Do you remember the historic match between AlphaGo and Lee Sedol in 2016? While many people simply saw it as "a computer beating a human at Go," Google's former CEO Eric Schmidt described that moment as "the moment the Earth quietly moved." Through his recent TED talk, let's explore the true meaning of the AI revolution we missed and what lies ahead.
The First Move in 2,500 Years of Go History
The truly pivotal moment in the AlphaGo match was during the second game. The AI made a completely novel move that no one had seen in 2,500 years of Go history. Technically speaking, AlphaGo was designed to always maintain a win rate above 50%, and in that calculation process, it discovered a move that humans had never even considered.
Why was this so shocking? Go is a game that billions of people have played. Yet a computer created a completely new idea that humans had never thought of. This wasn't just a victory of computational power—it was the first case of AI surpassing humans in the realm of creativity.
Schmidt says this moment was the real starting point of the AI revolution. While we didn't know exactly what would happen next, it was clear that these algorithms were new and powerful.
AI is Being Underestimated
Now that everyone is talking about AI, Schmidt argues that AI is actually being underestimated. Think about most people's reaction when they first experienced AI through ChatGPT: "Wow, it can write!" But that's not all there is to it.
There have been tremendous advances in reinforcement learning over the past two years. Looking at systems like OpenAI's o3 or DeepSeek R1, you can see them going forward and backward, making plans. This means we've moved beyond simple language generation to genuine thinking and planning capabilities.
Schmidt gives an example of buying a rocket company and using AI systems in fields where he's not an expert. These systems produce in-depth research reports in 15 minutes—and think about how much computational power 15 minutes of supercomputer processing represents.
We started with language, moved to sequence processing (the foundation of biology), and now we're at the stage of planning and strategy development. Eventually, computers will run all business processes, with individual agents collaborating by communicating with each other in English.
The Energy Crisis: We Need 90 Nuclear Power Plants
But this progress comes at an enormous cost: energy. Schmidt testified that the United States needs an additional 90 gigawatts of power—equivalent to 90 nuclear power plants. But this is practically impossible.
Think of each data center as requiring the power of an entire city. Arab countries are building 5-10 gigawatt data centers, and India is considering 10-gigawatt facilities.
While some argue that algorithmic improvements can reduce power consumption, there's an old law: "What Grove gives, Gates takes away." When hardware gets faster, software uses up all that performance. Planning capabilities, in particular, require 100 to 1,000 times more computation than traditional deep learning.
Data Depletion and the Limits of Knowledge
Beyond the power problem, there are two other major challenges. First is data depletion. Since we've already trained on all publicly available internet data, AI now needs to generate its own data.
Second is the limit of knowledge. How can AI that thinks based on existing knowledge invent something completely new? Geniuses like Einstein saw patterns in one field and applied them to completely different areas to make new discoveries. Current AI systems lack this ability.
To solve this problem, we need to address the technical challenge of "non-stationarity of objectives"—the ability to adapt when rules keep changing. If we solve this problem, we'll need more data centers, but we could create entirely new schools of scientific and intellectual thought.
The Autonomous AI Dilemma: When to Pull the Plug
AI research pioneer Yoshua Bengio has argued that we should stop developing autonomous agent AI systems. But this is precisely the next step that all AI labs are heading toward.
While Schmidt acknowledges Bengio's concerns as legitimate, he offers a different approach to solutions. What happens when agents start using their own computer language instead of human language? We won't be able to know what they're doing. That's when we need to shut down the system.
The key is observability—we need to be able to monitor AI systems. The industry is defining several warning signs:
But stopping development in a competitive global market isn't realistic. Instead, we need to find ways to establish guardrails.
The Dangers of US-China AI Competition
Schmidt says the AI competition between the US and China will define this field. The US has currently imposed 145% mutual tariffs, which significantly impacts supply chains. Our industry depends on Chinese packaging and components, and if China blocks access to these, it would be a major problem.
Conversely, the US is trying to block China's access to cutting-edge chips. Interestingly, China is responding with an open-source approach. In DeepSeek's case, they found more efficient algorithms to circumvent constraints. If China leads in open source, it will spread rapidly worldwide.
This leads to a truly dangerous scenario. Let's apply the concept of mutually assured destruction that Dr. Kissinger designed to AI. Say you're the good guy and I'm the bad guy. You're six months ahead of me, and we're both heading toward superintelligence.
Six months might seem manageable, but it's not. This is a network effects business, and the slope of improvement determines everything. OpenAI and Google are replacing 1,000 programmers with 1 million AI software programmers. The closer we get to superintelligence, the steeper the slope becomes.
If you get there first, I can never catch up. You'll have the tools to recreate the world, especially the ability to destroy me. So what would I do?
1. First, I'd try to steal your code
2. I'd try to infiltrate human spies
3. I'd try to corrupt your models
4. Finally, I'd bomb your data centers
It sounds crazy, but these conversations are actually happening between nuclear powers. Some people say preemptive strikes are the only solution.
The Open Source Dilemma
So how should we view open-source AI? Our entire industry and science are built on academic research and open source. Many of Google's technologies came from open source.
But what if truly dangerous open-source models fall into the hands of people like Osama bin Laden? The current industry consensus is that open-source models haven't yet reached the level of national or global risk. But we can see that pattern emerging.
This fight will happen between the US and China—only these two countries are "crazy" enough to invest tens of billions of dollars. Europe wants to but lacks the capital structure, as do India and Arab countries.
Dr. Kissinger said the possibility of war with China would start from an accidental incident, just like World War I began with a small event that escalated throughout that summer.
Surveillance State vs. Individual Freedom
There's a strange tension in the process of aligning AI systems at scale. Solutions to prevent 1984 often sound like 1984 itself. In trying to prevent dystopia, we might accidentally create the ultimate surveillance state.
Schmidt says he's deeply committed to individual freedom. It's easy for well-intentioned engineers to accidentally create freedom-restricting systems while building optimized ones. This isn't a technical problem but a business decision issue.
Creating a surveillance state is possible, but so is creating a system that gives freedom. The key is identity verification. But this verification doesn't need to include details. Using cryptographic techniques like zero-knowledge proofs, you can prove you're human without connecting any other information.
The Dream: Defeating Disease and Infinite Possibilities
There are also hopeful visions for the future. At Schmidt's age, friends start getting serious diseases. Why can't we cure all these diseases right now? Why can't we eradicate these diseases?
One nonprofit organization plans to identify all human drug targets within the next two years and make them available to scientists. Knowing drug targets allows the pharmaceutical industry to start developing treatments. Another company claims to have found a way to reduce Phase 3 clinical trial costs by an order of magnitude.
We also want to find dark energy and dark matter. There's tremendous physics hidden there, which would lead to a revolution in materials science.
Why doesn't every human on Earth have a personal tutor in their own language? A tutor that helps them learn new things starting from kindergarten. The technology works—we just haven't built it because there's no economic logic.
Most healthcare worldwide is either absent or provided by nurses or stressed village doctors. Why isn't there a doctor assistant system that helps them provide perfect healthcare in their own language?
What Will Humans Do?
When we reach a world of radical abundance and AI systems handle most economically productive work, what will humans do?
Schmidt's answer is clear: humans don't change. Do you think lawyers will disappear? No, they'll conduct more sophisticated lawsuits. Will politicians disappear? No, they'll mislead people on more platforms.
The key point is that our society doesn't have enough humans. Look at Asia's birth rates—essentially 1.0 for every two parents. That's not good.
The core problem during our lifetime is making people in their productive years more productive so they can support old people like me. These tools will dramatically increase that productivity.
One study suggests that under agent AI, discovery, and the scale I've described, we could achieve 30% annual productivity growth. When I talk to economists, they say they don't have models for that level of productivity growth. We've never seen anything like it.
Ride the Wave with Marathon Spirit
Finally, Schmidt's advice is that this is a marathon, not a sprint. From his experience in a 100-mile bike race, he learned about spin rate—getting up every day and keeping going.
From his Google experience, he knows that when you're growing at high speed, you forget how much you accomplished in a year. Humans can't comprehend that.
As exponential growth accelerates like this, we'll forget what was true 2-3 years ago. So here's the advice: ride the wave, but ride it every day. Don't view it episodically—understand and build.
Whether you're an artist, teacher, doctor, businessperson, or technologist, everyone has a reason to use this technology. If you don't use it, you'll become irrelevant compared to your peer group, competitors, and people who want to succeed.
Adopt quickly. Even Schmidt, with his enterprise software background, was shocked by the development speed of these systems. Now, through Anthropic's model protocol, you can connect models directly to databases without any connectors. Entire industries are disappearing. You just say what you want, and it gets built.
In Conclusion
We're witnessing the most important change in human society that happens perhaps once every 500, maybe 1,000 years. And it's happening in our lifetime. So let's not mess it up.
The journey that began with that one move by AlphaGo in 2016 is now leading to fundamental changes in human civilization. While there are countless challenges—energy crises, geopolitical tensions, ethical dilemmas—there are also infinite possibilities opening up: defeating diseases, revolutionizing education, and making scientific discoveries.
The important thing is not to fear this change but to actively embrace and utilize it. Let's try to understand and use this technology a little bit every day. Only then can we ride this massive wave instead of being swept away by it.
*Source: TED Talk - Eric Schmidt on AI*.