The term "AI Winter" was first used in 1984 during a public debate at the annual meeting of the American Association of Artificial Intelligence, reflecting periods of reduced enthusiasm and funding for artificial intelligence research.5 These winters have been characterized by a cycle of high expectations followed by profound disappointment, leading to significant reductions in funding and interest in AI.

The first AI winter occurred from 1974 to 1980, triggered by a combination of factors including the publication of the "Lighthill Report," which criticized the overly ambitious promises of AI research, and the subsequent withdrawal of funding by major agencies like DARPA in the United States and similar actions in the United Kingdom.24 This period saw a significant decline in AI research activities as the initial excitement generated in the 1950s and 1960s waned.

A second AI winter took place from the late 1980s to the mid-1990s. This downturn was largely due to the limitations of expert systems, a form of AI technology that was popular at the time but failed to live up to the high expectations. These systems were initially seen as a promising avenue for AI but soon encountered significant practical and technical limitations, leading to disillusionment and a subsequent reduction in funding and interest.2

Despite these setbacks, AI has seen periods of resurgence, often termed as "AI summers," where new advancements and approaches, such as deep learning and improvements in computational power, have revitalized interest and investment in the field. These advancements have helped to overcome some of the limitations that led to previous winters, suggesting a pattern of cyclical highs and lows in AI research and development.4

Overall, the history of AI winters underscores the challenges of balancing expectations with the actual pace and capabilities of technological advancements in AI. It also highlights the importance of sustainable and realistic approaches to developing AI technologies to prevent future periods of stagnation.

The 1966 failure of machine translation marked a significant early setback in the field of artificial intelligence, contributing to the onset of the first AI winter. This event was highlighted by the ALPAC report, which critically assessed the state of machine translation and concluded that the technology had failed to meet the expectations of general scientific text translation, casting doubt on its immediate prospects. This report led to a significant reduction in funding from key sponsors like the Department of Defense, the National Science Foundation, and the CIA, thereby dampening enthusiasm and development in AI research during this period.12

The criticism of perceptrons in 1969 marked a pivotal moment in the history of artificial intelligence, particularly in the field of neural networks. Perceptrons, initially introduced by Frank Rosenblatt in 1958, were simple neural networks capable of performing various logical operations but failed to handle non-linear separability, such as the XOR problem. This limitation was highlighted in Marvin Minsky and Seymour Papert's seminal work, "Perceptrons," which argued that these networks could not compute a simple XOR function, casting doubt on their utility for more complex tasks. This critique significantly influenced the perception of neural networks, contributing to the onset of the first AI winter by fostering skepticism about the broader capabilities of neural networks12.

The period from 1971 to 1975 marked a significant chapter in the history of AI, characterized by DARPA's growing frustration with the Speech Understanding Research (SUR) program at Carnegie Mellon University. Despite an ambitious start and substantial funding, the program struggled to meet the high expectations set for it. The technology's inability to effectively understand and process natural language in real-world scenarios led to doubts about its practical applications. This dissatisfaction contributed to DARPA's decision to cut back funding, not only affecting the SUR program but also casting a shadow over the perception of AI's effectiveness in practical applications during this time.12

The consequences of these developments were far-reaching, contributing to the onset of an AI winter. The reduction in funding and support from a key agency like DARPA signaled a broader lack of confidence in AI's immediate utility, influencing other funding bodies and stakeholders across the field. This period highlights the critical impact of government and institutional support in the advancement of AI technologies and the vulnerabilities associated with dependency on such funding.12

The Lighthill Report, published in 1973, played a pivotal role in shaping the trajectory of artificial intelligence research in the United Kingdom. Commissioned by the British Science Research Council, the report offered a critical evaluation of AI's progress, particularly criticizing the field for not fulfilling the ambitious promises it had made. Sir James Lighthill highlighted significant shortcomings in foundational areas such as robotics and language processing, and pointed out the issue of combinatorial explosion, which he argued AI methods were not equipped to handle at scale. His skepticism about the practical applications of AI led to a substantial reduction in funding, marking the beginning of what is often referred to as an "AI winter" in the UK.12

This period of reduced funding and support had a profound impact on AI research within the country, stifling development and innovation. The report's influence was not limited to financial aspects; it also affected the morale and confidence within the AI community, leading to a decade-long period of cautious and limited advancement in the field. The repercussions of the Lighthill Report underscore the significant influence of governmental and institutional reviews on the direction and momentum of scientific research.12

In the early 1970s, the Defense Advanced Research Projects Agency (DARPA) significantly altered its funding strategy for artificial intelligence research, leading to substantial cutbacks in academic AI projects. This shift was influenced by the Mansfield Amendment of 1969, which mandated that DARPA fund only "mission-oriented direct research" rather than "basic undirected research." This legislative change marked a departure from the previous approach of providing broad, flexible funding, which had allowed researchers considerable freedom to explore diverse AI concepts without immediate practical applications.

The impact of these cutbacks was profound, particularly in academic settings where foundational AI research was being conducted. Universities and research institutions that had previously enjoyed generous funding faced stringent requirements to demonstrate the near-term military utility of their projects. This new funding environment led to the cancellation or scaling down of several ambitious AI projects. For instance, DARPA's decision to reduce funding for AI research was partly influenced by critical evaluations such as the Lighthill report and DARPA's own internal reviews, which questioned the feasibility of achieving significant breakthroughs in AI technology in the foreseeable future.

The reduction in funding not only slowed the pace of AI research but also shifted the focus of remaining projects towards more application-specific goals, such as autonomous tanks and battle management systems. While some projects like the Dynamic Analysis and Replanning Tool (DART) eventually proved successful, many others did not meet their objectives, leading to a broader disillusionment with AI's potential during this period. This phase significantly contributed to the onset of what is known as an "AI winter," a period marked by reduced funding, diminished interest, and skepticism about the field's prospects. The consequences of DARPA's funding cutbacks underscored the vulnerability of AI research to shifts in policy and funding priorities, highlighting the complex interplay between technological ambition and practical feasibility in the evolution of AI.

The collapse of the LISP machine market in 1987 marked a significant turning point in the history of artificial intelligence, contributing to the onset of the second major AI winter. LISP machines, which were specialized computers designed to efficiently run the LISP programming language, faced a dramatic decline in demand as more versatile and cost-effective general-purpose computers from companies like Apple and IBM began to match their performance capabilities. This shift was exacerbated by the high costs associated with maintaining LISP machines and the broader availability of LISP programming environments on standard computers, which made the specialized hardware redundant and economically unviable.12

The market collapse not only affected the companies that manufactured these machines, such as Symbolics and LISP Machines Inc., but also had broader implications for the AI industry. It underscored the vulnerability of specialized AI hardware to rapid advancements in general computing technology and market dynamics. The event highlighted the importance of adaptability and the risks associated with over-reliance on niche technologies in the rapidly evolving field of AI. This period is an instructive example of how technological and economic factors can precipitate a sector-wide reassessment and realignment, leading to significant downturns in industry enthusiasm and investment.13

The cancellation of new spending on AI by the Strategic Computing Initiative (SCI) in 1988 marked a significant turning point in the history of artificial intelligence, contributing to the onset of the second major AI winter. This decision by the U.S. government reflected a broader skepticism about the short-term potential of AI technologies, particularly in the wake of high-profile disappointments and unmet expectations from earlier ambitious projects. The cut in funding was a response to the perceived overreach in AI capabilities, leading to a reevaluation of investment priorities and a significant reduction in support for AI research at a critical time.12

This move by the SCI not only impacted the specific projects under its purview but also sent a chilling effect across the broader AI research community. The withdrawal of funding led to the abandonment of many ongoing AI projects and discouraged new initiatives, as the financial backing and institutional support necessary for such research became scarce. This period highlights the profound influence of government policy and funding strategies on the direction and momentum of technological innovation, particularly in fields as resource-intensive and speculative as artificial intelligence.13

The 1990s marked a significant downturn for expert systems, a branch of AI that had initially shown promise in automating decision-making processes using rule-based systems. Despite their early success in specific domains, these systems struggled to adapt to the complexity and variability of broader real-world applications. The maintenance of expert systems proved to be costly and labor-intensive, primarily because they required constant updates to their knowledge bases, which were not automated. This inability to scale and adapt led to a widespread abandonment of expert systems during this decade, contributing to the broader AI winter of the late 1980s and 1990s. The disillusionment was compounded by the systems' failure to meet the high expectations set during their introduction, resulting in a significant reduction in both interest and investment in this AI technology.123

The Fifth Generation Computer Project, initiated by Japan in the early 1980s, was a highly ambitious endeavor aimed at revolutionizing computing through the development of machines that could engage in human-like reasoning and knowledge processing. Despite the significant investment of approximately ¥54 billion, the project fell short of its transformative goals by the early 1990s. The technologies developed, particularly the parallel inference machines, did not achieve the anticipated impact on the computing industry or AI development. This shortfall led to a broader reassessment of AI's potential and contributed to the onset of an AI winter during this period. The project's conclusion marked a pivotal moment in AI history, underscoring the challenges of aligning ambitious technological goals with practical outcomes.12

The period from 2000 to 2020 marked a significant shift in the landscape of artificial intelligence, moving away from the traditional AI winters experienced in previous decades. This era was characterized by rapid advancements in AI technologies, particularly in machine learning, deep learning, and neural networks, which spurred renewed interest and investment in the field. The development of more powerful computational hardware and the availability of large datasets facilitated these advancements, leading to successful applications in various domains such as image and speech recognition, natural language processing, and autonomous vehicles. This period is often referred to as an "AI spring" or "AI resurgence," as the field not only recovered from previous setbacks but also began to realize some of its earlier promises, significantly impacting both industry and society.12

Moreover, the integration of AI into mainstream technology and business practices became more pronounced during this time. Companies like Google, IBM, and Microsoft heavily invested in AI research and development, leading to the creation and improvement of AI products that are now ubiquitous in everyday technology. This era also saw the rise of AI in the public consciousness, partly due to these companies' efforts to commercialize AI technologies, making them accessible to a broader audience. The shift from AI winters to a more sustained period of growth and acceptance of AI technologies marked a maturation of the field, setting the stage for ongoing innovations and applications.23