We Have Been Here Before: The Edge of Impressive Is Not the Edge of Everything

How the brightest minds of AI's first revolution mistook capability for emergence and why we may be making the same mistake again.

AI could easily be the first contact of many people with an incredibly transformative and undeniably powerful technology. I do understand the impact it has on our society and how impressive it is to experience the consequences of its “magic.”

However, this is old news, and understanding what happened in the not-so-distant past is crucial for discerning what is happening now and for having a sober, grounded, and healthy view of what most likely will happen in the future.

This position is not about being a downer, or an “AI denier,” or some other labels people are quick to use when anyone dares to throw some light on the AI hype. Labels sometimes are important, and I use them, but at the same time, they mean nothing if not supported by facts.

So this article is about facts. Facts that cannot be ignored if we truly want to understand what is currently happening in the AI world. After all, life can be ironically cyclical.

Scenes of a Movie We Watched Before

What do technologies like expert systems and good old-fashioned artificial intelligence (GOFAI), fusion energy, the semantic web, and virtual reality, to name just a few, have in common? They all share a structurally identical pattern:

• Real capability demonstrated in controlled conditions

• Brilliant, credentialed advocates extrapolate to transformation

• The gap between impressive in-domain and generalized emergence is systematically underestimated

• Adjacent, unpredictable technologies solve the actual problems instead

• The original vision is partially realized as useful infrastructure, not as the emergent phenomenon promised

Sounds familiar?

I won't talk about all of them in this article. Instead, I will focus on expert systems and GOFAI since the similarities with the current state of practice of AI are nothing short of outstanding.

The Role of the Technology Advocate

As I said in an earlier article, every human genius is more human than genius. Similarly, when we take a look at the historical events that shaped what our technology is today, we can clearly see that a technological revolution, as a movement, is more about humans than technology.

Even consequential technologies can remain invisible for decades if, for instance, secrecy demanded it. The first prominent example of such a case is ARPANET, the precursor to the Internet, which was operated as a classified DARPA project for years before it became public.

Another case worth mentioning is the global positioning system (GPS), which was a Department of Defense military program for ten years before civilians were allowed access in 1983. Unrestricted use only came in 2000.

Stealth technology, developed through Lockheed's classified Skunk Works division throughout the 1970s, flew operational missions before most engineers in the broader aerospace industry knew it was physically possible.

These technologies transformed civilization quietly, through institutional patience and public funding, without needing a prophet. But that is the exception. Most technological revolutions depend entirely on their advocates being loud, persuasive, and relentless, because without visibility there is no capital, and without capital there is no progress.

The Danger

Under that kind of pressure, though… the line between confidence and exaggeration becomes dangerously thin.

Elizabeth Holmes raised nearly a billion dollars on technology that did not work and could not work, and sustained the illusion through calculated deception for over a decade.

Elon Musk has predicted "complete" Full Self-Driving capability every year since 2016, consistently moving the horizon forward without ever acknowledging the previous prediction's failure.

And the entire dot-com era was sustained by founders and executives making revenue projections to investors that bore no relationship to any underlying commercial reality.

This list of events is far from being exhaustive. And yet, they deserve to be highlighted because advocates with strong beliefs and a platform will not think twice to make audacious predictions, and those predictions is what moves money, redirect careers, reshape national priorities, and when they fall short, leave behind not just disappointment but measurable, sometimes catastrophic, losses for individuals, organizations, and entire governments that chose to believe them.

The Advocates and Their Claims

Herbert Simon, Nobel Prize in Economics 1978, Carnegie Mellon, is one of the founders of AI as a discipline. In 1965, he claimed Machines would be capable, within twenty years, of doing any work a man could do. In fact, he went much further. He said that in less than twenty years from that time, machines would replace all human functions in organizations. Sounds familiar?

Marvin Minsky, Turing Award 1969, co-founder of MIT’s AI Lab, MIT, wrote a book in 1967 entitled Computation: Finite and Infinite Machines, where he claims:

Within a generation, I am convinced, few compartments of intellect will remain outside the machine's realm—the problems of creating 'artificial intelligence' will be substantially solved.

He was so confident that he served as scientific advisor to Stanley Kubrick’s 2001: A Space Odyssey, vouching for HAL 9000 as a plausible near-future machine.

However, around 1982, Minsky acknowledged that the AI problem was one of the hardest science had ever undertaken, and the “generation” timeline proved to be overly optimistic.

John McCarthy, Turing Award 1971, Stanford, coined the term “Artificial Intelligence” in 1955. He organized the 1956 Dartmouth Conference, the foundational event of AI, with a proposal suggesting that “every aspect of learning or any feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.” The proposal estimated this could be meaningfully achieved in a single summer by a small group.

The Technologies that Got Them Excited

In 1956, Newell and Simon created Logic Theorist, which proved 38 of 52 theorems in Whitehead and Russell’s Principia Mathematica. Simon believed it was a proof of concept for machine reasoning.

The following year, Newell and Simon were joined by programmer J.C. Shaw, and they designed the General Problem Solver, an ambitious attempt to build a universal reasoning engine capable of simulating human problem-solving across any domain. It handled toy problems elegantly. When confronted with the complexity of the real world, it completely failed.

But the failures did not slow the field. In 1972, a Stanford doctoral student named Edward Shortliffe built MYCIN, an expert system that diagnosed bacterial infections with an accuracy that matched (and in some evaluations exceeded) that of senior physicians. It was a genuine achievement, and it made believers out of skeptics. If a machine could reason through a medical diagnosis, what else could it do?

By 1980, that question had reached the corporate world. John McDermott at Carnegie Mellon developed XCON, also known as R1, for the Digital Equipment Corporation. It configured complex computer orders automatically, and it worked, saving DEC an estimated $40 million per year.

Between 1982 and 1992, Japan’s Fifth Generation Computer Project was a $400 million government-funded initiative explicitly designed to produce thinking machines within a decade. It caused global alarm, and as a response, DARPA launched a counter-initiative, the Strategic Computing Initiative, fearing technological defeat. Both sides seemed to agree that the machines were ready to think.

What They Got Wrong

We have an advantage over every single person and organization mentioned in this article: we know the end of the story. They didn't. So we can now just look back and analyze what they got wrong.

The assumption was that intelligence was fundamentally about rules and logic. That is, if you could encode enough knowledge explicitly, understanding would emerge.

What they discovered instead is what philosopher Hubert Dreyfus (UC Berkeley) had argued since 1965 in his paper Alchemy and Artificial Intelligence: human expertise is not reducible to explicit rules. It depends on embodied experience, context, and what Michael Polanyi called “tacit knowledge”, things we know but cannot fully articulate.

Dreyfus was ridiculed by the AI community. Simon and Minsky severely dismissed him. So much so, Dreyfus was barred from speaking at MIT. Think about this for a moment. However, guess who was right? Time vindicated Dreyfus completely.

What Time Taught

As the direct consequence of a consistent and catastrophic series of misleading predictions and expectations, funding just stopped. In 1973, the British government’s Lighthill Report dismantled most UK AI research. Japan’s Fifth Generation project was quietly wound down in 1992, having produced none of its central goals. XCON and MYCIN worked. They were actually useful, but the intelligence so fiercely advocated never generalized. They were expert systems, not a form of intelligence.

What We Can All Learn From History

Let me repeat what I said before because I really hope you can remember this: we have an advantage over the AI advocates of the past. We now know what happened.

And why does this matter so much? Just by observing these historical events and comparing them to what is happening now, it is not a stretch to believe that most advocates (I am not sure if all) start by believing what they know is true.

But then, when the claims are so extraordinary, and the pressure becomes overwhelming (competition, investors, media, reputation, etc.), there seems to be a moment when the advocate is confronted with reality, but they are already too deep into the rabbit hole to just turn back. At least some of them will revisit their beliefs and projections and provide a more critical view of everything they said, as it happened with Minsky.

When the “AI” subject becomes a fever, everyone will be quick to have an opinion. From what I have seen through an abundant number of examples, these opinions are mostly about feelings and personal beliefs.

Unless you invoke faith, conviction is useless without facts and proofs. It is only prudent to be judicious about AI.