Chapter 5 — My Take on “The Four Waves of AI”
Kai-Fu Lee outlines four waves of AI development: internet AI, business AI, perception AI, and autonomous AI. While I appreciate this framework, I can’t help but feel that these are not neatly sequential stages. Rather, they seem like overlapping processes, all happening with different intensities and at different levels of maturity.
1. Internet AI — Basic and Often Annoying
Lee starts with what he calls the “internet wave” of AI. For me, this represents the most primitive forms of narrow AI — recommendation engines like those used by Amazon, Netflix, or YouTube. These systems never impressed me. When I didn’t actively interact with them, their predictions were poor, and their constant nudging felt invasive. This might be due to my European sensibility around privacy — even if I’m not obsessed with it, it shapes how I react to these systems.
Still, it’s important to clarify: narrow AI as a category fascinates me — especially large language models. It’s these simple recommender systems that underwhelm me.
And when Lee says the U.S. has a small lead here, I’m left wondering how solid those numbers are.
2. Business AI — Where “Real AI” Emerges
This second wave excites me the most. I see it as the birthplace of what I call “real AI.” Here, algorithms extract patterns from data that humans can’t detect — outperforming human intelligence in specific domains. Lee’s example of a Chinese lending app that calculates risk using unconventional parameters like smartphone battery level was eye-opening.
While he called that data point absurd, I see a logic: people who panic about keeping their phone charged may also be more conscientious debt repayers.
He nailed it when he explained why AI outperforms humans here: we focus on strong, obvious features — AI thrives on identifying weak signals that humans ignore.
Lee argues the U.S. is far ahead in this wave because of its historical obsession with tracking and labeling everything in finance and healthcare. That makes sense. And honestly? This is the wave that amazes and terrifies me at once. I see how many jobs it could replace — without offering much in return.
3. Perception AI — Gateway to the Physical World
Lee calls the third wave “perception AI,” mostly about image and audio recognition. On the surface, nothing here shocks me. But he shifts focus to how AI blends the online and offline worlds — for instance, smart stores where sensors and perception AI know your habits, where items are, and what your fridge needs.
What strikes me is that this wave allows AI to escape the screen — and that could lead to a massive offline revolution.
It also reveals a geographic imbalance: the U.S. dominates data-heavy domains like finance and medicine, but China has more physical-world data — an advantage in this wave.
4. Autonomous AI — Futuristic, but Not Distant
The final wave is autonomy, mostly applied to vehicles. Lee contrasts Google’s perfectionist approach (develop the full stack before release) with Tesla’s iterative method (deploy pieces progressively). Surprisingly, Tesla caught up — and that made me think of how Chinese AI operates more like Tesla.
But Lee adds something wild: in China, the government is willing to adapt the infrastructure to match autonomous needs. Roads might be redesigned for self-driving cars. That’s futuristic — but no longer science fiction.
It reminds me of 2021–2022, when I first heard about Volvo’s autonomous truck tests. They worked shockingly well. I remember feeling sympathy for truck drivers — but also safety about my own IT career. A few years later, I realized: I, too, am replaceable. That’s a big reason I’m reading AI Superpowers and Melanie Mitchell’s AI: A Guide for Thinking Humans.
Final Thoughts — I’m in This Era, Too
Reading this chapter, I felt something deeper: I’m not just observing — I’m inside this shift. I’m one of those future AI engineers, the kind who’s lived through the Linux server days, then virtualization, then containerization… and now? Possibly the most transformative shift of all.
Lee’s wave metaphor still leaves me slightly unconvinced. Yes, each wave builds on data from the previous, but in practice they feel like concurrent movements. Still, I appreciate his clarity — especially where it lets me reflect on my own place in this evolution.
This chapter made something clear: AI is already outperforming us in many domains. And for those of us in tech, we’re not immune — we’re the next in line. Better understand it than ignore it.