Hello! I'm your friend, a financial journalist and economist.
This article from Havenlon Labs is an incredibly in-depth and well-written breakdown of the underlying logic of AI. It doesn't hype up the latest model parameters or boast about the market value of any particular company; instead, it takes a 80-year perspective to reveal an counterintuitive truth: The AI that seems to have “exploded” suddenly today is actually the result of countless small breakthroughs over the past few decades.
To help you understand this easily, I've summarized the key points and then broken down the economic and technological logic behind it from five different dimensions.
---
📝 Summary of Key Points: AI Is Not a “Moment of Genius,” but the Result of a Century of Progress
The main argument of this article is clear: The history of artificial intelligence is not a history of “heroes” but a history of the evolution of infrastructure.
Many people think that the emergence of ChatGPT in 2022 or the Transformer in 2017 marked the beginning of AI. However, if you only look at the past decade or so, it might seem like AI suddenly became much smarter. But if you extend the timeline, you'll see that:
1. **There were no “inventors”: Alan Turing never trained neural networks, NVIDIA didn’t intend to use GPUs for AI when they were developed, and the ImageNet dataset wasn’t created with the intention of training large language models.
2. The Transition from Barriers to Advancement: The evolution of AI isn’t a linear increase in intelligence; rather, it’s about humans gradually handing over cognitive processes that used to require human effort—writing rules, defining features, making decisions—to machines to learn and execute.
3. The Moment of Convergence: The current explosion in AI is due to the simultaneous maturity of three independent developments: algorithms (like Transformer), data (on the scale of the internet), and computing power (GPUs and clusters) between 2012 and 2022.
In one sentence: The “revolution” we’re witnessing is the result of conditions that had been developing independently, finally coming together at a certain point.
---
🔍 In-Depth Analysis: Understanding the Evolution of AI from Five Dimensions
1. From “Writing Rules” to “Learning Patterns”: A Fundamental Shift in the Source of Knowledge
[Easy-to-Understand Explanation]
Early AI (1950s–1980s) was like a “supercomputer.” If you wanted it to play chess, you had to write the rules (“If the opponent moves a horse, I move a car”) into the code. This was symbol-based AI and expert systems.
- Challenge: The real world is too complex for humans to write all the rules. For example, how do you define what is dangerous?
- Turning Point: The Perceptron in 1957 and the Backpropagation algorithm in 1986 brought about a paradigm shift. Machines no longer needed explicit instructions; instead, they were given a set of data (e.g., pictures of cats) and learned what a cat was through trial and error.
- Economic Perspective: This reduced the cost of acquiring knowledge. Previously, you needed to hire experts for expensive and time-consuming tasks; now, you just need to collect data, which is relatively cheap and fast. This was the first key barrier to the commercialization of AI.
2. The “Dual Drive” of Computing Power and Data: Why the 2012 Explosion?
[Easy-to-Understand Explanation]
Why was deep learning proposed in the 1980s but didn’t really take off until 2012? The “idea” was there, but the tools weren’t ready.
- Computing Power Barrier: Training complex neural networks used to take months or years; in 2006–2007, NVIDIA introduced CUDA, making GPUs capable of general computing, which is ideal for neural networks.
- Data Barrier: Before 2009, there were only a few thousand images; with ImageNet, millions of labeled images became available.
- Critical Point: In 2012, AlexNet emerged, combining better algorithms, more data, and stronger computing power. This led to a dramatic decrease in error rates.
- Economic Perspective: This is a typical “S-curve” of technology. Before the turning point, the cost-benefit ratio was low; once all three factors crossed the threshold, capabilities could be scaled linearly or even exponentially with additional investment in GPUs and data. This explains why NVIDIA’s stock price soared, as computing power became the “fuel” for AI.
3. From “Understanding” to “Acting”: Reinforcement Learning Completes the Puzzle
**[Easy-to-Understand Explanation】
Early neural networks focused on perception (“Is this a cat?”). True intelligence involves more than just understanding; it also involves action.
- Reinforcement Learning (RL): This approach, often overlooked, relies on rewards. It’s like training a dog: you give it food for correct actions and nothing for wrong ones. Machines learn by adjusting their strategies based on feedback.
- The Significance of AlphaGo: In 1997, Deep Blue defeated Garry Kasparov using brute-force search and human-written rules; in 2016, AlphaGo used deep learning to find its own strategies.
- Key Leap: AI evolved from “What is this?” to “What should I do next?”
- Economic Perspective: This laid the foundation for intelligent agents. If AI can only chat, it’s a tool; if it can make decisions based on feedback, it can replace human labor and move from information processing to task execution.
4. Scaling Laws: Turning “Luck” into “Engineering”
[Easy-to-Understand Explanation]
Before 2020, people thought AI’s intelligence came from smarter algorithms. Research in 2020 (such as GPT-3 and Scaling Laws papers) revealed a surprising pattern: with the same architecture, larger models, more data, and stronger computing power led to better results in a predictable manner.
- The Impact of GPT-3: With 175 billion parameters, it could learn new tasks without specific training. This was a breakthrough.
- Corrections: It turns out that more isn’t always better; the parameters, data, and computing power need to be balanced.
- Economic Perspective: This marked the industrialization of AI. Previously, AI development was like crafting handmade items; now, it’s like building factories. With enough capital (chips, data, data centers), you can create more powerful models. This explains why AI giants spend billions on data centers—scale becomes a competitive advantage.
5. From “Generating Information” to “Changing Reality”: The Era of Agents
[Easy-to-Understand Explanation]
This is the future direction the article points to.
- Past AI: Input → Model → Output (text/image). It only produced information.
- Current Agents: Observation → Reasoning/Planning → Using tools (search, code, browsers) → Action → New state.
- Essential Difference: AI’s output is no longer just answers; it can perform actions (book flights, write code, operate software).
- Alignment Issue: Strong models can be problematic if they don’t understand human intentions or cause irreversible consequences. We need feedback to teach them to act appropriately.
- Economic Perspective: This represents the ultimate value of AI. When AI can execute tasks directly, it’s no longer just an assistant but a “digital employee” that can reshape the labor market.
---
💡 Lessons for Everyone
1. **Don’t Mythologize “Geniuses”: AI’s progress is the result of the work of many people (mathematicians, engineers, chip designers, data annotators). When discussing AI, don’t focus only on figures like Elon Musk; also consider the infrastructure (electricity, chips, data).
2. Understand the Power of Scale: In the AI era, “bigger” means “stronger.” This applies to both technology and business. Those with access to massive data and computing power have a significant advantage.
3. Be Alert to the Misalignment between Ability and Intent: AI is intelligent, but it may not understand your intentions. Clear instructions and continuous feedback are crucial.
4. Focus on Action: In the coming years, the most important thing will be who can create AI that can reliably execute tasks. Whoever can move AI out of chatbots and into practical applications will seize the next opportunity.
To conclude, let me use the most inspiring sentence from the article:
> “What we call ‘artificial intelligence’ today was never the future created by a single genius. Instead, it’s the result of generations pushing forward the small pieces of the unknown, eventually forming an entire era that we can’t see.”
This is the truth about AI: It’s not magic; it’s engineering, accumulation, and time.