🧠 AI Superpowers — Reflections on “Utopia, Dystopia, and the Real AI Crisis”

This was by far the most important chapter for me. Not because it predicted some cinematic robot takeover, but because it struck at something far more real: the economic, psychological, and social tsunami that AI might bring. It’s actually why I picked up this book in the first place — to confront my own pessimistic vision of the future, sharpened by the recent mind-blowing progress of models like GPT.

🤖 It started with self-checkouts

Back in 2019–2020, I remember noticing advancing automation in various places — grocery stores with self-checkouts were just one example. Most people saw convenience — I saw job loss. Low-skilled workers getting quietly phased out by automation. Of course, I understood that humanity had always functioned this way, just like with the Luddites in the UK back in the day, but ultimately it was supposed to lead to a “better tomorrow” — though this social aspect was always a red flag for me.

At the time, I still felt safe in my “knowledge worker” bubble. Even as automation crept into IT and digital workflows, I thought my skill set was too specialized, too complex to be replaced.

That illusion collapsed in 2024 when I saw what GPT-4 could do — and how fast it got there. The last mindblowing moment for me was realizing how good GPT and Claude are at psychology (I always treated these tools dismissively in this aspect). Suddenly, it wasn’t just others whose jobs were on the line. It was mine. And probably yours too.

🧪 Forget AGI. The real crisis is already here.

Kai-Fu Lee opens this chapter with a discussion on AGI — Artificial General Intelligence — and the possible risks of a superintelligent AI. But even the techno-optimists admit: if AGI ever posed a threat, it wouldn’t be a robot uprising, but something subtle and systemic, like an undetectable virus hijacking critical systems.

Still, that’s not the real issue.

As Lee clearly states: there’s no scientific or technological foundation today to suggest AGI is around the corner. I fully agree with this from the perspective of someone who has already dipped into AI and delved into ML technicalities. AI research has been around for over 70 years. Breakthroughs have been rare and scattered. Deep learning was one such leap, and people now blindly extrapolate it into the future — just like a rookie salesman who assumes he’ll grow 20% every week forever because the last two were good.

That’s not just naive — it’s dangerous. (And yes, Taleb would absolutely tear this line of thinking apart.)

This is where techno-optimism turns into intellectual laziness. Lee tears into this mindset, and I couldn’t agree more.

Techno-optimists:

Assume every past innovation led to progress, so AI must do the same — which leads to the belief that things are generally better: better economy, better market conditions.
Equate narrow AI with earlier tools like the printing press or tractors (compare apples to oranges).
Ignore AI’s characteristics such as deployment speed, further evolution, global reach, and lack of physical limitations.
And worst of all, say things like: “Don’t worry, new jobs will appear, like always.”

Lee shows how flawed this thinking is. We need to understand that AI is another General Purpose Technology — revolutionary technologies that “disrupt and accelerate the normal course of economic development” — in the same league as:

The steam engine (1st Industrial Revolution)
Electricity (2nd)
Information and Communication Tech (3rd)

But here’s the kicker: while the first two reduced the need for expertise and increased the number of workers and most importantly increased EFFICIENCY, ICT increased the need for expertise while decreasing the number of workers. EFFICIENCY also grew, but wages didn’t, creating a growing divide.

Steam and electricity: positive disruption through expertise democratization

Steam and electricity devalued human expertise — but this had a massively positive social impact. Higher production meant more goods, lower prices, more people employed in simpler jobs. We shattered scarce expertise into more accessible roles and the system accelerated. This was democratization through decomposition — breaking down elite knowledge monopolies into distributed, learnable skills.

ICT: centralization instead of distribution

ICT boosted productivity, but didn’t raise wages in tandem — and here we need to emphasize the social aspect. Lucky bastards working in IT might think things improved, but wonder what they’d think if they were elsewhere. ICT favored centralization of competencies, rather than breaking them apart (opposite of previous top discoveries) — which adds contrast to AI’s potential impact.

And AI? It might just accelerate everything without creating enough meaningful new jobs.

🧠 This time, it’s not the blue-collar jobs.

Here’s the most brutal irony: The AI revolution is targeting knowledge workers first.

Historically, physical labor was the first to be automated. This time, it’s reversed. AI finds it easy to replicate:

pattern recognition,
data processing,
rule-based decisions.

It still struggles with:

physical dexterity,
navigating chaotic environments,
real social interaction.

So yes, your plumber is safer than your project manager.

White-collar workers didn’t realize that the path to building robots is much longer than simple algorithm implementation, which is already much better than knowledge workers at the most important tasks. Physical workers relying on dexterity and unstructured environments will be replaced in the more distant future — robots are further in the future. But they still have the advantage of local indivisibility and physical access — you can’t do plumbing from China, but you can already design documents remotely.

Lee drops a powerful comparison:

Just as tractors once caused a massive increase in productivity and reduction in the number of physical workers, AI will now be like a tractor for knowledge workers — reducing the need for their numbers. But unlike tractors, it’s invisible, cheap to replicate, and not limited by geography.

Even worse: if your “mental tractor” (AI) can do 80% of your work, your negotiation power crashes, even if you’re not fired. And if you are fired? You’re competing with even more people for fewer roles.

⚡ Turbocharged by software nature and Chinese potential

We need to mention the accelerators of change. AI = software is instant deployment — ‘click’ and it’s done. VCs are doing their work, and China with its over billion people, no longer being just copycats, are bringing massive intellectual potential and entrepreneurial drive — like adding new processing layers and new GPU/TPUs to a neural network, affecting the quality and speed of results. Plus their entrepreneurs with new implementations and new AI use cases.

In such a system, there won’t be time for adaptation that humanity previously had (e.g., 2 generations to adopt electricity). AI can devour entire industries in years, not the decades that previous revolutions required.

📉 The Macabre Matrix of Job Replacement

Lee introduces a brilliant 2x2 matrix:

For knowledge work:

X-axis: optimization ←→ creativity
Y-axis: asocial ←→ social

For physical work:

X-axis: structured/low dexterity ←→ unstructured/high dexterity
Y-axis: same

Jobs in the bottom-left quadrant (asocial + optimization-based) are at huge risk. Think government clerks — they essentially take data and make decisions based on it, so if you strip away all the chaotic wrapper, you can see that AI can trivially replace them, because it’s incredibly efficient and good at this, better than humans, without any bias.

Top-right (creative + social) is safer — psychiatrists, VPs, nurses.

The other zones? “Human veneer” and “slow creep” — the timeline may vary, but no one’s safe forever.

Human veneer — much depends on companies’ readiness for restructuring and other people’s readiness as clients for different types of interactions. A teacher who is “still” needed, but only because the company doesn’t want cultural changes.

Slow creep — here a lot depends on further progress in AI implementation. As AI capabilities grow, even a scientist isn’t safe, and a plumber can be replaced by a robot (though that’s the most distant perspective in these AI-related transformations).

📊 Studies agree: we’re underestimating the threat

Oxford (2013): 43% of jobs could be automated.
OECD: only 9% — but they looked at tasks not jobs.
PwC: 37%.
McKinsey: 45–50%.
Bain & Co (2018): Even accounting for new jobs, 25% may disappear.

And all this was before diffusion models, ChatGPT, and the recent AI leap.

Even worse, most studies assume 1:1 job replacement — not complete industry redesign. (See: loan agents replaced by algorithmic lending; journalism replaced by AI-driven curation.)

What does Lee say now? 2024 update

Panicked, I looked at what Lee says now, because 7 years have passed since he wrote the book! In 2024, Lee reaffirmed his 2017 prediction, saying his forecast that AI will displace 50% of jobs by 2027 is “uncannily accurate” and that the emergence of generative AI only strengthened his conviction about the correctness of this pace.

Lee emphasizes that AI will likely eliminate white-collar jobs faster than blue-collar jobs, describing it as a very significant problem that some countries have started realizing needs to be addressed.

🏆 AI takes everything: the monopoly of the future

AI is designed for monopoly. Those who are pushing forward now will rule and will have an insurmountable advantage. USA and China are crushing it, while the rest of the world falls behind. There’s no time to catch up — AI technology scales exponentially, and early access to the best talent, data, and capital will determine who wins. Plus, AI has network effects: the more data, the better the model, the more users, the more data — creating an unstoppable feedback loop.

This isn’t like previous tech revolutions where smaller countries could “catch up” over decades. AI rewards those who are first and punishes everyone else.

💣 Middle class? What middle class?

As AI widens productivity gaps, it creates two extremes:

Elite super-specialists in high-paying roles.
Low-wage, commoditized service workers.

Everyone else? Squeezed out.

No more middle management. No more “decent job with decent pay.”

If you’re not exceptional, you’re replaceable.

That’s not dystopia — that’s a forecast based on current reality.

🧩 Existential crisis incoming

Here’s what Lee gets absolutely right: This isn’t just an economic crisis. It’s a psychological and social one.

Studies show that depression triples among the unemployed.
Suicide attempts double among job seekers.
People will watch AI outperform them at tasks they spent years mastering.

And this isn’t just about losing income — it’s about losing identity.

🧾 Conclusion: This chapter isn’t science fiction — it’s a warning.

This isn’t about Terminators or rogue AGI. It’s about a far more mundane, but far more brutal reality:

AI will eat your job, not your soul — but it’ll take the soul with the job.

We’re looking at:

runaway inequality,
fragile economies,
broken identities.

All while techno-optimists keep saying: “It’ll be fine.”

But will it?

Let’s not be Luddites. But let’s also not be fools.

There’s no time to waste — we need to prepare for this and it’s not about simple binary answers. Those who want to survive should digest this topic, spend more time reflecting to find themselves in the future world. This might be the first moment when that clichéd interview question “where do you see yourself in 5 years?” makes REAL sense.

TL;DR

AGI? Not the real issue (yet).
Real crisis = job loss, inequality, social collapse.
AI will hit white-collar jobs first.
Techno-optimists are blindly extrapolating.
Don’t assume the “new jobs” will be there.
It’s not about fear — it’s about seeing clearly.

Your move.

Search

Reflections on Utopia, Dystopia, and the Real AI Crisis