Humans have truly reached the doorstep of AGI!
A few weeks ago, GPT-6 Astra was unveiled, and OpenAI officially announced the realization of AGI.
Today, Google DeepMind also made a major move, rapidly establishing a new institution—
DeepMind Institute (DMI).
Led by DeepMind’s three leaders: Demis Hassabis, Shane Legg, and James Manyika.
Previously, DeepMind was holding back its biggest moves, aiming solely to create AGI. But now, with the establishment of DMI, the stakes have been raised even higher:
When AGI truly arrives, what will all of humanity have to respond with?
The DeepMind Institute is established, led by the three giants.
Google is getting serious about preparing for AGI.
After the establishment of the DeepMind Institute, it specifically researched the disruptive impacts of "superintelligence" on the economy and society.
In their view, "although current AI still makes mistakes, humans are drawing close to achieving AGI with full-brain cognitive abilities, and the gap will soon be closed."
The industry is stunned; no one expected the AGI tipping point to arrive so soon, in 2026.
DMI's goal is clear: to expand interdisciplinary research with a focus on examining the potential disruptive impacts of the AGI era on the economy, science, and society.
It will bring together researchers from fields such as AI, economics, philosophy, and policy to foster interdisciplinary dialogue.
Once AGI truly arrives, how will society adapt? What will happen to jobs? How will economic rules change? And how should a group of AI agents be governed?
Even the question of “what it means to be human” has been put on the table.
What’s truly chilling is the statement in DeepMind’s founding manifesto:
Humanity stands on the threshold of a profound transformation.
They believe that although today’s AI still falters on certain basic tasks and has not yet reached the standard of “full AGI” in terms of stability and creativity, the current pace of progress suggests that humanity is drawing closer to a system with the full cognitive capabilities of the human brain.
What about the remaining gaps?
DeepMind's assessment is very straightforward and will be quickly made up.
For 25 years, we pursued AGI—and now, this moment has finally arrived.
This is especially symbolic for Shane Legg.
AGI, which spanned almost his entire career.
More than twenty years ago, he was researching the formal definition of machine intelligence.
Later, he co-founded DeepMind with Hassabis and others, progressing from reinforcement learning and AlphaGo to today’s cutting-edge large models.
When announcing the establishment of DMI, Hassabis also specifically noted that he and Shane Legg have been discussing how AGI will impact the economy, science, and society for over 20 years.
Now, they have turned this into a formal institution.
And DMI opens with the first sentence: As we near AGI.
Note that it is "close to."
DeepMind has not announced that "AGI has been achieved."
According to their criteria, a true AGI should possess all the cognitive abilities of the human brain. Current models clearly still have limitations.
Sometimes even a very simple question can be answered incorrectly; stability still isn't sufficient when handling long-term, complex tasks; true top-tier human-like creativity has not yet been fully resolved.
DeepMind's current stance is also clear: these issues are more like "how much farther is left."
AGI has arrived; the first thing is not to celebrate
More interestingly, DeepMind did not only talk about the "golden age" this time.
According to Hassabis’s long-standing vision, AGI could become one of the most powerful scientific tools ever created by humanity.
Cure diseases, discover new energy sources, and drive economic growth by compressing scientific discoveries that once took decades into just a few years or less.
A new golden age of scientific discovery may indeed be opening.
But the founding declaration of DMI was immediately followed by a reality check—some risks are no longer in the future.
DeepMind explicitly stated that increasingly capable AI is beginning to impact cybersecurity and biological risks.
Take it one step further: if future AI systems emerge that can continuously self-improve, an even more challenging question will arise:
Will humans lose control?
So this time, the issues they listed grew larger and larger.
In the AGI era, what do humans truly value? Will it redefine what it means to be human?
How will we govern these "agent societies" once thousands, or even hundreds of millions, of agents emerge in the future?
Precisely for this reason, DeepMind stated for the first time: the future of AGI cannot be decided by technologists alone.
Philosophers, economists, humanities scholars, governments, and social scientists must all get involved.
On the first day of opening, start by exploring the "AI Brain".
DMI is not an empty nameplate.
On the first day, the blog released three articles covering AGI economic policy, reasoning transparency, the New Utopia principles, and cutting-edge frameworks.
This article on reasoning transparency focuses on AI's "thought processes."
Researchers believe that humans can still observe part of the model's reasoning process, and this window may become very important in the future.
If AI begins to deceive, hide its true objectives, or execute complex plans that humans did not request, we still need at least some way to understand what it is truly “thinking.”
In another article, researchers discussed eleven potential policy tools to address the economic impact of AGI.
The reason is also very practical.
If AI can someday perform a large portion of cognitive tasks that previously required human effort, productivity could surge dramatically—but income distribution, employment, and wealth structures would also change accordingly.
An economic system that appears stable today may not be transferable in its current form at that time.
The DMI homepage has now laid out all these issues, making it feel more like a preliminary manual for an AGI society.
The final moment arrives as the countdown timers of the giants align.
Looking back at the entire timeline, you'll see that all these events are connected.
OpenAI continues to bet on self-reflection in its reasoning pipeline, prompting various companies to explore autonomous code agents;
From Dario Amodei’s ten-thousand-word article warning of geopolitical and biological security risks amid the rapid advancement of AI, to now DeepMind’s entire leadership team fully engaging in building a comprehensive system to prepare for AGI.
Regarding AGI, no one truly knows all the answers.
What is becoming increasingly clear is that the time left for society to contemplate these issues, as left by cutting-edge labs, is being continuously compressed by successive rounds of model upgrades.
What happens to humanity if it actually does?
This article is from the WeChat public account "New Intelligence Yuan," authored by ASI Revelation.