A Cycle of Enthusiasm and Disenchantment

The history of artificial intelligence (AI) has been marked by cycles of euphoria and disillusionment, known as "AI Winters." The first of these winters, between 1974 and 1980, represented a critical stage in the field's evolution: the end of overblown promises and the beginning of a deep reflection on its limits. Although AI was formally born in 1956 as the simulation of intelligent processes by machines, the gap between expectations and actual results became unsustainable, causing a drastic reduction in interest and funding.

The Golden Age and the Peak of Expectations

During the 1950s and 1960s, optimism was palpable. Researchers like Marvin Minsky, John McCarthy, and Allen Newell claimed that the AI problem would be solved within a generation. Funding flowed, especially from the U.S. Department of Defense (ARPA/DARPA), which saw AI as a strategic tool. Pioneering systems like ELIZA (1966), Logic Theorist, and SHRDLU were developed, operating in controlled "microworlds." However, these advances failed to generalize to real environments, revealing an underestimation of the complexity of human thought.

Technical Limitations and the Combinatorial Explosion

The initial enthusiasm collided with deep technical barriers. Systems were fragile, dependent on rigid rules, and not very adaptable. Problems such as common-sense reasoning, the "frame problem," and the combinatorial explosion—where the number of possibilities grows exponentially with the size of the problem—made the algorithms impractical. Furthermore, the computational power of the era was insufficient to support more ambitious models, severely limiting progress.

The Authoritative Blow of the Lighthill Report

In 1973, the British Parliament commissioned mathematician Sir James Lighthill to evaluate the state of AI. The result was devastating: the Lighthill Report harshly criticized the lack of significant advances and the disconnect between grandiose goals and actual results. It recommended massive cuts in public funding, leading to the closure of almost all AI projects in the United Kingdom. This report became the symbol of institutional disenchantment with the discipline.

The Scientific Philosophy Behind the Critique

Lighthill argued that the best science should be linked to practical problem-solving. He advocated for "working worlds," i.e., contexts like health or defense where problems were concrete and urgent. His critique was not only technical but philosophical: he questioned the legitimacy of research without a clear application. This stance contrasted with the view of some researchers who saw AI as a tool to explore the theory of knowledge, beyond its immediate utility.

Lighthill's Categories: The Failure of the "Bridge"

To structure his analysis, Lighthill divided AI into three categories: A (advanced automation), C (central nervous system simulation), and B (bridge robotics). He was especially critical of category B, which sought to build generalist intelligent systems. According to him, this category was a "failed bridge" between theory and application, without a defined field of use. This classification influenced how AI projects were evaluated in the following years.

The Contagion Effect on Global Funding

Although the Lighthill Report was British, its impact was felt globally. In 1974, DARPA in the U.S. also drastically cut funding for fundamental AI research. The discipline fragmented, and many researchers migrated to safer areas like theoretical computer science, statistics, or software engineering. The loss of scientific credibility was profound, and the term "artificial intelligence" began to be avoided in academic publications.

The Defense of the Theoretical Paradigm Against Application

The scientific community responded firmly. Donald Michie, an AI pioneer in Edinburgh, criticized Lighthill's classification as misleading. He proposed that category B should be understood as "Theory of Intelligence," whose goal was to build "epistemoscopes": instruments for discovering new theories. Michie argued that the real motivation for many researchers was epistemological, not utilitarian, and that demanding immediate applications was a reductionist view of science.

The Hidden Seeds of Maturation

Despite the setback, the First Winter allowed the field to mature. Exaggerated promises were abandoned, and a more critical attitude was adopted. Between 1974 and 1980, the foundations for statistical learning, algorithmic optimization, and the development of more robust architectures were laid. The Neocognitron, proposed by Kunihiko Fukushima in 1979, anticipated the convolutional neural networks that would revolutionize computer vision decades later.

The Renaissance Through Expert Systems

Starting in 1980, AI re-emerged thanks to Expert Systems, which focused on specific, well-defined tasks. These systems emulated the decision-making of human experts through rules and knowledge bases. Examples like MYCIN (medical diagnosis) and DENDRAL (chemical analysis) showed that AI could be useful in concrete domains. This pragmatic approach regained institutional trust and attracted new funding.

From General Intelligence to Narrow Productivity

The failure of general AI led to a paradigm shift: weak or narrow AI, focused on specific tasks. This form of AI has proliferated in commercial applications such as virtual assistants, recommendation engines, and computer vision systems. Although AGI remains a distant goal, narrow AI has reached the "plateau of productivity," generating tangible value in multiple sectors.

Adversarial Search and Technical Continuity

Even in times of recession, technical research continued. In the field of problem-solving, search algorithms—informed and uninformed—continued to evolve. In competitive environments, such as games, adversarial algorithms like Minimax and Alpha-Beta pruning were developed, allowing agents to make optimal decisions against intelligent opponents. These techniques remain fundamental in modern AI, from video games to automated negotiation.

Sources and Further Reading

The historical information regarding the First AI Winter is primarily based on the analysis of two key events:

The Lighthill Report (1973): Formally titled "Artificial Intelligence: A Paper Symposium," this report commissioned by the British Science Research Council was pivotal in the reduction of AI funding in the UK. A retrospective analysis can be found in publications like "The Lighthill Report – an overture to AI winters" by The Royal Society.
DARPA's Funding Cuts: In the mid-1970s, the US Defense Advanced Research Projects Agency (DARPA) shifted its funding strategy towards more directed, application-oriented projects, significantly reducing support for the foundational, exploratory AI research that had characterized the previous decade.