successful attempt to give a machine memory-based decision-making capabilities.

1955: The Logic Theorist. The Logic Theorist was a computer program developed by Allen Newell, Herbert A. Simon, and Cliff Shaw. Many consider it the first AI program, though some debated whether it was actually “intelligent.” The Logic Theorist could reason and use logic to prove mathematical theorems. It sparked the Cognitive Revolution—a movement to model the mind with machines.

1956: AI is named at the Dartmouth Workshop. Inspired by the growing interest in thinking machines, computer/cognitive scientist John McCarthy and colleagues Marvin Minsky, Claude Shannon, and Nathan Rochester decided to organize a conference to discuss this new field. McCarthy called the conference the “Dartmouth Summer Research Project on Artificial Intelligence”—the first time the term “AI” was used. It marked the birth of the field of artificial intelligence.

1957: The Perceptron. Psychologist Frank Rosenblatt developed the perceptron in 1957 and introduced it in 1958. It was a single-layer neural network that could learn to categorize items via supervised learning—that is, it updated connection weights between its artificial neurons to minimize errors. Rosenblatt was enthusiastic about the promise of his “connectionist” model. However, critics like Marvin Minsky highlighted its limitations, stalling AI progress for years.

1959: Arthur Samuel coins “machine learning.” Games like chess and checkers were an early benchmark of AI research. In 1959, Arthur Samuel published a paper called “Some Studies in Machine Learning Using the Game of Checkers.” This paper, which described a machine that could beat amateur checkers players, was the first to use the term “machine learning.” This term referenced a computer’s ability to solve a task without being explicitly programmed.

1966: The Chatbot ELIZA. ELIZA, considered by many to be the first chatbot (or chatterbot, as they were originally called), was created by MIT professor Joseph Weizenbaum. It could simulate conversation between human and machine by recognizing patterns in speech, but it could not pass the Turing test.

1970s: The first AI winter

The 1950s and 1960s saw many exciting advances, but early successes led to undue optimism about AI’s capabilities. Computers weren’t powerful enough to handle complex model architecture, leading to underwhelming results. Interest in AI fizzled out, and research funding was cut, leading to an AI winter where few advances were made.

1980s: Expert systems and machine learning

1980: The expert system R1 (or XCON). While rule-based “expert systems” emerged in the ’60s, they hit the mainstream in 1980 with R1 (later XCON). Developed for Digital Equipment Corporation (DEC), R1 used a massive library of “if-then” rules to automate complex hardware orders. By saving the company millions of dollars annually, it proved AI had massive commercial value, sparking a global wave of corporate investment.

1986: Backpropagation is applied to neural networks. Early neural networks like the perceptron were notoriously difficult to train, but the 1986 paper “Learning representations by back-propagating errors” changed everything. Authors Geoffrey Hinton, David Rumelhart, and Ronald Williams demonstrated how to efficiently train multi-layer (or “deep”) networks by “back-propagating” errors to adjust internal weights.

Alongside the influential book “Parallel Distributed Processing” by Rumelhart and James McClelland, Hinton et al.’s work rescued “connectionism” from the fringes of statistics, proving that deep learning was the future of AI. For his foundational role, Hinton would eventually be dubbed the “Godfather of AI.”

1990: CNNs and Handwritten Digit Recognition. Computer scientist Yann LeCun pioneered the use of convolutional neural networks (CNNs) to automate the reading of handwritten digits. Modeled after the human visual system, CNNs excelled at identifying spatial patterns, setting a new standard for image processing. This breakthrough provided one of the first high-stakes, practical applications for neural networks: sorting mail and processing bank checks with incredible accuracy.

Early 1990s: The second AI winter

Economic hardship, coupled with disappointment in the inflexibility of expert systems, led to a second AI winter in the early 1990s. Researchers struggled to produce products that could handle real-world complexities and function in uncontrolled environments.

Late 1990s–2010s: The rise of deep learning and popular AI

1997: Deep Blue beats a chess champion. Deep Blue was a chess-playing program developed by IBM. Deep Blue’s triumph over world chess champion Garry Kasparov signaled a promising breakthrough in the computational capabilities of AI. It used brute force to evaluate millions of potential moves with incredible speed.

1997: The Rise of Long Short-Term Memory (LSTM). While early recurrent neural networks (RNNs) could process sequential data like text, they suffered from the “vanishing gradient problem”—a flaw that caused them to “forget” the beginning of long sentences. To solve this, researchers introduced LSTM, a specialized RNN capable of selectively remembering or forgetting information over long periods. This breakthrough revolutionized natural language processing (NLP), allowing AI to finally grasp context in complex, real-world data.

1999: NVIDIA released the “first” GPU. Graphical processing units, or GPUs, had existed under different names in gaming and graphics systems for decades, but NVIDIA touted its GeForce 256 as the “first-ever” GPU. These computer chips enable the complex calculations required for advanced AI models.

2007: ImageNet and the Big Data Revolution. While the internet created an explosion of data, AI models lacked a “gold standard” for training. Computer scientist Fei-Fei Li bridged this gap by releasing ImageNet, a massive, meticulously labeled database of millions of images. By providing a common benchmark for evaluating image-recognition models, ImageNet became the essential fuel for the deep learning revolution that followed.

2012: AlexNet and the “Big Bang” of Deep Learning. The modern AI boom began when AlexNet—a deep CNN designed by Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton—shattered records at the ImageNet competition, beating the runner-up by a massive 10.8%. Its success proved that “deep” networks with millions of parameters were finally viable, thanks to the use of GPUs for high-speed training.

2016: AlphaGo Defeats Lee Sedol. Google DeepMind’s AlphaGo stunned the world by defeating a legendary Go champion. While chess was solved by “brute force,” Go’s complexity required deep reinforcement learning to mimic human intuition. AlphaGo’s creative, unconventional strategy proved that AI could master tasks once thought to be the exclusive domain of human intelligence.

2017: The transformer architecture. In their 2017 paper “Attention Is All You Need,” Google researchers proposed the transformer deep learning architecture. Transformers were an evolution of the RNN/LSTM that could process words in parallel using a mechanism called “self-attention.”

2020s: Large language models and modern AI

2020: OpenAI introduces GPT-3. The capabilities of the transformer architecture enabled the creation of large language models. One of the earliest and most successful examples is OpenAI’s generative pretrained transformer (GPT)-3; a later version was released as ChatGPT in 2022.

2021: OpenAI introduces DALL-E. DALL-E (the name is a combination of the cartoon robot WALL-E and artist Salvador Dalí) is a text-to-image model that can generate images from text descriptions (prompts). It uses a version of GPT-3 to generate images and has been a massive step forward in generative AI.

2023: An open letter about the threats of AI. Concerned by the rapid, exponential growth of AI (the AI boom), leading experts publish “Pause Giant AI Experiments: An Open Letter.” The letter called for a stop to AI development due to concerns about its use to create propaganda, automate jobs, and harm society.

Though this issue is not mentioned in the letter, other experts have expressed environmental concerns about the massive amounts of energy used to train and run AI systems.

2024–26: Agentic AI. In recent years, focus has begun to shift from programs that “generate” to agents that “do.” Unlike previous models, agentic AI (or AI agents) can execute a wide range of tasks independently, like navigating the web to book a trip or debugging software. This era represents a significant leap toward Artificial General Intelligence (AGI): systems that can emulate or even outperform human capability across a wide range of tasks.

Frequently asked questions about the history of AI

Who is the father of AI?

The term “artificial intelligence,” or AI, was first used by John McCarthy to propose a 1956 workshop on thinking machines. McCarthy and his fellow workshop organizers, Claude Shannon, Nathaniel Rochester, and Marvin Minsky, are considered some of the founding fathers of AI.

Other important figures in the field of AI include Alan Turing, whose famous Turing Test provided a framework for considering whether machines think; Allan Newell and Herbert Simon, who developed the first “reasoning” computer program (the Logic Theorist); Geoffrey Hinton (the “godfather of AI”), who conducted formative research on deep learning and neural networks; and Fei Fei Li (the “godmother of AI”), whose creation of the database ImageNet transformed how AI models are trained and evaluated.

Thanks to the efforts of early pioneers, today AI and AI-generated content feel unavoidable. Thankfully, tools like QuillBot’s AI Detector can help you determine whether content is human- or AI-generated.

When did AI start?

The idea of artificial intelligence (AI) has existed in fiction for millennia—”automatons,” or machines with agency and free will, can be traced back to Greek mythology.

However, AI as we now know it began to take off in the 1950s. The term “AI” was first used at a workshop in Dartmouth in 1956. John McCarthy, who helped organize this workshop, is often considered the father of AI.

The work of McCarthy and many other innovators has led to the AI capabilities we see today. Generative AI tools like QuillBot’s AI Image Generator are revolutionizing the way we work and create.