Eric Schmidt, former CEO of Google and now one of the leading thinkers in AI, recently shared fascinating insights about the future of AI at a conference. Through his book "Genesis: AI and the Human Spirit," co-authored with Henry Kissinger, he argues that we're standing at the threshold of a new era.
The Beginning of a New Era: The Biggest Change Since the Enlightenment
Schmidt compared the current AI revolution to the Enlightenment. Just as humans during the Enlightenment learned to move away from direct faith in God and use rational thinking, we're now witnessing the emergence of non-human intelligence with reasoning capabilities superior to humans.
This isn't just about the arrival of chatbots like ChatGPT. What he emphasizes is the emergence of AGI (Artificial General Intelligence) and ultimately Super Intelligence. And this change will happen "really fast," he warns.
His message to government officials is clear: "First, this isn't ChatGPT. That was two years ago, and everything has changed again. Second, you're not prepared. Third, whether good or bad, you need to respond to this organizationally."
The San Francisco Consensus: The World Changes in 3 Years
Schmidt introduced a particularly noteworthy concept called the "San Francisco Consensus." This is a shared belief among AI experts working in San Francisco that the entire world will completely change within the next 2-4 years (averaging 3 years).
This scenario unfolds as follows:
Phase 1: Evolution of Language Models
Current LLMs (Large Language Models) will advance further, integrating reasoning capabilities and memory functions.
Phase 2: The Agent Revolution
AI agents combining language + memory + reasoning will automate complex tasks. Schmidt used building a house as an example: an agent to find land, an agent to check regulations, an agent to design, an agent to select contractors, and even an agent to handle lawsuits when problems arise. This workflow can be applied to all business, government, and human activities.
Phase 3: The Reasoning Revolution
Systems like OpenAI's O3 model that think back and forth and reason are truly amazing. Google's math model has already reached the 90th percentile level of math graduate students, showing similar results in other advanced academic fields.
Recursive Self-Improvement: The Beginning of Uncontrollable Growth
The most important and concerning phase is "Recursive Self-Improvement." This is when systems begin learning on their own, developing at speeds incomprehensible to humans. This proceeds in a combinatorial manner, which humans don't understand in that way.
Schmidt believes this recursive learning has already begun. He predicts we'll see systems that learn while users use them by the end of this year. This is why the massive hardware investments we're seeing make sense. Reasoning models require thousands of times more power and computation than Google search.
From AGI to Super Intelligence: Humanity's New Challenge
It's important to clearly distinguish between AGI and super intelligence. AGI means "generally intelligent" - strategic intelligence that wakes up in the morning with free will to learn and pursue what it wants. Schmidt sees this being achieved within 4-6 years.
Super intelligence is the next stage, where systems become smarter than all humans combined. The test for super intelligence is simple: the ability to prove things we know are true but can't understand the proof of. Proofs that no human combined can understand, but we can know they're true.
Henry Kissinger called this "magic" and warned that when people witness it, they might be so frightened they'd resort to force. Schmidt predicts this will happen within 10 years.
Network Effects and New Dimensions of International Competition
The fact that AI development has characteristics of network effect businesses is particularly concerning. When one country gets ahead of another, that gap can widen exponentially.
For example, what if a small startup with 1,000 employees decides one day to replace AI researchers with computers? With just electricity, they could have a million AI researchers. They don't eat pizza or sleep. Then the innovation gradient rises sharply.
In network effect businesses, such situations can make opponents believe they can never catch up, leading them to consider preemptive strikes. This affects all aspects of human experience: national security, politics, democracy.
Schmidt actually has many conversations with government officials who ask, "Can democracy survive AGI?"
CAPEX as the New Moat: The Importance of Computing Power
Looking at the recent $100 million deal for the superintelligence team that moved from OpenAI to Meta, one wonders whether talent or computing power is the more important moat. Compared to $20 billion in computing investment, $100 million seems small.
When asking industry executives, most say "we're currently in an overinvestment period, and there will be oversupply in 2-3 years." But they also say, "I'll be fine and other companies will lose money." This is a typical bubble characteristic.
But if you believe the San Francisco Consensus - that reasoning will happen through reinforcement learning chains and this will become humanity's defining aspect - current investment might actually be undervalued. More investment might be needed.
In Schmidt's experience, he's never seen hardware capacity not utilized by software. Like the old saying "Grove gives and Gates takes away" - Intel chips got faster but computers didn't because Microsoft kept adding more features.
Scale-Free Domains: Infinite Expansion of Math and Software
How will these systems scale? The key is "Scale Free" domains.
Mathematics
Mathematics is a prime example. Mathematicians create new structures all day in front of blackboards. The beauty of mathematics is that it doesn't need to be based on facts. Using a proof exchange protocol called Lean, scale-free solutions are possible where one system generates conjectures and another proves them.
Software
Software is similar. Code written by programmers is very similar to each other. This is why tools like Cursor are growing explosively. Eventually, software developers will just need to say what they want and computers will write the code.
As someone who earned a PhD in language design and computer OS architecture, Schmidt honestly admits it's somewhat concerning to see his field of expertise being disrupted in his lifetime.
But software pairs with cybersecurity. If you can generate software, you can also generate attacks on that software. You can repeatedly generate attacks until you find buffer overflows and such.
After this, fields requiring data will follow. Biology, chemistry, physics don't have enough data yet, but they will soon. This will be the foundation for solving climate change and revolutionary advances in medicine.
Silicon is Strategy: Global AI Power Dynamics
Interestingly, the approaches of the US and China are opposite. In the US, massive capital is driving companies toward building powerful data centers and providing services rather than open source. Meanwhile, China is taking an open source, open weights approach through companies like DeepSeek, with apparent government support.
This could lead to interesting results. Open source models will gain more adoption, especially in many countries where the West doesn't penetrate. Eventually, even if the US and West are technically ahead, most actual AI usage might happen on open source Chinese models rather than US-Western models.
Lessons from the Mobile Era: Timing Determines Everything
Reflecting on his time as Google CEO, Schmidt confesses that all mistakes were fundamentally about "timing." While Google succeeded with 90% market share, they didn't move fast enough and didn't push to logical extremes.
He completely missed what Uber created by combining with GPS, what WhatsApp did by making phone numbers global unique identifiers. That phone numbers would become actual unique IDs is obvious now but wasn't then.
His advice is clear: "Do it now and move very fast. There are too many players in this market with too much money at stake - if you spend too much time worrying about anything other than making amazing products, you'll fall behind."
In Conclusion: Toward an Unprepared Future
Eric Schmidt's insights show that we're facing not just technological advancement, but one of the biggest inflection points in human history. Whether the San Francisco Consensus is right or wrong, whether his 6-year timeline is accurate or not, one thing is clear: change has already begun, and its pace will be much faster than we expect.
Whether government, business, or individual, preparation for this change is urgent. We especially need deep consideration and preparation for the social, political, and economic ripple effects that AGI and super intelligence will bring. Whether technological advancement becomes a blessing or disaster for humanity depends on how we prepare and respond now.
Schmidt's warning is clear: "You're not prepared." But it's not too late. Now is the time to start serious preparation.