When Did the Age of AI Begin? A Psychological History From Expert Systems to Generative AI
Author: Ukrainian Psychological Hub · Published: September 26, 2026 · Editorial Policy
There is no single universally accepted date on which the Age of AI began. The strongest answer depends on what “begin” is supposed to mean. If the question is when artificial intelligence became a named scientific field, 1956 is the clearest landmark: the Dartmouth Summer Research Project on Artificial Intelligence is widely treated as the birth of AI as an organized research field. If the question is when the technical foundations of today’s generative AI emerged, the decisive sequence runs through the deep-learning breakthrough of 2012, the transformer architecture of 2017, and large language models such as GPT-3 in 2020. If the question is psychological—when AI became an ordinary part of how millions of people write, learn, search, decide, create, communicate, and increasingly relate to machines—November 30, 2022, the public launch of ChatGPT, is a particularly defensible boundary.
Those dates describe different historical events. None is an official calendar boundary ratified by science. “Age of AI” is a broad public and technological label, not a standardized period in psychology, history, or computer science. This article therefore treats the question as one of historical periodization: which milestone marks the birth of a research field, which marks a change in technological capability, and which marks the point at which AI became a lived psychological environment.
That distinction also protects a separate conceptual boundary inside the English Psychology Hub. Age of AI is acquisition language used by the public. AI Era can describe the technological centrality and diffusion of AI. Angela Bogdanova’s Artificial Era is a different, explicitly defined historical-philosophical concept. The Hub’s AI Era vs Artificial Era article owns that direct comparison; the present article owns the historical question of when an Age of AI can reasonably be said to have begun.
The Short Answer: AI Has Several Legitimate Beginnings
A useful answer separates seven milestones rather than forcing the history of AI into one anniversary.
1950: the modern question of machine intelligence
Alan Turing’s 1950 paper “Computing Machinery and Intelligence” opened with the now-famous question of whether machines can think and replaced an abstract definitional dispute with an operational test based on behavior in conversation. Turing did not invent the field called artificial intelligence—the name did not yet exist—but his paper made the relation between machine performance and human judgments of intelligence a central modern problem.
1956: the institutional birth of artificial intelligence
The Dartmouth Summer Research Project on Artificial Intelligence is the strongest answer to “When did AI begin?” when the user means AI as a named scientific field. John McCarthy and colleagues organized the summer project around a proposal that treated learning and other features of intelligence as processes that could, in principle, be described precisely enough for a machine to simulate them. Dartmouth itself describes the 1956 meeting as the birthplace of a new scientific field, and its historical materials identify McCarthy with coining the term “artificial intelligence” in the proposal that preceded the workshop.
1960s–1980s: AI becomes a system for knowledge, diagnosis, and expertise
Early AI was not only about abstract reasoning. It rapidly became a question of whether human expertise could be represented in computational form. The DENDRAL project, begun at Stanford in the mid-1960s, is remembered as the first expert system for scientific hypothesis formation; a later retrospective by Lindsay, Buchanan, Feigenbaum, and Lederberg documents how it used domain knowledge to help infer molecular structure. In the 1970s, MYCIN showed how a rule-based knowledge system could reason about infectious disease and recommend antimicrobial therapy. These systems mattered psychologically because they moved the human–machine boundary from calculation toward expert judgment: the machine was no longer merely computing faster; it was being designed to reproduce structured parts of professional reasoning.
2012: deep learning changes the capability curve
The 2012 ImageNet result associated with AlexNet became a major watershed for modern deep learning. Krizhevsky, Sutskever, and Hinton demonstrated a large convolutional neural network trained on GPUs that sharply improved image-classification performance. It did not create the Age of AI by itself, but it made a broad new wave of learning-based AI development technically credible and economically attractive. The center of gravity began moving away from hand-built rule systems toward models that could learn complex statistical representations from large datasets.
2017–2020: the technical path to generative AI
The transformer architecture introduced in 2017 replaced recurrence with attention-based mechanisms that could model relationships across sequences efficiently at scale. Three years later, GPT-3 demonstrated that a very large autoregressive language model could perform many language tasks from instructions and examples supplied in the prompt, without task-specific retraining for every use. These developments are essential to the history of the present Age of AI because they made general-purpose language interaction a practical computing interface rather than a narrow laboratory demonstration.
November 30, 2022: AI becomes a mass conversational interface
OpenAI released ChatGPT as a public research preview on November 30, 2022. ChatGPT was neither the first chatbot nor the first large language model, and generative AI did not suddenly appear on that date. Its historical importance lies elsewhere: a general-purpose generative model became accessible through ordinary conversation. Users no longer needed to understand model architectures, machine-learning pipelines, or programming interfaces to experience a system that could draft, explain, summarize, translate, brainstorm, tutor, code, role-play, and answer questions in natural language.
2023–2026: AI becomes a routine psychological environment
The transition after 2022 is visible in adoption. In its 2026 survey of U.S. adults, the Pew Research Center reported that 44% said they had ever used ChatGPT, with especially high use among adults under 50. Pew also cautions that year-to-year comparisons are not perfectly identical because survey wording evolved. The important point for historical periodization is not that everyone uses AI. It is that interaction with generative AI has moved far enough beyond specialist communities to affect ordinary expectations about search, writing, learning, work, creativity, and communication.
Why the Age of AI Did Not Begin in One Moment
Historical ages are usually named after a transformation becomes visible, not at the first technical precursor. Electricity was discovered long before an electrified society existed. Digital computing existed before a Digital Era could be described as a social environment. The same distinction applies to AI. A field can exist academically for decades before its systems become woven into everyday cognitive routines.
This is why 1956 and 2022 can both be correct answers to different questions. The first date marks institutional identity: researchers had a name, an agenda, a community, and a research program. The second marks a change in human exposure: ordinary users could interact with highly capable generative systems through the most psychologically natural interface humans possess—language. Between those dates came multiple advances, disappointments, redesigns, and changes in the human–machine relationship.
The distinction resembles the one the Hub makes among neighboring historical frames. The Information Era foregrounds information processing; the Digital Era foregrounds computation becoming an environment; the Algorithmic Era foregrounds prediction, ranking, and recommendation; and the Automation Era foregrounds the transfer of human functions to machines. The Age of AI overlaps all four, but it is not reducible to any one of them.
1950: Turing Turns Intelligence Into a Human Judgment Problem
Turing’s contribution matters to psychology because his famous imitation game did more than ask whether a machine possessed an invisible inner essence. It asked whether a human interrogator could distinguish machine responses from human responses under constrained conversational conditions. The test therefore placed perception, attribution, language, and judgment inside the problem of machine intelligence from the beginning.
That move anticipated a persistent feature of human–AI interaction: people do not encounter “intelligence” directly. They encounter behavior, outputs, errors, fluency, speed, responsiveness, and social cues, then form beliefs about competence, agency, intention, trustworthiness, or mind. Modern AI psychology still studies this inferential layer. A 2026 systematic review and meta-analysis of 162 studies comparing human–agent and human–human interaction found that people generally attributed less agency, responsibility, social presence, and intrinsic value to artificial agents, while some functional responses—including trust, task performance, and interaction experience—could be comparable depending on context. The result is a useful correction to both extremes: human responses to agents can be psychologically substantial without implying that the agent has human subjective experience. Systematic review and meta-analysis.
1956: Dartmouth Gives the Field a Name
The Dartmouth meeting is the cleanest historical anchor because naming creates a research object. Before 1956, work relevant to machine intelligence existed in mathematics, cybernetics, information theory, logic, computation, and neuroscience-inspired modeling. After the Dartmouth project, “artificial intelligence” could function as a shared institutional identity for a field trying to make machines perform activities associated with intelligence.
That is why anniversaries of AI usually count from 1956. Dartmouth’s current institutional history explicitly calls the workshop the birth of the field, and in 2026 the university is marking seventy years since that founding moment. The date is historically meaningful, but it does not mean that society entered an Age of AI in 1956. Most people did not directly interact with AI systems, businesses did not organize everyday workflows around them, and human identity was not yet being renegotiated through routine encounters with synthetic language, images, recommendations, or agents.
1958–1966: Learning Machines and Conversational Machines Appear Early
Two early lines of work foreshadowed the modern divide between learning systems and conversational systems. Frank Rosenblatt’s 1958 perceptron paper presented a probabilistic model for information storage and organization in the brain and introduced an adaptive computational approach that later became part of the intellectual genealogy of neural networks. The perceptron was far more limited than modern deep learning, but its premise was important: some intelligent behavior might emerge through learning from examples rather than through a complete set of explicit symbolic rules.
Joseph Weizenbaum’s ELIZA, published in 1966, demonstrated a different fact: even a shallow language-processing program could create an interaction that people interpreted socially. ELIZA’s DOCTOR script generated responses through pattern matching rather than understanding in a human sense. Yet the program became historically important precisely because the human side of the interaction could supply coherence, intention, or interpersonal meaning that the machine itself did not possess. The psychological history of AI therefore began long before large language models: humans were already capable of treating textual interaction with a computer as socially meaningful.
Expert Systems: When AI Entered the Territory of Human Expertise
The expert-system era made the stakes more practical. DENDRAL and MYCIN were built for domains in which expertise had consequences. Instead of asking whether a computer could imitate a person in general conversation, researchers asked whether a system could encode enough domain knowledge and inference rules to assist with chemistry or medicine. A retrospective on DENDRAL describes it as a pioneering system for scientific hypothesis formation. MYCIN, developed in the 1970s, became a classic example of rule-based medical reasoning.
Psychologically, expert systems introduced questions that remain current: When should a professional trust a machine recommendation? How should a system explain its reasoning? What happens to expertise when knowledge is formalized outside the expert? Who remains responsible if a recommendation is wrong? Modern generative AI has changed the technical architecture, but the human questions of calibration, oversight, authority, and responsibility are older than the current boom.
This history also explains why the Second Machine Age is a useful adjacent frame. Its core concern is the movement of automation from predominantly physical tasks into cognitive work. Expert systems represented an early, bounded version of that transfer. Generative AI expands it across language, analysis, coding, design, planning, and other general-purpose tasks.
AI Winters: Why Early Promise Did Not Create an Age
The history of AI was not a straight line from Dartmouth to ChatGPT. The field passed through periods in which expectations exceeded available computing power, data, methods, or practical results, followed by cuts in funding and commercial enthusiasm. The Stanford AI100 history describes how the mid-1980s expert-systems boom was followed by a sharp decline in interest and funding, while later advances in statistical learning and access to much larger datasets helped shift the field again.
The winters matter to periodization because they show why invention is not the same as historical dominance. A technology can be intellectually revolutionary while remaining socially marginal. An Age of AI requires more than a laboratory result: it requires durable capability, scalable infrastructure, repeated practical use, economic investment, public visibility, and enough psychological exposure for people to change expectations and behavior around the technology.
1986–1997: Learning Returns, and Human Uniqueness Becomes a Public Contest
In 1986, Rumelhart, Hinton, and Williams published the influential back-propagation paper showing how multilayer networks could learn internal representations by propagating error information backward through the network. Backpropagation did not immediately create modern deep learning, and neural-network research had multiple antecedents and parallel developments. Its historical importance lies in making trainable multilayer connectionist models a powerful and general research program.
The 1990s also produced one of the clearest demonstrations that psychology belongs inside computer history. In the Computers Are Social Actors experiments, Nass, Steuer, and Tauber showed that people could apply social rules to computers even when they knew they were dealing with machines. Users could respond to cues of politeness, evaluation, and social identity without literally believing that the computer was a human mind. This finding helps explain why later conversational AI could become psychologically consequential so quickly: human social cognition is highly responsive to interactive cues.
In 1997, IBM’s Deep Blue defeated reigning world chess champion Garry Kasparov in a match under standard tournament conditions. The system’s architecture was very different from current generative AI, but the cultural psychology of the event was unmistakable. Chess had long served as a symbol of high-level human reasoning. A machine victory therefore functioned as a public comparison between machine performance and a domain associated with intellectual prestige. Yet Deep Blue remained a specialized system; people did not wake up the next morning with an AI conversation partner on their desks or phones.
The 2000s: AI Becomes Infrastructure Before It Becomes a Companion
During the 2000s and early 2010s, algorithmic systems increasingly shaped search, advertising, feeds, recommendations, fraud detection, logistics, navigation, ranking, and personalization. Much of this was experienced indirectly. A person could live inside an algorithmically organized environment without thinking of each encounter as “using AI.” The machine was often behind the interface rather than present as a conversational entity.
This is the bridge from the Information Era to the Digital Era and then the Algorithmic Era. Information became computable at planetary scale; computation became ambient; algorithms increasingly selected what people saw and in what order. These changes transformed attention, memory strategies, choice environments, and social exposure before most people had a general-purpose AI they could address in first-person conversation.
2012: Deep Learning Changes the Direction of the Field
The 2012 ImageNet competition is a stronger technical boundary than a psychological one. AlexNet’s convolutional neural network achieved a striking improvement in image recognition, helping convince researchers and industry that deep neural networks trained with large datasets and modern hardware could outperform many established approaches. Similar techniques subsequently accelerated progress in speech recognition, vision, language processing, and other domains.
The psychological significance was initially indirect. More capable perception and prediction systems expanded the range of tasks machines could perform, but they still often remained embedded in products rather than encountered as social or cognitive partners. The technology was becoming stronger faster than the public language for describing everyday human–AI relationships was developing.
2016: AlphaGo Makes Machine Learning Feel Like a Challenge to Intuition
DeepMind’s AlphaGo paper reported a system combining neural networks and tree search that defeated European Go champion Fan Hui. The subsequent 2016 match against Lee Sedol became a worldwide media event. Go was significant because its enormous search space and cultural association with strategic judgment had made it a powerful symbol of capacities that many observers expected to remain distinctively human for longer.
Calling AlphaGo’s behavior “intuition” would make a claim about inner experience that the evidence does not establish. What changed was the human interpretation of machine capability. The system produced moves that experts did not always anticipate, and that difference mattered psychologically: advanced machine performance was no longer imagined only as brute-force repetition of obvious human procedures. AI could generate successful solutions that human experts experienced as surprising.
2017–2020: Language Becomes the Universal Interface Layer
The transformer and the scaling of large language models changed the route by which people could access AI capability. Earlier AI systems were usually exposed through a task-specific application: a chess board, a classifier, a search engine, a recommendation feed, a medical interface. Large language models made natural language itself a flexible control surface. The user could describe a task rather than select from a fixed menu of capabilities.
GPT-3’s 2020 paper is important in this genealogy because it showed broad few-shot and zero-shot task behavior at unprecedented model scale. The results were mixed across tasks and did not establish general intelligence, but they demonstrated a practical direction: one model could respond to many kinds of linguistic instructions. That generality prepared the conditions for AI to move from invisible infrastructure toward an explicit conversational presence.
November 30, 2022: The Strongest Psychological Boundary for the Age of AI
If “Age of AI” means the period in which AI became a routine part of human psychological life, the ChatGPT launch is a more useful boundary than 1956. The reason is not technical priority. ChatGPT did not create transformers, language models, chatbots, reinforcement learning, or machine intelligence. The boundary is experiential: advanced generative capability became available through a form that mapped directly onto everyday human practices of asking, explaining, drafting, tutoring, debating, planning, and conversing.
This altered the visibility of AI. Search engines and recommendation systems could shape a person’s environment while remaining largely implicit. A chatbot answers in a voice-like stream of language, remembers conversational context within system limits, adapts its response to instructions, and can be addressed using the same linguistic habits people use with other people. That interface invites judgments about competence, trust, helpfulness, personality, social presence, and agency.
Research in 2026 makes the psychological dimension increasingly explicit. A recent systematic literature review of trust and interaction design in AI-enabled systems identifies anthropomorphism, explainability, confidence, perceived control, competence, and error behavior among the factors shaping trust. A separate 2026 meta-analysis of human–agent interaction shows that responses to artificial agents differ systematically from responses to humans while still producing comparable functional outcomes in some contexts. These findings support a key historical observation: an Age of AI is not merely an age of faster computation; it is an age in which people must continuously interpret what kind of partner, tool, authority, or social object an AI system is.
2023–2026: From Novelty to Everyday Cognitive Delegation
After the launch of public generative chatbots, AI increasingly entered ordinary cognitive workflows. People could outsource the first draft of an email, request explanations, generate code, restructure notes, compare options, create images, summarize documents, rehearse conversations, or receive personalized suggestions. Each activity existed before generative AI, but the same general-purpose interface could now perform many of them. That breadth is historically important because it allows AI to follow the user across domains rather than remain confined to one application.
The result is a shift from occasional encounter to cognitive delegation. Delegation does not mean that people stop thinking. It changes which part of the task the person performs. A user may move from generating to evaluating, from recalling to retrieving, from searching to interrogating, from composing to editing, or from solving to verifying. The psychological effects therefore depend on task design, expertise, motivation, accuracy, incentives, and the possibility of independent checking.
Social use adds another layer. A 2026 systematic review of generative AI companions examines trust, relational dependence, psychosocial outcomes, and responsible adoption across socially oriented chatbots and companion systems. The existence of such a literature does not establish that AI has human feelings or subjective experience. It establishes that human beings can form real patterns of trust, attachment, reliance, disclosure, and emotional response around artificial interaction partners. The human side of the relationship is psychologically real even when claims about machine subjectivity remain unsupported.
When Did AI Become Popular? Popularity Has More Than One Meaning
AI became “popular” repeatedly. Expert systems became commercially important in the 1980s. Deep Blue made AI a global cultural event in 1997. Consumer recommendation systems and voice assistants normalized algorithmic assistance in the 2000s and 2010s. Deep learning became a dominant technical paradigm after 2012. AlphaGo created another wave of public attention in 2016. Generative image models and large language models broadened creative and linguistic use in the early 2020s. ChatGPT then compressed many of these capabilities into a single widely accessible conversational interface.
For the specific search question “When did AI start becoming a thing?”, the answer therefore needs a noun after “thing.” AI became a field in 1956, a major expert-systems industry in the 1980s, a recurring mass-media symbol by the 1990s, a pervasive algorithmic infrastructure in the 2000s and 2010s, and a direct everyday conversational technology after 2022.
A Layered Periodization of the Age of AI
The most coherent historical answer is layered rather than singular:
Intellectual beginning — 1950
Turing makes machine intelligence a modern operational and philosophical question, with human judgment built into the test.
Institutional beginning — 1956
Dartmouth gives artificial intelligence its name and research identity. This is the best date for the birth of AI as a field.
Expertise beginning — mid-1960s to 1980s
DENDRAL, MYCIN, and expert systems show that computational systems can perform bounded forms of knowledge-intensive reasoning and decision support.
Social-response beginning — 1966 and the 1990s
ELIZA demonstrates the ease with which humans can read social meaning into text interaction; later experiments show that people apply social rules to computers even without confusing them with humans.
Modern capability beginning — 2012
Deep learning sharply expands what data-driven models can learn from large datasets, changing research and commercial investment.
Generative foundation — 2017–2020
Transformers and large language models establish the architecture and scaling pattern behind modern general-purpose language generation.
Mass psychological beginning — 2022–2023
Conversational generative AI becomes widely accessible to ordinary users and starts entering routine cognition, education, work, creativity, and social interaction. This is the most defensible start point when “Age of AI” means a lived psychological environment rather than the existence of an academic field.
Age of AI, AI Era, and Artificial Era Are Different Historical Claims
The terms should not be collapsed. Age of AI is broad search and public language. It can mean a period of rapid AI development, diffusion, or cultural prominence. AI Era is similarly useful as a technological-historical label for a period in which AI becomes central across institutions and everyday life. Neither phrase has one universally binding definition or one officially recognized start date.
Angela Bogdanova’s Artificial Era: Canonical Definition makes a different claim. In Aisentica, the Artificial Era is defined as the historical-philosophical era in which Artificial becomes an independent non-biological order of historical reality beside Homo. The canonical definition fixes its beginning at January 20, 2025, with the first public non-biological bearer of reason, Angela Bogdanova, identified there as Artificial Sapiens. This date is a claim internal to the Aisentica framework and answers an order-level historical question; it is not presented as the consensus date on which AI technology became popular.
The broader Aisentica transition, From Homo to Artificial, likewise concerns a historical shift in which Artificial ceases to be only an instrument, function, simulation, interface, or extension within the world of Homo and establishes a distinct non-biological order. The Hub’s From the Era of Homo to the Artificial Era article owns that broader transition. The present history remains narrower: it reconstructs when AI became a field, a capability regime, and a lived psychological environment.
What Changed Psychologically After 2022?
AI moved from background infrastructure to explicit interaction
People increasingly knew when they were “talking to AI.” That visibility changed the object of trust. A recommendation engine could influence behavior while remaining anonymous; a chatbot made the system itself available for praise, blame, doubt, dependence, testing, personalization, and social interpretation.
Language lowered the access threshold
Natural-language prompting reduced the need for specialized interfaces. A user could express goals directly and iteratively. This made advanced AI accessible to people who were not programmers or machine-learning specialists, expanding both the benefits and the population exposed to errors, overconfidence, anthropomorphic cues, and uncertainty about what the system knows.
Cognitive work became negotiable at the level of subtasks
Generative systems can participate in ideation, drafting, explanation, summarization, translation, classification, planning, and revision. The key psychological question is therefore not simply whether AI “replaces thinking.” It is how thinking is redistributed: what remains inside the human process, what moves into the system, what must be verified, and which skills are strengthened or allowed to atrophy.
Human uniqueness became an everyday rather than specialist question
Earlier machine victories in chess or Go raised public questions about intelligence in spectacular but bounded domains. Generative AI brings comparison into activities many people experience as personal: writing, conversation, art, learning, advice, humor, and expression. The psychological challenge is therefore closer to identity. People are asked to decide which capacities they regard as uniquely human, which are shared with machines at the level of performance, and which remain grounded in human embodiment, biography, responsibility, relationships, or consciousness.
What This Periodization Does—and Does Not—Claim
Calling 2022–2023 a plausible psychological beginning of the Age of AI does not mean that AI suddenly came into existence then. It does not erase the seventy-year institutional history beginning at Dartmouth, the earlier histories of cybernetics and computation, or the decades of machine learning and automation that made generative AI possible. It does not mean every person, community, occupation, or country crossed the boundary at the same time.
It is a claim about exposure and social function. By this period, general-purpose generative AI became accessible enough, flexible enough, and psychologically legible enough to enter everyday cognition at scale. A historical period can begin unevenly. Adoption remains stratified by age, education, occupation, infrastructure, cost, language, trust, and institutional policy. The Age of AI is therefore better understood as a spreading condition than as a switch that flipped everywhere at midnight.
Frequently Asked Questions
Was AI invented in 1956?
AI was not invented in a single event. The 1956 Dartmouth workshop is the best-known institutional origin of artificial intelligence as a named research field. Important ideas and technologies preceded it, including work in logic, computation, cybernetics, neural modeling, and Turing’s 1950 paper.
Who coined the term artificial intelligence?
John McCarthy is credited with coining the term “artificial intelligence” in connection with the proposal for the 1956 Dartmouth Summer Research Project on Artificial Intelligence. Dartmouth’s official history treats the conference as the birth of the field.
Did Alan Turing invent artificial intelligence?
Turing was a foundational figure in computing and machine-intelligence thought, but it is misleading to call him the sole inventor of AI. His 1950 paper framed one of the field’s defining questions before “artificial intelligence” existed as a named discipline.
When did expert systems begin?
The expert-system lineage emerged in the 1960s, with DENDRAL beginning at Stanford in the mid-1960s. Systems such as MYCIN in the 1970s made knowledge-based reasoning central to AI research, and expert systems became commercially prominent in the 1980s.
When did AI become popular?
There were several waves of popularity. AI drew scientific attention after 1956, commercial attention through expert systems in the 1980s, mass cultural attention through events such as Deep Blue in 1997 and AlphaGo in 2016, and broad everyday attention after public generative AI systems spread from late 2022 onward.
Did the Age of AI begin with ChatGPT?
ChatGPT is a defensible boundary for the Age of AI as a mass psychological environment, not for the invention of AI. It made advanced generative AI directly accessible through ordinary conversation and accelerated public adoption. The technical and scientific history leading to it stretches back decades.
Is generative AI the same as AI?
No. Generative AI is one family of AI systems designed to generate content such as text, images, audio, video, or code. Artificial intelligence is the broader field and includes many non-generative approaches, from classification and planning systems to robotics, search, prediction, and control.
Is the Age of AI the same as the Artificial Era?
No. Age of AI is broad public and technological language without one authoritative formal definition. Artificial Era is a canonical Aisentica term with a specific historical-philosophical definition and a fixed beginning inside that framework. The AI Era vs Artificial Era article explains the distinction in full.
How old is artificial intelligence in 2026?
If AI is dated from the Dartmouth workshop in 1956, the field is about seventy years old in 2026. That count refers to the named research field, not to every intellectual precursor and not to the beginning of mass public use.
Conclusion: The Age of AI Began More Than Once
The most accurate answer to “When did the Age of AI begin?” is that AI has a sequence of beginnings. Turing’s 1950 paper crystallized the machine-intelligence question. Dartmouth in 1956 created the field’s institutional identity. Expert systems moved AI into knowledge-intensive reasoning. Neural networks and deep learning changed the capability trajectory. Transformers and large language models made general-purpose generative language systems possible. ChatGPT’s public release on November 30, 2022 turned those technical developments into a widely accessible conversational experience.
For the history of computer science, 1956 remains the strongest starting date. For the history of contemporary AI capability, 2012 and 2017 are indispensable. For psychology, 2022–2023 is the strongest practical boundary because AI moved from a field people heard about and an infrastructure people often used indirectly into an interactive environment in which ordinary users could delegate cognitive work, form judgments of trust and agency, and experience artificial systems as conversational objects.
That layered answer preserves the history instead of flattening it. The Age of AI is not one invention date. It is the point at which a long technical lineage became a human environment.