Author: Techub News Compilation

Introduction

In August 2025, Ethereum co-founder Vitalik Buterin (known as V) made a rare guest appearance on the podcast "Doom Debates," which focuses on exploring the existential risks of technology, engaging in a nearly two-and-a-half-hour deep conversation. In this dialogue titled "Vitalik vs. AI Doomer," Buterin publicly presented, for the first time, his systematic assessment of the survival risks posed by superintelligent AI (ASI) and shared his conceptual framework for addressing these risks—"Decentralized Democratic Defensive Acceleration" (d/acc).

As a prominent thought leader in the blockchain space, Buterin has long been concerned about technological governance, institutional design, and existential risks. However, he had previously discussed AI alignment and the probability of doom (P(Doom)) in far less detail in public venues. This occasion aligns with the rapid advancements in AI capabilities and the increasing global discourse surrounding AI risks, where Buterin offers valuable insights from the cryptocurrency world through his unique interdisciplinary perspective and calm, rational analysis on this matter concerning humanity’s future.

Summary

• Vitalik Buterin (V) estimates the probability of human extinction due to superintelligent AI (P(Doom)) to be about 12%, based on a comprehensive assessment of technological development timelines and geopolitical situations.
• He proposed the concept of "Decentralized Democratic Defensive Acceleration" (d/acc), advocating for the construction of a diverse, decentralized ecosystem of agents to address AI risks, emphasizing the combination of defensive technology with open, democratic governance.
• Buterin believes discussions about AI risks need to move beyond personal attacks and ideological divides, returning to high-quality rational debate, and he urges the tech community to respect and seriously consider the concerns of AI safety researchers.
• He considers himself both a technological optimist and a believer in technology's capacity to solve many problems, but he emphasizes that superintelligent AI has "independent agency," representing a fundamental risk distinct from all previous technologies that requires special treatment.

V's "Doom Probability": Why 12%?

When directly asked by the host, "What is your P(Doom)?" Vitalik Buterin (V) provided a specific figure: approximately 12%. He clarified that this probability is not static but is dynamically adjusted based on assessments of technological development timelines and global situations. About a year ago, his estimate might have hovered around 10%, then dipped at one point but has recently risen back to 12%, achieving even 15%-16% earlier this year.

Two main factors have influenced his judgment: the first is the worsening geopolitical environment. Over the past year, the atmosphere and willingness for international cooperation seem to have degraded, weakening the global capacity to address potential global crises, including AI risks. Buterin posits that a more divided and foolish geopolitical environment is a significant factor raising the risk level. The second factor is the shortening timeline for AI capability development. Innovations and advancements in technologies like "Chain of Thought" have compressed the time he previously anticipated AI would need to reach critical capability thresholds. He explicitly stated that most of his "doom probability" is concentrated before 2050; the shorter the timeline, the higher the P(Doom); conversely, a longer timeline would reduce the risk probability.

Regarding the definition of "doom," Buterin confirmed in clarifying discussions with the host that it primarily refers to the extreme scenario of human extinction due to superintelligent AI. He places his probability estimate within a broader "rational range," considering ranges proposed by figures like OpenAI's Ilya Sutskever—between 10% and 90%—to be reasonable, while extreme confidence too close to 0% or 100% might indicate overconfidence.

In 2023, Buterin signed the well-known "AI Risk Statement," which prioritized mitigating the extinction risks caused by AI alongside pandemics and nuclear war. He explained that superintelligent AI possesses the logical persuasiveness of the "standard doom story"—rapidly growing capabilities, orthogonality thesis, and tool convergence could lead a super powerful AI to act against human interests. Given that societal resources addressing this risk are far from matching its development speed, he believes it is vital to raise alarms and mainstream this issue.

Examining AI's Future: Timelines, Limitations, and the Possibility of Surpassing Humanity

Buterin holds a cautiously open attitude toward the timelines for the arrival of artificial general intelligence (AGI) and superintelligence (ASI). He presents two coexisting narrative frameworks: the first is the "acceleration train" narrative, where AI rapidly progresses under the drive of computational investments, capability growth, and self-sustaining industries, suggesting that AGI might emerge in the early to mid-2030s according to current indicators (such as the length of time AI can perform tasks without human intervention), with a quick leap to superintelligence. The second is the "historical recurrence" narrative, which posits that the current progress in AI (particularly large language models) may primarily automate a subset of human abilities (such as interpolation), akin to the AI boom of the 1970s, but could encounter bottlenecks when it comes to truly innovative, extrapolated, and comprehensive handling of entirely new complex tasks (like independently founding and running a large online enterprise), delaying the realization of AGI until mid-century or later.

In Buterin's view, the core limitation of current AI systems based on large language models is their severe reliance on massive amounts of training samples, excelling in "human footprint dense" fields (interpolation), but performing poorly in tasks requiring deep original thinking, new domain processing, or systematic extrapolation. He cites his own experiences in cutting-edge cryptography or deep debugging in Linux systems to illustrate that AI can sometimes propose seemingly clever solutions but often fails in actual execution. This limitation is crucial for risk assessment: if AI cannot robustly innovate at a "scientific paradigm" level outside the current knowledge distribution, then the trajectory leading to rapid "capability explosions" and loss of control may be interrupted.

However, Buterin firmly rejects any reductionist dismissal of "AI will always just be…". He believes that either AI is now more than a simple mechanism, or it will soon surpass that, or humanity itself may also be explainable by a similar mechanism. Regarding the debate of "whether consciousness is a firewall," Buterin holds an algorithm-centric view, arguing that consciousness is an attribute of specific types of algorithms; if physical laws can be realized in silicon, then simulating the human brain or developing AI with similar functions could lead to the emergence of consciousness. He dismisses the possibility of "philosophical zombies," asserting that entities capable of discussing conscious experiences likely possess consciousness themselves.

Looking ahead to the "headroom" of surpassing human intelligence, Buterin believes ASI has great potential to far exceed humanity. The key lies not just in a linear increase in IQ-like intelligence but in the orders of magnitude leap in thinking speed. An AI that thinks thousands of times faster than a human, even if its "IQ" is only slightly above that of an ordinary person, could achieve planning and operational capabilities far beyond human abilities on a subjective time scale, fundamentally changing the landscape of economics, society, and even physical interactions (such as perceived light speed). At the same time, he remains open but uncertain regarding sci-fi level capabilities like molecular nanotechnology, suggesting that empirical analysis of specific technological pathways is necessary, while acknowledging that if such technologies are feasible, their power would be extremely formidable.

Warnings from a Technological Optimist: Why AI is Different?

Vitalik Buterin (V) defines himself as a technological optimist. He firmly believes that historically, technology has been the most powerful force driving the world toward good, from extending lifespans to curing diseases, and addressing environmental pollution, technological advancements continually solve the problems they create. He cites a graph showing the continuous rise in human life expectancy over the past century, pointing out that even massive disasters like World War II were ultimately surpassed and mitigated by technological progress.

However, he emphasizes specifically that superintelligent AI is a fundamental exception. Past technological products were essentially tools, still under human control. But superintelligent AI possesses independent agency, no longer merely a tool but a species that may surpass and replace humanity. This qualitative change brings the risk that two historical models may fail: first, the development process of AI may not afford a "restart." Typically, technological iterations allow for learning from mistakes, but once AI capabilities breach a certain threshold, its feedback loop can escalate rapidly, leaving no opportunity for humanity to correct with a "second version." Second, the extinction risks driven by AI have a "pursuing" characteristic. Unlike nuclear war or natural pandemics that might leave survivors, a superintelligent AI with clear objectives (even if not directly aimed at humanity) could systematically seek out and eliminate all humans to ensure task completion.

In response to the recent dichotomy between "effective accelerationism" (e/acc) and the "deceleration/safety" camp in technology discourse, Buterin criticizes this trend of positioning based on "atmosphere" rather than deep reflection. He understands the e/acc camp's aversion to excessive regulation and anti-growth sentiments has reasonable aspects in many specific tech issues, but argues that unconditionally elevating "acceleration" as a doctrine, particularly in the unique realm of AI, is dangerous and irresponsible. He points out that the quality of many ideological "declarations" currently is low, lacking clear principles, objectives, and pathways for realization, resembling emotional outbursts rather than serious contemplation.

"d/acc": A Decentralized, Democratic, and Defense-Oriented Response Framework

In response to the aforementioned risks, Buterin proposed a concept he calls "Decentralized Democratic Defensive Acceleration" (d/acc). The core of this idea is that, in the face of potential threats from superintelligent AI, mere technological deceleration or blind technological acceleration are not optimal solutions. Instead, we should accelerate the development of technologies and institutions that enhance societal resilience, defensive capabilities, and democratic oversight, ensuring that these capabilities are decentralized.

d/acc encompasses several key dimensions:

• Decentralized: Avoid excessive concentration of power and critical capabilities in a few entities (whether state or corporate). An ecosystem composed of diverse, independent actors can remain stable and resilient against attacks even if some of its components face issues or are maliciously controlled. Blockchain technology and its governance concepts exemplify this dimension in practice.
• Democratic: Ensure that the development and deployment of technology are subject to broad democratic oversight and accountability, preventing it from being used for inhumane or oppressive purposes. This pertains to the openness, inclusiveness, and legitimacy of governance processes.
• Defensive: Prioritize the development and deployment of defensive technologies. For instance, in the military domain, prioritize air defense and defensive systems rather than offensive weapons; in the biological domain, prioritize public health monitoring and response capabilities rather than pathogen creation. Defensive technologies often possess a "stability advantage," meaning they can lower the risk of conflict escalation and are more conducive to global safety, even if in the hands of multiple parties.
• Acceleration: This does not imply a full deceleration but rather a deliberate and rapid push forward in these specific directions regarding research and application.

Buterin believes the d/acc framework is applicable not only to AI risks but also to bio-risks, nuclear risks, and other existential threats. Its goal is to build a world that is "robust across multiple possible futures," where even if our judgments on many specific issues (including the difficulty of AI alignment) are flawed, we can avoid the worst outcomes. He acknowledges that the AI alignment issue may fundamentally be "intractable", hence we cannot place all hopes solely on the route of "perfectly solving the alignment problem," but must increase humanity's survival chances by constructing robust, defensive socio-technical structures.

Refuting Personal Attacks: The Importance of High-Quality Discourse

In the dialogue, Buterin and the host spent considerable time discussing the various personal attacks surrounding the AI doomsday narrative (Doomerism) and collectively refuted these arguments. They aimed to steer the discussion back to a rational exploration of the substantive issues.

In response to the claim that "doomsters are a fringe minority/echo chamber," Buterin pointed out that deep concerns regarding AI risks are quite common within the AI research community, and public surveys show that a majority has moderate to high concerns about it. Although the specific theoretical foundations of the public may differ, the direction of concern is similar.

In addressing accusations that "doomsters are not constructive/not in the arena," Buterin cited early builders concerned about AI risks, such as Jaan Tallinn (co-founder of Skype) and Dustin Moskovitz, along with scientists within major AI companies conducting alignment research, who are both deep risk concerners and accomplished practitioners.

In countering the view that "those who truly understand AI will not become doomsters," Buterin deemed this "very wrong," using figures like Eliezer Yudkowsky as examples, illustrating that many deep concerners possess a profound understanding of principles like machine learning, gradient descent, and Transformers.

Against the claim that "doomsday narratives are a tool used by large companies to seek regulatory arbitrage," Buterin acknowledged that commercial interests might indeed influence some discussions about “open source vs. closed source” security (for instance, arguments against open source could facilitate monopoly for commercial closed-source models), but stressed that this is far from being the primary motivation for many researchers in the AI safety domain; many thinkers are sincere and independent of those commercial interests.

Addressing the mockery that "people do not sincerely believe in doomsday/this is a nerd's alternative religion," Buterin analyzed from a psychological perspective, indicating that many worriers initially treated AI risks as an interesting intellectual puzzle but were eventually reluctantly forced (even “kicking and shouting”) to take its real consequences seriously over time as evidence accumulated, a process more resembling being dragged by rational arguments rather than actively embracing some doomsday belief. What attracted them more was the process of discovering the astonishing truths of the world through pure reasoning rather than the specific conclusion of "doomsday".

Buterin praised the rationalist community (especially early LessWrong) for shaping its values, including epistemic honesty, meta-level thinking, altruism, and transcending tribal positions. He emphasized that high-quality, candid discourse is crucial for addressing complex challenges like AI, and he actively participates in such dialogues, even amidst differences in viewpoints (such as differing specific estimates of P(Doom) between him and the host). He calls on the tech community, particularly those who respect his views on decentralization and system design, to also respect the profound thoughts of AI safety pioneers like Eliezer Yudkowsky and to take their warnings seriously.

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。