Illusion or Infrastructure Revolution: Breaking Down the Core Skepticism Around AIArticle · digest for October 18
Surrealist still life combining a 19th-century railway switch with a monumental computing cluster.
Article4 sections≈ 4 min

Illusion or Infrastructure Revolution: Breaking Down the Core Skepticism Around AI

You have probably scrolled past plenty of apocalyptic takes in your feed lately: "AI is a giant Ponzi scheme," "LLMs understand nothing and just stitch other people's words together," "VCs burned trillions for nothing and the whole thing is about to crash."

When marketing hype around a technology hits ear-splitting volume, fierce skepticism is completely natural. My own eye twitches every time another "startup" slaps a thin wrapper onto someone else's API and claims a hundred-million-dollar valuation. We have dug into that underbelly before, looking at modern AI anatomy from silicon to agentic systems and AI's real business impact through Jevons paradox and hidden costs.

Still, when criticism turns into flat denial of the underlying technology, it pays to step back and apply cold logic. Entrepreneur Nick Saraev recently dissected the most viral anti-AI manifestos, and his arguments are worth walking through honestly.

The Intelligence Test: How Do We Actually Measure Mind?

One of the critics' favorite lines goes like this: "A neural network has no real consciousness—it merely mimics the outputs of intelligent behavior."

Sounds profound. Yet pause for a second: how does human society judge another person's intelligence?

Neither you nor I can crack open a colleague's skull to touch their subjective "consciousness" or measure their "soul" with a ruler. In universities, engineering interviews, medical boards, and PhD defenses, we evaluate intelligence strictly through output-based evaluation:

  • Can the person solve the math problem in front of them?
  • Can they write a coherent, logically consistent argument?
  • Can they spot a subtle bug in someone else's code?

When a human passes those tests with flying colors, we call them competent. When a neural network does the exact same thing, skeptics immediately move the goalposts: "That doesn't count, it just memorized patterns." It is the classic retreating-horizon trick—the moment a machine masters chess, translation, or programming, that domain is promptly reclassified as "mechanical work that never required real intelligence anyway."

Novel Problems vs. Rote Memorization: Silver at the International Math Olympiad

A second popular myth claims: "AI can only recombine what was already in its training set."

Recent mathematical breakthroughs shattered that claim. DeepMind's AlphaProof and AlphaGeometry 2 tackled brand-new problems from the International Mathematical Olympiad (IMO) at a silver-medal level, scoring 28 out of 42 points.

IMO problems are drafted in strict secrecy right before the competition—they physically did not exist on the internet or in any training dataset. To solve them, the systems couldn't just generate plausible-sounding prose; they had to construct formal, mathematically rigorous proofs in the Lean proof assistant, where a single logical flaw triggers an immediate compiler rejection. Those proofs were checked and verified by leading mathematicians—Fields Medalist Sir Timothy Gowers and Dr. Joseph Myers—and published in Nature. Calling that "random word-stitching" simply ignores the scientific record.

History's Lesson: The 19th-Century Railway Mania

What about the financial bubble and eye-watering data-center spending?

Here Nick Saraev draws a sharp historical parallel to the "Railway Mania" of the 1840s in Britain and the United States. Back then, investors frantically bought shares in hundreds of rail companies, newspapers declared a new age, and then the bubble burst with a loud pop. Speculators went broke and shady startups went to zero.

Yet what happened to the railroads themselves? They didn't vanish. The tracks, ties, bridges, tunnels, and stations stayed right where they were. They linked cities, slashed freight costs by orders of magnitude, and laid the physical foundation for the greatest industrial expansion in history.

The exact same dynamic is playing out with generative AI today. The VC froth around empty wrapper startups will inevitably evaporate, and investors will reset inflated timelines for ROI. What remains is the foundational infrastructure—massive data centers, upgraded power grids, silicon clusters, and trained base models.

Separating Speculative Froth from Core Utility

Stock-market hype and a tool's practical usefulness are two completely different things. Share prices can swing wildly, yet if an engineer saves two hours every day using AI, the value of that saved time is 100% real.

Voting with Their Wallets

Finally, look at the most down-to-earth counterargument to the "total scam" narrative: the user base.

ChatGPT now counts over 900 million active users, and more than 50 million paying subscribers.

Millions of founders, developers, copywriters, teachers, and marketers hand over their own money every single month for a subscription. People will fall for hype once out of curiosity, sure—but nobody keeps renewing a paid plan month after month unless the tool pays for itself and saves real time in daily work.

AI isn't a magic wand, and it isn't a replacement for the human mind. It is, however, a massive productivity lever actively reshaping the global economy. While some people write manifestos explaining why it "shouldn't work," plenty of others are simply picking up the tool and getting their work done faster.

Want a deep diveVotes help us choose what to break down in detail. This material has 0 votes.
Sources · 3