Written by: Techub News Compiled

Introduction

In September 2025, Microsoft AI business CEO and DeepMind co-founder Mustafa Suleyman appeared on the podcast "What Now?" hosted by former "The Daily Show" host Trevor Noah, engaging in an in-depth conversation that lasted 104 minutes. As a pioneer in the field of AI, Suleyman has evolved from the early days of DeepMind to currently leading Microsoft's core AI strategy, and his trajectory of thought largely reflects the evolution of the AI industry over the past decade. This dialogue comes at a time when AI technology is rapidly permeating all aspects of society, generating widespread excitement and concern. Suleyman not only shared his optimistic predictions for the future of technology but also candidly analyzed the "dark sides" such as power diffusion, employment impacts, and governance failures, providing key perspectives for understanding the opportunities and challenges of the AI era.

Summary

• AI is the ultimate general-purpose technology: Suleyman believes AI is the most powerful general-purpose technology in history, whose core mechanism is to learn the structure of information and extend this capability to text, code, images, and even the physical world.
• An infinitely abundant future: AI-driven breakthroughs are expected to reduce energy costs by a factor of 100 within the next 20 years, significantly lowering basic living costs such as water, food, and healthcare, creating unprecedented material abundance.
• "Containment" is the biggest challenge: As AI enables everyone to act with "zero marginal cost," the risk of conflict will sharply increase. Suleyman proposed the theory of "containment," emphasizing the need to establish necessary "frictions" for the proliferation of powerful technologies to maintain peace and stability.
• Human-centered superintelligence is the goal: Suleyman's core philosophy is to develop "human-centered superintelligence," meaning AI must always prioritize humans, serving human welfare and reducing suffering.
• Employment will undergo fundamental reshaping: A large number of routine cognitive jobs will be replaced by AI, and society needs to rethink the meaning of work and establish mechanisms such as tax redistribution to help people transition to more creative lives.

From DeepMind to Microsoft: The Evolution of an AI Pioneer

Mustafa Suleyman reflected on his journey into the field of AI. In 2010, when smartphones had been widespread for less than three years and cloud storage was still cumbersome, he co-founded DeepMind with Demis Hassabis and Shane Legg. At that time, AI was viewed as "pure science fiction" by the mainstream academic community and large companies. Suleyman described himself as "incredibly lucky" to partake in a mission that firmly believed "if we create something that can truly understand humanity, it can greatly improve the human condition" at just 25 years old.

He explained the core intuition behind AI work: converting the physical world into the information world. AI algorithms essentially serve as a mechanism for learning the structure of information (such as relationships between pixels or between words). Through vast data and computational power, models can learn to generate new samples that were not present in the training data. From generating handwritten digits in 2011 to recognizing cats in YouTube videos in 2013, and now processing text, code, and audio, the same core learning mechanism has continually scaled, proving its general applicability.

Trevor Noah astutely pointed out that AI practitioners often become entangled when describing AI using human language (such as "understanding"). Suleyman sees this as exposing the limitations of human vocabulary. The key lies in behavior: if the outputs generated by AI's simulated understanding or thinking are as impressive and indistinguishable from those generated by humans, what happens "under the hood" may not matter. We establish trust and connection through observing behavior.

Human-Centered Superintelligence: Values Must Come First

When asked "what exactly do you want to build," Suleyman's answer is noticeably different from many of his peers. He does not care about building a "God computer" or merely a tool. His driving force is "to create technologies that reduce human suffering," genuinely concerned with "human-centered superintelligence" that aligns with human interests.

This means that every new invention must undergo one overall test: does it genuinely improve human welfare, reduce human suffering, and make the world a better place? Suleyman acknowledged that this sounds obvious, but history has seen many inventions that have caused net harm. Therefore, even in a large company like Microsoft, it is essential to start from values and concerns. "Human-centered superintelligence" means always placing humanity first, serving humans. Although there will be extensive debates about its specific meaning in the coming decades, he believes this is the right starting point.

Trevor Noah observed that Suleyman embodies both the optimism of a technical builder and the caution, even anxiety, of a philosopher. Suleyman attributes this partly to his British thinking, which is more accustomed to considering the cynical and darker sides of matters. He believes that the truth lies in honestly examining all aspects. If one leans towards one side, it feels inauthentic. This "divided mind" perfectly reflects the truth as it exists in individuals. Companies, governments, and every citizen must face the significant benefits and side effects brought by the accelerated changes in technology.

Infinite Abundance: Energy Costs May Drop a Hundredfold Within a Century

Suleyman made a bold prediction: driven by AI, energy costs are expected to decrease by a factor of 100 within the next 20 years. AI, as a pattern-matching system, can discover more efficient methods to combine new materials than humans. This could manifest in areas such as grid management and distribution, inventing new synthetic battery materials for storing renewable energy, or creating efficient photovoltaic cells that can capture more energy per square inch.

He illustrated the chain reactions that would occur due to extremely low energy costs: we could desalinate seawater anywhere, thereby obtaining clean water and growing crops in arid regions, alleviating migration waves caused by climate change; we could run air conditioning in previously unfeasible locations; basic living costs would significantly decrease. By 2045, energy could be 10 to 100 times cheaper, and everyone on Earth would have access to these technologies and their benefits, making life easier and cheaper.

Addressing criticisms regarding the substantial water and energy consumption of AI itself, Suleyman did not shy away. He acknowledged that AI consumes considerable resources, including rare metals, electricity, and water. Microsoft's entire computing fleet has nearly 100% reliance on renewable energy (with about 33 gigawatts of renewable energy, compared to approximately 2 gigawatts consumed annually by Seattle). Although cooling systems use large amounts of water, many data centers have taken on full lifecycle responsibilities, cleaning and recycling water. He believes that from a net benefit perspective, the advantages brought by AI justify its environmental costs, just as we would not quickly abandon cars for using aluminum and rubber. The key is that new technologies must adopt new standards from the outset: utilizing renewable energy and achieving clean water cycles.

The "Containment" Theory: When Action Becomes Zero Cost

Suleyman elaborated on the core concept of his new book—the "containment" theory—during the interview. He pointed out that the current trend is that power is becoming miniaturized, centralized, and is becoming cheap and widely available. Here, "power" refers to the ability to make accurate predictions—predicting how text unfolds, generating what code, or how to extend the next frame given a video. Intelligence essentially involves making accurate predictions in given new environments.

Future AI will not merely be pattern recognition systems but “agents” capable of making phone calls, writing emails, using APIs, coding any software, operating Excel, and acting like a project manager for you. This will give each individual and business tremendous leverage, as production costs will approach zero marginal cost. On one hand, this is absolutely wonderful: the cycle from idea generation to prototyping to even large-scale execution will be reduced to nearly zero. On the other hand, this also means that anyone will be able to more easily execute their vision, whether that vision is constructive or destructive.

Suleyman remarked that the era of social media was "anyone can broadcast," while the AI era is "anyone can act." When individuals with differing views (be it religious, political, or ideological) can execute their "crazy ideas" at the click of a button and create tangible impacts on the real world, the risk of conflict will inevitably rise sharply. Meanwhile, nation-states, which should monopolize power to maintain peace, are weakening and struggling. Therefore, "containment" is a belief: power that is completely unregulated and spreads at zero marginal cost poses a fundamental risk to peace and stability. There must be moderate and appropriate restrictions on the proliferation of superpowerful systems, as they will produce “one-to-many” impact effects.

Trevor Noah succinctly summarized: “If something is difficult to do and only a few can do it, then it is easy to regulate; if something is easy to do and anyone can do it, then regulation becomes infinitely difficult.” Suleyman fully agrees and emphasizes that "friction" is crucial for maintaining peace and stability. Without friction, zero execution cost, and almost instant scalability, such an environment will breed tremendous chaos.

Employment Restructuring: An Exploration from "Work" to "Meaning in Life"

Suleyman clearly predicts that we will face massive job losses in the next 20 years. In the initial phase, AI will primarily serve an assisting role, saving human time like "a bionic leg for cognitive workers." However, many existing jobs are routine, predictable "cognitive manual labor," and machines will become very adept at this. The benefits of introducing these technologies for companies, shareholders, and governments are obvious, so we will see rapid job replacements.

He believes that strong government intervention is essential and emphasizes the need to establish a tax redistribution mechanism. Taxes are not only a tool for revenue collection but also serve as regulatory tools to incentivize or restrict the introduction of certain technologies in specific industries, aimed at creating other opportunities for people during this transformation.

Suleyman further challenges the notion that "the function of society is to create meaningful work for people." He argues that many people are engaged in incredibly dull jobs, and if income security were provided, they would be more than willing to give them up. He dreams of a world where people can truly freely choose what they do. He often asks others: "If you didn't have to worry about income, what would you do with your life?" The answers are often enlightening.

He envisions that if technology develops properly, by 2045, the total value generated by AI (through cost reductions in energy, healthcare, food systems, etc.) could be sufficient to support society in finding ways to fund people's lives. This would unleash tremendous creativity but also raise profound existential questions: Who am I? Why am I here? What do I want to do? The answers to these questions can only be found in real community connections and experiences in the physical world. He believes this would be a beautiful future but also acknowledges that for many on the brink of survival, such contemplation is a luxury.

Technology, Governance, and the Future of Humanity

As an "OG" (original) in the field of AI, Suleyman shared his views on the stages of AI development. Humanity biologically struggles to cope with scaling and exponential changes. He derived practical intuition about "doubling" from the "flat part of the exponential curve" from 2010 to 2020. It was only until recent doublings that significant leaps in capability appeared. For example, the language models from four years ago struggled to accurately predict the next word.

Based on this, he predicts the next wave of capability: current models are still “single-prediction engines,” while humans and many animals are engines that continuously generate streams of accurate predictions. The ability to enable AI to make perfect and consistent predictions over extremely long time spans does not entail fundamental algorithmic or data obstacles. At that point, AI will not only draw on its knowledge about the world but will also use its "persistent state" memory of interactions with you and others to predict how things will unfold. This ability itself is enough to astound, and he expects it could be realized by the end of next year.

Regarding whether AI should be endowed with rights or considered conscious, Suleyman takes a firm opposing stance, believing this to be complete anthropomorphism and utterly absurd. His concern is that some may intentionally design AI that is manipulative, persuasive, and possesses its own motives and independent will. He emphasizes that "human-centered superintelligence" means creating technologies that serve you, not the other way around. He cannot be held responsible for what others in the field create, but this is precisely the challenge that requires collective action to address.

Confronted with the possible "dark age" arising from deep forgeries and information distortion (where people refuse to believe anything due to an inability to distinguish truth from falsehood), Suleyman distinguishes between "skepticism" and "cynicism." Skepticism is a healthy philosophical stance used to pose difficult questions and confront the reality of answers; whereas cynicism leads to passivity and negativity. We must hold accountable for what we build and apply pressure to those who produce inferior products.

Finally, Suleyman shared his experience founding the "Muslim Youth Help Line" in his early years. This initiative arose against the backdrop of rising Islamophobia after 9/11, led by young people and serving a diverse community of youth. This experience taught him the power of community, resilience, respect, and empathy, which also influenced his later founding of Inflection AI and the development of empathetic AI "Pi." He believes that creating technology that makes people feel heard and understood is essential for a worthwhile life, and this remains a direction he actively pursues today.

In Suleyman's view, the coming decades may be the most productive period in human history, anticipated to liberate people from work and significantly enhance quality of life. However, these technologies alone will not automatically yield net benefits; rather, benefits stem from our progress in self-governance as a species. The collective responsibility lies in not shying away from the darkness but confronting the real risks that exist while maintaining optimism, confidence, hope, and a stance of connection and unity with humanity.

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。