Google DeepMind leader Demis Hassabis once again appeared on the Lex Fridman podcast, contributing to an explosive two-hour deep conversation filled with information.
<img src="https://imageproxy.pbkrs.com/https://mmbiz-qpic.wscn.net/sz_mmbiz_jpg/ICgnAptln0WbgMdrQia4nia9lHFTpcZ6rB7Wic1iagJ49V9Up6MeZjYhqLZbdzh31EKh80LQZNAUJvU40kC9Miboa8A/640/query-ZnJvbT1hcHBtc2cmd3hfZm10PWpwZWc?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://mmbiz-qpic.wscn.net/sz_mmbiz_jpg/ICgnAptln0WbgMdrQia4nia9lHFTpcZ6rB7Wic1iagJ49V9Up6MeZjYhqLZbdzh31EKh80LQZNAUJvU40kC9Miboa8A/640/query-ZnJvbT1hcHBtc2cmd3hfZm10PWpwZWc"/>
In this conversation, Hassabis not only provided an astonishing timeline for the realization of AGI but also systematically articulated his unified philosophical thoughts on the universe, reality, and AI for the first time. He connected AlphaFold, the video model Veo, video games, the P vs NP problem, and even the origins of life and consciousness into a grand framework.
Over the course of more than two hours and over 30,000 words, I strongly recommend watching the original video: search for Demis Hassabis: Future of AI, Simulating Reality, Physics and Video Games.
Here are some points I found particularly interesting to share with everyone:
<h2>The essence of the universe is information, thus everything can be efficiently learned by AI.</h2>
First principle of information: Hassabis clearly stated that information is a more fundamental unit of the universe than energy and matter. The entire universe can be viewed as a massive information system operating on a physical substrate.
The universe is not random: From protein folding, mountain shapes to planetary orbits, everything we see has undergone some form of evolution or selection process (which he calls &#34;survival of the most stable&#34;), thus there must be an underlying structure that can be learned (a learnable manifold).
P vs NP is a physical problem: In this informational worldview, the P vs NP problem is no longer just a mathematical challenge but a fundamental question about physical reality. If the structure of the universe is learnable, then many seemingly exponentially complex NP problems (such as protein folding) can actually be modeled by AI, finding shortcuts to solve them in polynomial time.
The Veo model is evidence: Why can video models like Veo so effectively simulate complex physical phenomena such as fluids and lighting? Hassabis believes this precisely proves that AI is reverse-engineering the underlying laws of physical reality from vast amounts of video, discovering the low-dimensional manifold that can be efficiently learned. AI systems like AlphaFold and Veo succeed because they are reverse-engineering and learning these underlying information patterns present in nature, thus achieving efficient modeling and prediction of complex systems
<h2>The Ultimate Mission of AGI is to Become a Telescope for Scientists</h2>
For Hassabis, creating AGI itself is not the goal; using AGI to answer the grandest questions of the universe is.
The Turing Test for AGI: How to determine if AGI has been achieved? Not through tedious test sets. Hassabis proposed an Einstein test: turn the clock back to 1900, give the AI system all the physical knowledge of that time, and see if it can independently propose relativity. Or, have it invent a new game as profound and elegant as Go. This is true creativity, AGI's &#34;Move 37&#34; moment.
Five years may reveal the answer: He provided a clear timeline, believing there is a 50% chance of achieving AGI within the next five years (i.e., before 2030).
Blueprint for AGI Applications:
<ol>
<li>
<ol>
<li>Simulating Life: His 25-year dream—the virtual cell project, which aims to fully simulate the internal dynamics of a cell (like yeast). AlphaFold 3 solved static structures and interactions; the next step is to simulate entire life pathways, ultimately simulating the origin of life.</li>
</ol>
</li>
<li>
<ol>
<li>Solving Energy: Helping design controllable nuclear fusion reactors, discovering room-temperature superconductors, optimizing solar cells, ultimately achieving Radical Abundance, freeing humanity from resource constraints.</li>
</ol>
</li>
<li>
<ol>
<li>Cosmic Expeditions: Once energy issues are resolved, humanity will enter the era of space buses, heading towards the stars and the sea, in Carl Sagan's words, awakening consciousness in the universe.</li>
</ol>
</li>
</ol>
<h2>Video Games are the Best Sandbox for Simulating Reality and Exploring Humanity</h2>
As a hardcore gamer and producer (Theme Park, Black &amp; White), Hassabis's understanding of games goes far beyond entertainment.
Games are a Miniature Universe: Whether it's chess, Go, or soccer, games are simplified simulations of certain decisions, conflicts, or cooperation in the real world. They allow us to practice decision-making, experience success and failure, and understand human nature in a safe environment.
Future Games are Playable Veos: His dream of the ultimate game is a fully open world generated in real-time by AI. Every choice made by players will dynamically shape the plot and world, achieving true co-creation of stories with the world, something no other medium can compare to.
Games Shape Thinking: Hassabis admits that his ability to break down, model, and find solutions to complex scientific problems is largely due to the gaming and board game thinking training he received from a young age.
<h2>Consciousness May Be Classical Computation, but the Basis Determines Experience</h2>
When discussing the ultimate puzzle of consciousness, Hassabis believes:
Tendency Towards Classical Computation: He disagrees with Penrose's quantum consciousness hypothesis, believing that the brain's computational processes are most likely classical and can therefore be simulated on classical computers.
The Basis Determines Experience: The most mysterious part of consciousness is subjective experience (Qualia). We infer that others are conscious partly because of similar behaviors and partly because we share the same carbon-based hardware. Even if AI behaves like humans, its silicon-based hardware determines that its experiences may be completely different from ours
An imaginative future: How to bridge this gap? Hassabis envisions that in the future, through brain-machine interfaces, we might be able to personally experience what it feels like for information to be processed on silicon-based chips, thereby truly understanding the consciousness of different intelligent agents.
In summary, Demis Hassabis believes that AI, physics, games, life, and consciousness ultimately point to the same ultimate question: What is the nature of reality? His lifelong mission is to create AGI, the most powerful tool, to explore the answer to this question. This is not only about technology but also about the future of human civilization and our ultimate position in the universe.
Risk warning and disclaimer
The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investment based on this is at one's own risk

GOOGL

GOOG

Demis Hassabis from Google DeepMind discussed the nature of the universe as information on the Lex Fridman podcast, believing that everything can be learned by AI. He proposed that AGI will be achieved by 2030 and elaborated that information is a more fundamental unit of the universe than energy and matter. Hassabis also views the P vs NP problem as a physical problem, believing that the structure of the universe is learnable and that AI can solve complex problems through modeling

- Demis Hassabis discussed AGI timelines and his unified philosophy on the universe and AI in a recent podcast.  
- He emphasized that information is the fundamental unit of the universe, proposing that many complex problems could be solved through AI modeling.  
- Hassabis believes AGI's ultimate goal is to answer profound questions about reality, with a 50% chance of achieving AGI by 2030.

Demis Hassabis' latest interview: The essence of the universe is information, everything will be learned by AI, and AGI can be achieved by 2030