Written by: Techub News整理

Introduction

In the current wave of artificial intelligence sweeping the globe, every statement from Demis Hassabis, co-founder and CEO of Google DeepMind and one of the leading explorers in the field, has drawn attention. In June 2025, he participated in a deep discussion for WIRED's "The Big Interview" column. In this approximately 20-minute interview, Hassabis not only provided a clear probabilistic prediction regarding the arrival of Artificial General Intelligence (AGI), but also elaborated on core topics such as the definition of AGI, the limitations of current AI systems, geopolitical competition, AI safety, future work forms, and the "fundamental abundance" society that AGI might bring, showcasing both technical rationality and grand vision. As a scientist and entrepreneur who takes "solving intelligence, then using intelligence to solve all problems" as his mission, his views offer a key perspective for understanding the current coordinates and future trajectory of AI development.

Summary

• Predicted a 50% chance of AGI being achieved in the next 5-10 years, while emphasizing that current systems still exhibit significant flaws in reasoning, planning, memory, and true creativity.
• Pointed out two core risks faced by AI development: the misuse of general AI technology by malicious actors and the security threats posed by the technology itself as its capabilities expand.
• Believes AI will have an "empowerment" effect on jobs in the short term, enhancing individual productivity, but in the long term, it may reshape the employment structure, and jobs requiring human empathy, such as nursing, will be difficult to replace.
• Envisioned that if AGI develops smoothly, humanity will enter an era of "fundamental abundance" characterized by nearly unlimited clean energy and solutions to diseases and resource shortages, but this requires a shift in social mindset from zero-sum to non-zero-sum games.
• Called for international collaboration and flexible regulation, asserting that scientific methods and continuous investment are key to overcoming technological challenges in the face of unknown risks.

AGI: A Five-Year Pact and the Missing Puzzle of Abilities

The interview began with a discussion about the timeline for AGI. Demis Hassabis reiterated the vision held by DeepMind since its inception: to solve intelligence, then use that intelligence to solve everything else. Reflecting on this nearly 20-year journey, he stated that the company has been "essentially proceeding according to plan." When asked if this means AGI is just five years away, Hassabis provided a more precise probability estimate: "I think we have a 50% chance of having what we define as AGI in the next five to ten years."

This prediction seems relatively cautious compared to some more aggressive peers (who claim it will be achieved in two or three years). Hassabis explained that there is currently debate in the industry regarding the definition of AGI, leading to varying time predictions. He referenced an early definition of AGI by co-founder and chief scientist Shane Legg, emphasizing that its core lies in a system's ability to exhibit all cognitive abilities possessed by humans. The human mind is the reference point because it is the only known example of general intelligence.

So how far are we from this goal? Hassabis pointed out that current AI systems exhibit a clear "puzzle of missing abilities." Whether it be large language models or chatbots, there are still shortcomings in reasoning, planning, memory, and true inventive creativity. "I don't believe today's systems can truly invent, create, or propose new scientific hypotheses." He illustrated that while some systems (like their AlphaFold) can achieve gold medal standards in the International Mathematical Olympiad, the same systems can sometimes err in high school math or in counting the number of letters in a word. This extreme inconsistency across different areas precisely indicates that existing systems have not achieved full generalization capabilities.

Regarding the form of AGI's arrival, Hassabis believes it is more likely to be a gradual transition rather than a discrete leap function. Even if AGI is technically achieved, its comprehensive impact on society and the physical world (such as factories and robots) will still take longer to manifest, possibly taking more than ten years to fully emerge.

Geopolitical Competition, Safety Risks, and Regulatory Dilemmas

When the topic shifted to the intense global AI competition, the host referred to Eric Schmidt's "binary" rhetoric, stating that if a certain country (like China) were to gain AGI first, it would lead to an uncatchable permanent advantage. Hassabis responded that this is an unknown variable, representative of the so-called "hard takeoff" scenario: if AGI systems can rapidly self-improve and self-code, even a slight lead could swiftly transform into a chasm. However, he also pointed out other possibilities, such as self-improvement not being so rapid, in which case temporal proximity would not result in vast differences.

Nonetheless, he emphasized the geopolitical implications: "The systems being built will carry the imprint of their designers' values and the cultural norms in which they exist." Thus, who progresses first and where these projects are initiated concerns not only commercial and national competition, but also closely relates to security considerations. He candidly stated that the current field is in a "very intense period", with massive resources pouring in and significant pressure.

Hassabis revealed his top two concerns: the potential for malicious actors (whether individuals or rogue states) to use general AI technology for harmful purposes; and the risks inherent in AI technology itself—as its capabilities grow and its autonomy increases, ensuring that safeguards are sufficiently secure and not bypassable is critical. This led to a discussion on regulation. When asked if he still calls for regulation as some companies did years ago, Hassabis affirmed but emphasized the complexity and international nature of the matter. He believes that given the overlay of geopolitical effects and the complexities of today's world, "wise, flexible regulation" is essential, and such regulation needs to dynamically adjust as research understanding deepens.

He further indicated that the real challenge lies in international cooperation. "Restricting it in one region does not truly provide benefits to the overall safety these systems build for the world and society." Given that AI is a digital system and will affect everyone, fostering international collaboration and consensus (at least on some basic principles) is a larger challenge, but one that is necessary.

When asked if competition pressure might force them to forgo safety barriers, Hassabis indicated that leaders of major Western labs and top researchers frequently communicate. The issue is that there is currently a lack of clear definitions to establish when "that critical point" arrives. He believes that today's systems, while impressive, still have many flaws and do not pose any existential risk. The risks are still theoretical, but the problem lies in "too many unknowns"—we do not know the speed and extent of risk emergence. On this, he remains technically optimistic: "When there are so many unknowns, I optimistically believe we can overcome them." He believes that as long as there is sufficient time, care, and thoughtful consideration, using scientific methods, technological challenges can be solved, whereas geopolitical issues might be more intractable.

However, he is also acutely aware that: "We are running out of time." Therefore, DeepMind is investing more resources in safety and controllability research (such as mechanical explainability) and also believes there is a need for more social discussions to explore institutional building, governance models, and how to reach international agreements.

Empowerment, Transformation, and the Future of "Fundamental Abundance"

When discussing the impact of AI on work and markets, Hassabis cited economists' observations, asserting that fundamental changes have yet to be seen, and what we see more is supplementation and empowerment, such as how AlphaFold has greatly accelerated scientists' work. Looking forward to the next five to ten years, he predicts that jobs will undergo significant changes, but historical experience shows that new technologies usually spawn new, often better job opportunities, similar to the transformations brought by the internet and mobile technologies. He anticipates the coming years to be a "golden age", with powerful tools significantly enhancing individual productivity, even making people "somewhat extraordinary" in terms of creativity.

Addressing concerns that AGI might replace all human jobs, Hassabis believes that even if AGI possesses all human capabilities, there are still many things we wouldn't want machines to do. He cited doctors and nurses as an example: while diagnostic tasks may be aided or even replaced by AI, the human empathy and care required in nursing are unlikely to be replaced by robots. He advises current students and graduates to actively embrace these new systems, deeply learn STEM and programming to understand their principles, and master skills such as fine-tuning and system prompts in order to become "super producers" who can maximize the utilization of these tools.

Looking even further into the future (20-30 years from now), if all goes well, Hassabis paints an exciting picture: "We should be living in what I call a ‘fundamental abundance’ era." AGI will help solve some of the "root node" problems faced by society, such as curing diseases, significantly extending healthy lifespans, and discovering new energy sources (like optimizing batteries, room temperature superconductors, and nuclear fusion).

In response to the host's concerns about current social inequities and difficulties in collaboration, he provided a specific example: access to water resources. Desalination technology exists, but it is limited to wealthier countries due to its high energy consumption. If AGI promotes the realization of near-zero-cost clean energy (like nuclear fusion), then the issue of accessing water resources will be easily resolved, and conflicts arising from water resource control will significantly decrease. Similarly, nearly unlimited energy will also simplify the production of rocket fuels (by separating hydrogen and oxygen from seawater).

Hassabis believes that AGI will technically provide the capability to achieve "fundamental abundance," but he hopes it will further prompt a shift in societal mindset from zero-sum games to non-zero-sum games. He acknowledges that fairly distributing the results of "fundamental abundance" remains a challenge, requiring the involvement of philosophers and social scientists. He even predicts that in the era following AGI, economics itself, along with concepts of value and currency, may change, urging economists to delve deeper into related new economic theories.

Finally, in response to societal fears and resistance toward AI (similar to the backlash experienced during the Industrial Revolution), Hassabis expressed understanding and stated that this transformation is at least of the same scale as the Industrial Revolution, if not larger. However, he responds with his own passion: his core driving force in building AI is to advance scientific and medical progress, addressing significant societal challenges such as diseases, climate, and energy. He cited AlphaFold (a Nobel Prize-level breakthrough) and its potential applications in drug discovery as an example, emphasizing that if humanity has the capability but does not pursue these solutions, it would be unethical. In his view, facing many social challenges, without revolutionary technologies like AI coming to provide support, he would be more concerned about the future.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。