Reviewing 80 years of AI development, these 5 historical lessons are worth learning
TechFlow Selected TechFlow Selected
Reviewing 80 years of AI development, these 5 historical lessons are worth learning
The lessons learned from 80 years of artificial intelligence development might help AI companies navigate the ups and downs of the next 30 days or 30 years.
Navigating Web3 tides with focused insightsBy: Gil Press
Translated by: Felix, PANews
On July 9, 2025, Nvidia became the first publicly traded company to reach a market capitalization of $4 trillion. What comes next for Nvidia and the turbulent field of AI?
Although predictions are difficult, there is ample data available. At the very least, it can help clarify why past forecasts failed—where, how, and why they fell short. This is history.
What lessons can be drawn from the 80-year journey of artificial intelligence (AI)? A journey marked by fluctuating funding, vastly different research and development approaches, and public sentiment swinging between curiosity, anxiety, and excitement.
The history of AI began in December 1943, when neurophysiologist Warren S. McCulloch and logician Walter Pitts published a paper on mathematical logic. In "A Logical Calculus of the Ideas Immanent in Nervous Activity," they theorized idealized and simplified networks of neurons and how they could perform simple logical operations by either transmitting or not transmitting impulses.
Ralph Lillie, who was pioneering the field of histochemistry at the time, described McCulloch and Pitts' work as giving "reality" to "logical and mathematical models" in the absence of "experimental facts." Later, when the paper's assumptions failed empirical testing, MIT's Jerome Lettvin noted that while the fields of neurology and neurobiology ignored the paper, it inspired "those who were destined to become enthusiasts of a new field—now known as AI."
In fact, McCulloch and Pitts' paper inspired "connectionism"—a specific variant of AI that now dominates the field, today called "deep learning," and more recently rebranded simply as "AI." Although this approach has no relation to how the brain actually works, the statistical analysis methods underpinning this AI variant—"artificial neural networks"—are commonly described by AI practitioners and commentators as "mimicking the brain." Leading authority and top AI practitioner Demis Hassabis declared in 2017 that the fictional descriptions of brain function by McCulloch, Pitts, and similar studies "continue to form the foundation of contemporary deep learning research."
Lesson one: Beware of conflating engineering with science, science with speculation, and science with papers full of mathematical symbols and formulas. Above all, resist the temptation of the illusion that "we are like gods"—the belief that humans are no different from machines and can create machines as intelligent as themselves.
This persistent and widespread arrogance has been a catalyst for technological bubbles and periodic AI frenzies over the past 80 years.
This brings to mind the idea of general AI (AGI)—machines that will soon possess human-like or even superhuman intelligence.
In 1957, AI pioneer Herbert Simon declared: "There are now machines that think, learn, and create." He also predicted that within ten years, computers would become world chess champions. In 1970, another AI pioneer, Marvin Minsky, confidently stated: "In three to eight years, we will have a machine with the general intelligence of an average human... Once the computer takes control, we may never get it back. We would survive at its mercy. If we're lucky, it might decide to keep us as pets."
Expectations of imminent AGI have been so significant that they influenced government spending and policy. In 1981, Japan allocated $850 million to its Fifth Generation Computer project, aiming to develop machines that think like humans. In response, after a long "AI winter," the U.S. Defense Advanced Research Projects Agency planned in 1983 to restart AI funding to build machines that could "see, hear, speak, and think like humans."
It took enlightened governments around the world about a decade and billions of dollars to gain not only a clear-eyed view of AGI but also an understanding of the limitations of traditional AI. But by 2012, connectionism finally triumphed over other AI schools, and a new wave of predictions about the imminent arrival of AGI swept the globe. OpenAI claimed in 2023 that superintelligent AI—"the most impactful invention in human history"—might arrive within this decade and "could lead to humanity losing power, or even human extinction."
Lesson two: Be wary of shiny new things. Examine them carefully, cautiously, and wisely. They may not differ greatly from previous speculations about when machines would achieve human-like intelligence.
Yann LeCun, one of the "fathers" of deep learning, said: "We are missing some critical ingredients to make machines learn as efficiently as humans and animals do—we just don't know what they are yet."
For decades, AGI has consistently been described as "just around the corner," largely due to the "first step fallacy." Machine translation pioneer Yehoshua Bar-Hillel was among the first to discuss the limitations of machine intelligence, pointing out that many people believe that if someone demonstrates a computer doing something previously thought impossible—even poorly—it’s assumed that further technical progress will eventually perfect the task. It's widely believed that patience alone will bring success. But Bar-Hillel warned as early as the mid-1950s that this is not true—and reality has repeatedly proven him right.
Lesson three: The distance from being unable to do something to doing it poorly is usually much shorter than the distance from doing it poorly to doing it well.
In the 1950s and 1960s, many fell into the "first step fallacy" due to improvements in semiconductor processing speeds powering computers. As hardware reliably advanced along Moore's Law year after year, it was widely assumed that machine intelligence would advance in tandem.
However, beyond continuous hardware improvements, AI entered a new phase with the introduction of two new elements: software and data collection. Starting in the mid-1960s, expert systems—a type of intelligent computer program—shifted focus toward acquiring and programming real-world knowledge, especially the knowledge and heuristics (rules of thumb) of domain experts. Expert systems grew increasingly popular, and by the 1980s, it was estimated that two-thirds of Fortune 500 companies used the technology in daily operations.
Yet by the early 1990s, this AI boom had completely collapsed. Numerous AI startups went bankrupt, and major companies froze or canceled their AI projects. As early as 1983, expert systems pioneer Ed Feigenbaum identified the "key bottleneck" leading to their demise: the scalability of the knowledge acquisition process, "an extremely tedious, time-consuming, and expensive endeavor."
Expert systems also faced challenges in knowledge accumulation. The constant need to add and update rules made them difficult and costly to maintain. They also exposed fundamental flaws of thinking machines compared to human intelligence. They were "brittle," making absurd errors when faced with unusual inputs, unable to transfer expertise to new domains, and lacking understanding of the world around them. At the most fundamental level, they couldn’t learn from examples, experience, or environment as humans do.
Lesson four: Initial success—widespread adoption by enterprises and government agencies, along with massive public and private investment—does not necessarily give rise to a lasting "new industry," even after ten or fifteen years. Bubbles tend to burst.
Amid cycles of hype and disappointment, two fundamentally different approaches to AI development have competed for attention from academia, public and private investors, and the media. For over forty years, rule-based symbolic AI dominated. But instance-based, statistically driven connectionism—as an alternative main approach—briefly rose to prominence and popularity in the late 1950s and again in the late 1980s.
Prior to the resurgence of connectionism in 2012, AI research and development were primarily driven by academia. Academia was characterized by dogma ("normal science"), enforcing an either-or choice between symbolic AI and connectionism. In his 2019 Turing Award lecture, Geoffrey Hinton spent most of his time recounting the hardships he and a small group of deep learning enthusiasts endured at the hands of mainstream AI and machine learning scholars. Hinton also deliberately downplayed reinforcement learning and the work of his colleagues at DeepMind.
Just a few years later, in 2023, DeepMind took over Google's AI operations (and Hinton left), largely in response to OpenAI's success, which included reinforcement learning as part of its AI development. Reinforcement learning pioneers Andrew Barto and Richard Sutton received the Turing Award in 2025.
Yet currently, there is no indication that either DeepMind or OpenAI—or the numerous "unicorn" companies dedicated to AGI—have shifted their focus beyond the prevailing paradigm of large language models. Since 2012, AI's center of gravity has moved from academia to the private sector; however, the entire field remains fixated on a single research direction.
Lesson five: Don't put all your AI eggs in one basket.
Undoubtedly, Jensen Huang is an outstanding CEO, and Nvidia is an outstanding company. Over a decade ago, when the AI opportunity suddenly emerged, Nvidia quickly seized it because its chips—originally designed for efficiently rendering video games—were well-suited for deep learning computations due to their parallel processing capabilities. Huang remains vigilant, telling his employees: "Our company is only 30 days away from bankruptcy."
Besides staying vigilant (remember Intel?), the lessons from 80 years of AI development might also help Nvidia navigate the ups and downs of the next 30 days—or 30 years.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
@GilPress
