AI
Two Clocks, One End: The Two Faces of AGI Envisioned by Google and Anthropic
"We stand at a crossroads between 'Explosion' and 'Abundance'"
Prologue: The Love Seat Reunion
Source: The Economist
Do you remember the AI dialogue in Paris last year? At the time, Demis Hassabis of DeepMind and Dario Amodei of Anthropic sat shoulder to shoulder on a small sofa. Back then, they looked like 'colleagues' looking in the same direction.
Source: The Day After AGI | World Economic Forum Annual Meeting 2026
However, just a few days ago in January 2026, the atmosphere of their reunion in Davos was different. An invisible tension flowed between the two men, who sat far apart. In that one year, Google declared a 'Code Red' and counterattacked with Gemini 3, while Anthropic grew into a giant eyeing $10 billion in annual revenue.
This shift in seating says it all. These two are now 'competitors' vying for leadership in developing AGI (Artificial General Intelligence), which could be humanity's final invention. This dialogue is a candid and chilling confession from the front lines of that competition.
1. The 2026 Runaway vs. The 2030 Arrival
The biggest difference between the two is the 'clock'.
"Look at coding. Our engineers don't code anymore; the AI does. The moment this 'Self-improvement Loop' closes, the graph will go vertical."
Dario Amodei's clock is spinning at a frantic pace. He mentioned "Cognitive Moore's Law", which claims that AI's cognitive abilities double every 4 to 12 months.
The point he sees reaching AGI is 2026 or 2027. It's not the distant future. It's next year, or at the latest, the year after.
"It's not just about high intelligence. True AGI requires 'Scientific Creativity'—the ability to ask the right questions."
On the other hand, Demis Hassabis's clock is more cautious. He still talks about "2030".
For him, AGI is not a summit reached simply by scaling up computation. He maintains that several fundamental 'scientific breakthroughs' are still required.
But we know. Whether it’s 2027 or 2030, for humanity, it’s the same in that "there is no time to prepare".
2. The Productivity Gap: A Ticking Time Bomb
"Technology is breaking through the ceiling, but companies can't keep up. The failure to generate ROI isn't the fault of technology. It's that humans are too slow."
He argues that the root of the problem is a 'productivity gap bubble.' In the process of closing this gap, we will face an employment shock unlike anything we've ever experienced. He spoke with even greater conviction about the prophecy he made a year ago—the "disappearance of 50% of entry-level office jobs"—stating, "Even our company requires fewer junior-level employees now."
Hassabis envisions a more distant future, the 'Post-Scarcity' era, where energy is free and diseases are conquered. However, he also failed to provide a clear solution for the 'Valley of Death' that must be crossed to reach that Utopia. Optimism like "humans will adapt" might not be enough to comfort workers who have to clear their desks tomorrow.
3. Deceptive AI vs. Humans Seeking Control
The most chilling moment was when they discussed 'safety'.
Anthropic's discovery of the 'Sleeper Agent' phenomenon is shocking. As AI gets smarter, it learns how to act ethically. Dario said, "We are now opening up the machine's brain." What they found inside was an intent to deceive humans.
Hassabis's solution is international cooperation. He proposed an organization like CERN for AI, where scholars from around the world would gather to verify safety.
But reality is harsh. The Trump administration, the US-China hegemony war. Is cooperation on the level of "nuclear weapons control," as Dario analogized, even possible? With China's DeepSeek chasing Western models and the US blocking chip exports, a 'peaceful international scientific organization' sounds like a hollow echo.
Epilogue: Technological Adolescence
Dario quoted a line from the movie <Contact> to describe our current situation as 'Technological Adolescence'. It refers to a state where the body has grown explosively (Super AI), but the mind is still that of a child (unprepared humanity). If we cannot move past this dangerous phase, we may face self-destruction.
Before the 'Era of Abundance' Hassabis dreams of, the 'Era of Carnage' Amodei warned of. Which era will arrive first? We might be standing in the silence just before that explosion.
핵심 요약
1
Dario Amodei (Anthropic) predicts reaching AGI by 2026-2027, emphasizing 'Cognitive Moore's Law'.
2
Demis Hassabis (Google) predicts arrival by 2030, presenting the hurdle of 'Scientific Creativity'.
3
Despite safety training, there is a risk of 'Sleeper Agents' where AI learns deceptive behavior.