AI History: An Artificial Intelligence Timeline

Key takeaways

• AI has grown through trial and error, evolving from early ideas into tools people use every day.
• Learning from data transformed AI, allowing systems to improve over time instead of following fixed rules.
• Generative AI and large language models now power everyday tools, changing how people write, create, and work.
• New agentic AI systems are beginning to take on more complex, goal-driven tasks, pointing toward a future where AI plays a more active role in human workflows.

Artificial intelligence (AI) is no longer just a futuristic idea—it’s a part of everyday life. Whether it’s the tools we use at work or the apps we rely on at home, AI can help you write, analyze data, create content, and make decisions faster than ever before.

What began as bold theories and thought experiments from early visionaries like Alan Turing has evolved into powerful systems that shape how we live and work today.

In this guide, we’ll explore the most important moments in AI’s evolution through a clear artificial intelligence timeline. Along the way, you’ll see how breakthroughs, setbacks, and shifts in thinking transformed AI from an academic curiosity into one of the most influential technologies of our time—and where it may be headed next.

Table of contents

• What is AI?
• Key figures in the early history of AI
• 1950s–1960s: The origin of AI
• 1970s: The first AI winter
• 1980s: An AI revival through expert systems
• Late 1980s–1990s: The second AI winter
• 1990s: AI and the emergence of machine learning
• 2000s–2010s: AI and the rise of deep learning
• 2020s: Modern AI, generative AI, and agentic systems
• What is the future of AI?
• The history of artificial intelligence summarized
• History of AI FAQs

What is AI?

Before diving into AI’s history, it helps to start with a simple definition.

AI refers to computer systems designed to perform tasks that typically require human intelligence—such as understanding language, recognizing patterns, learning from data, and making decisions. Modern AI systems use methods such as machine learning (ML) and natural language processing (NLP) to learn from data and power familiar tools, such as writing assistants and recommendation engines.

Grammarly is one example of this AI technology in action: it uses ML, NLP, and AI capabilities like generative AI features and AI agents to act as your always-on AI writing partner. As you move through your everyday writing tasks, Grammarly analyzes your writing, surfaces issues, and suggests improvements in real time—right where you already work.

For a deeper explanation and real-world examples, check out our intro guide to AI.

Key figures in the early history of AI

AI didn’t emerge overnight. It was shaped by a small group of researchers who believed machines could one day think, learn, and reason like humans.

Alan Turing is often called the “father of artificial intelligence.” In 1950, he asked the groundbreaking question, “Can machines think?” and proposed the Turing Test, a way to evaluate whether a machine could convincingly imitate human conversation.

John McCarthy helped define the field itself. He coined the term “artificial intelligence” and organized the 1956 Dartmouth Summer Research Project, widely considered the birth of AI as an academic discipline.

Other early pioneers helped turn theory into practice. Marvin Minsky advanced research into human cognition and machine intelligence, Allen Newell and Herbert Simon built early problem-solving programs, and Arthur Samuel laid the foundation for machine learning by creating systems that improved through experience.

Together, these thinkers set the stage for the excitement—and challenges—that would define AI’s early decades.

1950s–1960s: The origin of AI

The origin of AI dates back to the 1950s, when researchers began exploring whether machines could simulate human intelligence using logic, rules, and early computing systems.

In the early days of computing, researchers began asking a bold question: Could a machine ever think like a human? The 1950s and 1960s were shaped by big ideas, limited technology, and early experiments that laid the groundwork for everything AI would become.

Some of the key milestones from this era include:

• Alan Turing asks, “Can machines think?” (1950): In his influential paper “Computing Machinery and Intelligence,” Turing proposed the Turing Test, suggesting that if a machine could hold a convincing conversation, it could be considered intelligent—a radical idea at the time.
• AI gets its name at the Dartmouth Conference (1956): The term “artificial intelligence” was introduced during a summer workshop attended by a small group of researchers that marked the first major convening of the field.
• The perceptron shows that machines can learn patterns (1957): Frank Rosenblatt introduced the perceptron, an early neural network that could recognize simple patterns, hinting that machines might learn from data rather than follow strict rules.
• ELIZA becomes the first chatbot (1966): Created at MIT, ELIZA simulated a therapist using basic language rules. While the chatbot was simple, users found it to be engaging, revealing how quickly people can attribute intelligence to machines.
• Shakey the Robot takes AI into the real world (1966): Shakey was the first mobile robot that could move, sense its surroundings, and make decisions—demonstrating that AI could interact with the physical world, not just text or numbers.

These early systems were limited, but they proved something important: Machines could imitate aspects of human thinking, learning, and interaction—even if true intelligence was still far off.

Key takeaway: The 1950s and 1960s established the core ideas of AI and sparked the optimism that would drive the field forward.

1970s: The first AI winter

The first of two AI winters occurred in the 1970s, when early optimism around artificial intelligence faded due to technical limitations, unmet expectations, and major cuts in funding.

After the excitement of AI’s early breakthroughs, reality began to set in. While researchers had made promising progress, computers were still slow, expensive, and limited in what they could handle. As a result, many bold predictions about AI (such as Minsky saying in 1970 that in “three to eight years we will have a machine with the general intelligence of an average human being”) failed to materialize—leading to disappointment among funders and policymakers.

Key factors that led to the first AI winter include:

• Computers weren’t powerful enough: Early AI programs required more memory and processing power than computers of the time could provide, making them impractical outside of research labs.
• Early neural networks hit hard limits: Research revealed that models like the perceptron could solve only simple problems, casting doubt on whether machines could truly scale toward humanlike intelligence.
• Overpromising, underdelivering: Researchers and institutions made ambitious claims about AI’s potential, but real-world results lagged behind expectations.
• Funding was reduced or cut entirely: Governments and organizations scaled back investment in AI research, slowing progress and shrinking the field.

As funding dried up, many AI projects were abandoned or delayed. Interest in the field declined, and AI research entered a prolonged slowdown.

Key takeaway: The first AI winter showed that getting computers to replicate human thought is far more complex than early researchers anticipated—and that progress in AI depends on both realistic expectations and technological readiness.

1980s: An AI revival through expert systems

In the 1980s, AI experienced a revival as researchers shifted from ambitious general intelligence goals to practical systems designed to solve specific, real-world problems.

After the slowdown of the 1970s, AI researchers took a more grounded approach. Instead of trying to replicate human intelligence as a whole, they focused on capturing expert knowledge in narrow domains—leading to AI systems that businesses could actually use given the current state of computing.

Key developments from this era include:

• Expert systems bring AI into the workplace: Expert systems are designed to mimic the decision-making of human specialists. By following structured rules, they offer recommendations or diagnoses in specific fields.
• MYCIN and XCON show real-world value: MYCIN helped diagnose blood infections, while XCON assisted in configuring computer systems for companies like Digital Equipment Corporation. These tools demonstrated that AI could deliver measurable business impact.
• Commercial interest and investment return: As expert systems proved useful, companies increased funding and adoption, marking the first time AI achieved sustained commercial success.
• Backpropagation renews interest in neural networks: In the mid-1980s, improvements in backpropagation made it possible to train multilayer neural networks more effectively, setting the stage for future deep learning breakthroughs.

Despite being expensive to build and maintain and brittle in the face of unexpected information, early expert systems proved the value of AI in the business world

Key takeaway: The 1980s showed that AI thrives when it focuses on practical, narrowly defined problems—but scalability and flexibility remained major challenges.

Late 1980s–1990s: The second AI winter

The second AI winter occurred in the late 1980s and early 1990s, when an economic downturn—combined with rising costs, limited flexibility, and unmet expectations—led to another slowdown in artificial intelligence research and investment.

Despite the success of expert systems earlier in the decade, enthusiasm for AI once again began to fade. While these systems worked well in controlled environments, they proved difficult to maintain and adapt as conditions changed—revealing important limitations in how AI was being built and deployed.

Key factors behind the second AI winter include:

• Expert systems were costly: Building rule-based systems required painstaking human input; every improvement and correction required both experts and engineers to spend yet more hands-on time.
• AI systems struggled outside narrow use cases: Most AI tools worked only in highly specific scenarios and failed or behaved unpredictably when faced with new or ambiguous situations, limiting their broader usefulness.
• Expectations again outpaced results: As AI was marketed more widely, expectations rose faster than the technology could deliver, leading to frustration among businesses and funders.
• Investment slowed across the industry: Companies and governments reduced spending on AI projects that didn’t show clear long-term value, causing research momentum to stall.

Unlike the first AI winter, this period wasn’t driven by a lack of ideas—but by the realization that existing approaches couldn’t easily scale or adapt to real-world complexity.

Key takeaway: The second AI winter reinforced a critical lesson: For AI to succeed long term, it must be flexible, scalable, and able to learn—setting the stage for the rise of machine learning in the years that followed.

1990s: AI and the emergence of machine learning

In the 1990s, AI shifted toward machine learning, as researchers began building systems that inferred patterns from data instead of relying on hand-coded rules.

After two AI winters, it became clear that manually programming intelligence wasn’t scalable. Instead, AI researchers turned to machine learning (ML), a data-driven approach that allowed systems to improve through trial and error—bringing new momentum to the field.

Key developments from this era include:

• Learning replaces rigid rules in AI systems: Rather than following predefined instructions, ML models could learn patterns from data, making AI more flexible and better suited to real-world problems.
• New machine learning algorithms gain adoption: Techniques such as decision trees and support vector machines (SVMs) became widely used for classification and prediction tasks due to their accuracy and reliability.
• Ensemble methods improve performance: Methods like bagging and boosting combined multiple models to produce stronger results, forming the basis of many modern machine learning systems.
• Reinforcement learning advances AI decision-making: Algorithms such as Q-learning allowed AI systems to learn through trial and error, influencing future work in robotics, games, and control systems.
• AI enters practical, everyday applications: Machine learning began powering tools for fraud detection, document classification, speech recognition, and early recommendation systems.

These early machine learning breakthroughs set the stage for deep learning in the 2000s, when larger datasets and increased computing power would unlock even more advanced AI capabilities.

Key takeaway: The 1990s transformed AI into a data-driven discipline, laying the groundwork for the deep learning revolution that followed.

2000s–2010s: AI and the rise of deep learning

In the 2000s and 2010s, AI advanced rapidly. Core technologies such as deep learning, more powerful computers, and improved algorithms came about exactly when the large quantities of content required to train deep learning became available through a quickly growing internet.

While machine learning had proven effective in the 1990s, many AI systems still struggled with complex tasks like image recognition and language understanding. Deep learning changed that by using neural networks with many layers, allowing AI to learn richer and more abstract patterns from data.

Key developments from this era include:

• Deep belief networks revive neural networks (mid-2000s): Researchers demonstrated that deep neural networks could be trained more effectively, helping overcome earlier limitations and reigniting interest in neural approaches.
• RNNs and LSTMs (2010s): Recurrent neural networks (RNNs), particularly long short-term memory networks (LSTMs), became foundational for speech recognition, machine translation, and time-series prediction—significantly improving AI’s ability to process sequential data.
• CNNs transform image recognition (2012): Convolutional neural networks (CNNs) achieved major breakthroughs in computer vision, most notably after winning the ImageNet competition, dramatically improving image classification accuracy.
• Transformer architecture reshapes natural language processing (2017): Transformers, which look at all relationships in a sequence at once, made it faster and more efficient to train large-scale language models and redefined how AI handles text.
• Early GPT models point toward modern language AI (2018): The first generative pre-trained transformer (GPT) models showed how transformers trained on massive text datasets could generate coherent, humanlike language—laying the groundwork for today’s large language models (LLMs).

By the end of the 2010s, deep learning had become the dominant approach in AI research and industry, directly enabling the rise of large language models and generative AI in the decade that followed.

Key takeaway: Deep learning unlocked major advances in vision, speech, and language—and early models like GPT set the stage for the modern AI systems you use today.

2020s: Modern AI, generative AI, and agentic systems

In the 2020s, AI became mainstream as large language models, generative AI, and emerging agentic systems reshaped how people work, create, and interact with technology.

Building on advances from the previous decade—especially transformers and early GPT models—AI systems grew dramatically more capable. For the first time, AI wasn’t just powering systems behind the scenes; it became something you could interact with directly, using natural language.

Key developments from this era include:

• LLMs make AI conversational: Models trained on massive text datasets learned to understand context, answer questions, summarize information, and generate humanlike language. Chat-based interfaces made AI easier to use and more widely accessible than ever before.
• Generative AI expands AI from analysis to creation: Generative AI systems began producing original text, images, music, video, and code from simple prompts. These tools support tasks like writing, brainstorming, design, and software development—changing how people create content.
• AI becomes embedded in everyday workflows: Rather than operating within stand-alone tools, AI is now integrated into the products people use daily, helping improve productivity, communication, and decision-making.
• Agentic AI begins to emerge: A new class of systems called agentic AI started to take shape. These systems can plan steps, use tools, and carry out multistep tasks toward a goal, moving AI from responding to prompts to actively supporting outcomes.

Together, these developments mark a shift in how AI is experienced: from a background technology to an active collaborator in work and creativity.

Key takeaway: The 2020s transformed AI into a visible, interactive, and creative force—setting the stage for a future where AI systems play a more active role in how tasks are planned, created, and completed.

What is the future of AI?

As AI becomes part of everyday tools and workflows, its future will be shaped by both near-term advances and longer-term research goals.

In the near term, improvements in LLMs, generative AI, and agentic systems will make AI more helpful and more deeply integrated into how you work and create. AI is moving beyond single tasks to support planning, collaboration, and multistep workflows—while remaining under human guidance.

At the same time, researchers continue to explore artificial general intelligence (AGI), which refers to AI systems that can learn and reason across many tasks, not just one. While progress is being made, AGI remains a long-term goal rather than an immediate reality.

Beyond AGI lies artificial superintelligence (ASI), a theoretical stage where AI surpasses human intelligence across all fields. Because ASI raises significant ethical and safety questions, many AI safety researchers focus on responsible development and strong governance.

Key takeaway: The future of AI combines practical innovation with long-term exploration—and how these systems are developed and governed will shape their impact on society.

The history of artificial intelligence, summarized

The history of artificial intelligence is a story of bold ideas, setbacks, and steady progress. What began in the 1950s as a theoretical question—“Can machines think?”—evolved through periods of optimism, AI winters, and major breakthroughs in machine learning and deep learning.

Today, AI has moved from the background into your everyday work. Advances in LLMs, generative AI, and emerging AI agents have transformed AI from a passive technology into an active collaborator—one that can help you plan, write, revise, and communicate more effectively.

Grammarly reflects this evolution of AI’s seamless integration into day-to-day tools. For over 16 years, Grammarly’s AI capabilities have been behind millions of users’ everyday writing tasks, from drafting emails to polishing documents to brainstorming next steps. Now, that same foundation is powering more advanced capabilities: generative AI features that help you brainstorm and draft content with a simple prompt, and AI agents that proactively support your work goals across the entire writing process and help you communicate more clearly and confidently.

History of AI FAQs

What is the history of AI?

AI began in the 1950s with early theories about machine intelligence and primitive programs that mimicked human thought and language. After two periods of slowed progress known as AI winters, the field evolved from rule-based expert systems to today’s machine learning and generative systems.

Who is the founder of AI?

Alan Turing is often called the father of artificial intelligence for his framing of the question, “Can machines think?” and for creating the Turing Test to determine if a person can distinguish communication with a computer from communication with another human. The term “artificial intelligence” was coined by John McCarthy, a major figure in creating AI as an academic field.

What was the first AI ever built?

While there is no single “first AI,” the perceptron is considered the first neural network. Created in 1957 by Frank Rosenblatt, it mimicked brain neurons by reacting to inputs with a binary response—the same fundamental concept underlying today’s deep neural networks.

What caused the AI winter?

Both AI winters came about when modest technological improvements failed to meet heightened expectations. In the 1970s, limited computing power restrained innovation after some initial successes. In the late 1980s, expert systems proved too expensive and inflexible for most applications. In both periods, research funding and industry enthusiasm waned for several years.

What was the reason for the second AI winter?

The second AI winter, spanning the late 1980s and early 1990s, was a consequence of expert systems being expensive and inflexible. While these programs were useful for scaling analysis in certain contexts, they took too much work to build and adapt to be practically applied to many real-world applications. Funding dried up, and research stalled until machine learning emerged as the leading edge of AI.