AI Winter and its Lessons Part 2

The Duration of AI Winter

During AI winter, AI research programs had to disguise themselves under different names to continue receiving funding. Many somewhat ambiguous words came up during this time that carried a strong hint of AI, such as “Machine Learning,” “Informatics,” “Knowledge-based system,” and “Pattern recognition.” The re-branding of these fields permitted AI to resume progress in the winter. However, fewer perceived advancements were under AI, further aggravating the overall support decline. 

The retail AI industry likely received a more severe impact in the winter. AI programs intrinsically require a large amount of computing power. In the early 70s, they started exceeding the standard research computer limit.

The situation was further exacerbated by the nature of LISP symbolic programming language, which was unsuitable for standard commercial computers optimized for assembly and FORTRAN language. So beginning in the 70s, many organizations started offering machines specially tailored to the semantics of LISP and could run more comprehensive AI programs. However, with the onset of AI winter, the industry saw a shift of interest away from the LISP programming language and machines. Associated with the opening of the “PC revolution,” many LISP organizations such as Symbolics, LISP Machines Inc, and Liquid Inc. failed as the place AI demand could no longer pay the premium for the specialized machines. 

The popularity of the LISP programming language originated mainly from academics, where fast prototyping and script-like semantics were very favorable. It achieved a limited range of success in commercial software development as there were inherent inefficiencies associated with functional programming languages. As a result, many new ideas pioneered by the LISP language, such as garbage collection, dynamic typing, and object-oriented programming, went into oblivion along with LISP. Although many new ideas were not specific to AI, they returned in the late 90s. 

The development of AI Winter showcased a cycle of positive feedback, wherein symptoms tend to exacerbate.

causes[WikiAIWinter]. The effect of the first triggers in the late 60s continued to be amplified throughout the next decades until activities in the AI ecosystem died down in the 80s. A special self-reinforcing element in this cruel cycle was the so-called “AI effect.” AI development refers to the trend for people to forgive advances in AI after the fact[AAAI]. When some intelligent behavior is achieved in a computer, the inner workings are as plain as lines of computer code. The secret is gone, and people quickly ignore the accomplishment as mere calculations. At the AAAI conference, Michael Kearns asserted that individuals possess an inherent desire to establish their unique position in the vast expanse of the universe at a subconscious level. This suggests that people are constantly seeking to define their identity and purpose in the world, which can be a powerful motivator for personal growth and development. Such insights can be valuable for researchers and scholars seeking to better understand the complex nature of human identity and behavior. In the past, AI was often mocked for being “almost functional.” This led to a growing sense of disappointment as AI research consistently fell short of its goals.


The downfall of AI was triggered by disappointing reports from both ALPAC and Lighthill. The ALPAC report gave a false impression that machine translation (MT) of good quality was much closer to reality than it actually was. This led to more liberal funding and sponsorship of MT research than was appropriate. The early success of the Georgetown-IBM experiment was also artificial, as grammar and vocabulary rules were created specifically for the text samples used, making the system appear in the best light. There was speculation that Leon Dosert had shown the system too early in order to secure funding for additional research at Georgetown. MT researchers only saw it as a first effort or prototype, but the press and funding agencies did not take notice of that fact. The success of miniature systems can be deceiving, especially when combined with enthusiastic media coverage. Lighthill raised concerns about the scalability of AI methods in his “Combinatorial Explosion” argument. Funding large-scale applications requires more than just success in specific domains.

The disappointments in the field of AI stemmed from a lack of understanding of the complexity of real-life problems. AI applications were often studied superficially, and there was a tendency to make overly optimistic predictions for publicizing purposes. John McCarthy, a pioneer of AI study, criticized the fact that much of the work in AI was not focused on studying intellectual mechanisms, but rather on generating amazement in the public. AI scientists frequently claimed to have discovered a general scheme of intelligent behavior that could be applied to all problem-solving, but these claims were often premature. Many formalisms led to predictions that computers would become intelligent by specific times, but none of them proved to be accurate. Researchers could have avoided disappointment if they had truly understood their AI algorithms’ inner workings and shortcomings and exercised caution when making hopeful claims about any supposed panacea scheme. 

Picture1 - Help Of Ai

Further, a deeper cause of the disappointments lies in the general lack of understanding of AI. The basis of the Lighthill report, i.e., the classification of AI technology, represented a standard view that AI was an applied science derived from biology. Such understanding inevitably led to high expectations of the productivity of AI technology. However, John McCarthy provided a more precise description of AI:

Narrowly applying AI methods to specific tasks or oversimplifying AI as merely replicating biological structures on a computer can lead to disappointment, as it neglects the larger intellectual hurdles in AI research.

Finally, many emerging computing technology besides AI also found similar wintery periods in their history. For example, the internet company reached via a roller-coaster in 2000. The commonality lies in the hypes often associated with new technologies rise and fall. The Hype Cycle, developed by the Gartner Group, provides a clear timeline for adopting new technologies.  

Figure 1. The topic being discussed is the Hype Cycle and AI Winter, as referenced in Menzies03.

According to Gartner, there are five phases in the model being illustrated.:

  1. Technological Triggers from events that generate significant press and interest.
  2. The peak of inflated expectations is marked by over-enthusiasm and more failures than successes.
  3. Trough of Disillusionment when technology was no longer fashionable.
  4. The slope of Enlightenment for people who persist in understanding technology’s benefits and practical applications.
  5. Plateau of Productivity as the benefits are widely accepted again.
Picture2 - Help Of Ai

The history of AI leading up to the AI Winter aligns perfectly with the model’s characteristics. National Conference on Artificial Intelligence attendance closely matched the Hype Cycle pattern. 

Figure 2. I attended the National Conference of Artificial Intelligence, as mentioned in the source Menzies03.

The prevalence of the Hype Cycle suggests that AI winter is determined by the nature of the technology and a common tendency in human cognition. Curiosity, excitement, and disappointment are all inherited parts of exploring the unknown. Without the initial great excitement and, thus, the subsequent heavy support, AI technology might not have taken off in the first place. Now that the hype around cooling down has subsided, individuals may have the opportunity to recognize the fundamental obstacles in this area and develop fresh perspectives for future research. Some advertising may be catalytic to the technology.  


In general, the examination of AI Winter brought to light a number of key takeaways:

Small-scale success in AI was deceptive. The intricate nature of AI suggests that numerous challenges can only be confronted and resolved when dealing with large-scale problems in real-world scenarios.

The study of AI presents a variety of complex intellectual hurdles to overcome. This is not restricted to particular applications or specific biological structures. It requires combined basic research in cognition, statistics, algorithms, linguistics, neurosciences, etc. 

Hype is a double-edged sword. It initially boosted the rise of AI but did great harm. Researchers, funders, and the public are responsible for restraining it so that the AI winter will not reoccur. 

Picture of Hoa

Leave a Comment