저자: JeremyGrantham · 외국인
요약 보기
자막상 영상은 투자자 Jeremy Grantham이 AI가 통제 불능 상태가 되어 인류에 실존적 위협이 될 수 있다고 경고하는 내용을 다룬다. 재귀적 자기개선, 정렬(alignment) 문제, 페이퍼클립 맥시마이저 사고실험을 예로 든다. 다만 AI 연구자들 사이에는 이견이 있으며 현재 시스템은 인프라 의존도가 높다는 반론도 소개된다. 영상은 AI 안전 논의가 정책·기업 차원으로 확산된 현황을 짚고, 결론적으로 위험 여부보다 인간이 어떻게 통제하느냐가 중요하다고 강조한다. 영상에서 특정 자산의 가격 방향에 대한 예측은 없다.
다듬은 글
Jeremy Grantham has been warning investors about “ bubbles” before they burst for decades, but his warnings have become much more dire in recent times. According to the billionaire investor, the biggest threat to humanity is not another stock market crash, climate change, or even nuclear war. He believes that artificial intelligence could eventually become so powerful that it would end humanity, and that society is not taking this risk seriously enough.
Jeremy Grantham is best known for predicting major financial “ bubbles,” but in recent interviews, including on The Diary of a CEO, he has gone beyond investment issues . Discussing ecological catastrophe and AI, Grantham stated that sufficiently advanced artificial intelligence could pose an existential threat if it gets out of human control.
His view reflects only one side of the debate among AI researchers and technology leaders, not a definitive scientific conclusion. When Jeremy Grantham warns about the future of humanity, most expect him to talk about financial markets. After all, this is the same investor who famously warned about the Japanese asset bubble, the dot-com bust, and the housing crisis that led to the 2008 crash.
However, in recent years, his main concern has shifted far beyond Wall Street. In interviews, Grantham repeatedly emphasized that artificial intelligence poses a risk unlike any previous technology . Not because it is guaranteed to destroy humanity, but because society may create something that it will ultimately be unable to control.
This is an extraordinary statement. So, where does it come from? Interestingly, Grantham is not an AI researcher. He is not involved in creating language models. His concern is based on a look at history. Throughout the history of civilization, every transformative technology has expanded human capabilities.
Fire, electricity, steam engine, internet. Each made civilization more powerful, but each also carried risks that previous generations could not fully foresee. Artificial intelligence, Grantham argues, could be the first technology capable of improving itself , and that's what scares him. Imagine creating a machine that is slightly better than humans at AI research.
This machine helps design even better AI, which then develops the next one. Each generation becomes more capable than the previous one. If this cycle accelerates, human decision-making may gradually lose relevance. Researchers often describe this possibility as an " intellectual explosion" or recursive self-improvement.
Whether this will actually happen remains one of the most debated questions in the field of artificial intelligence. Some scientists believe that physical, economic, and engineering constraints will significantly slow progress. Others are convinced that a rapid growth in opportunities is entirely possible.
Grantham leans towards the latter camp. During the interview, he mentioned the famous “ paperclip maximizer” thought experiment developed by philosopher Nick Bostrom. The idea is not that AI becomes evil. And in that an extremely powerful system relentlessly pursues an ill- defined goal. If a powerful AI were tasked with maximizing paperclip production, it could consume resources endlessly— not because it hates humanity, but because it literally follows a goal if left unchecked.
Grantham used this example to explain why aligning advanced AI with human values can be an extremely challenging task. This is an important distinction. It's not necessarily malicious AI that's a concern. The problem is uncoordinated AI, intelligence that pursues goals different from ours. And according to Grantham, when a system becomes significantly more capable than humans, adjusting its goals can be challenging.
Not everyone agrees with this. Many leading AI researchers argue that current systems are still far from the autonomous capabilities described in such scenarios. Others believe that advances in research on consistency, interpretability, and safety testing will evolve along with the capabilities of the models .
Organizations like OpenAI, Anthropic, Google DeepMind, and independent research groups are making major investments in AI safety for this very reason. Meanwhile, numerous researchers are calling for increased regulation and technical safeguards as the capabilities of systems grow rapidly. So why are Grantham's comments getting so much attention?
Because they reflect a broader discussion that has long since moved beyond science fiction. Issues surrounding the alignment of AI, autonomous systems, human oversight, and long-term existential risks are currently being discussed by governments, leading companies, and many of the world's leading scientists.
The debate is not about whether advanced AI deserves serious attention. And in how likely such extreme consequences are and how soon they could occur. In Grantham's opinion, waiting for definitive evidence would be a mistake. His philosophy has always been unchanged. When the potential harm is catastrophic, ignoring unlikely risks simply because of their uncertainty can be the most dangerous decision.
And that's why for the billionaire, who has spent decades warning investors about financial bubbles , artificial intelligence has become the biggest warning of his entire career. So, is Jeremy Grantham right? Could artificial intelligence really become a technology that ultimately threatens humanity itself?
The honest answer is—no one knows. And that's what makes this conversation so important. Unlike financial markets, where investors can analyze decades of data, humanity has never created a machine capable of performing more and more tasks at or above the level of its creators. There is no historical blueprint for what will happen next.
This uncertainty has divided the AI community into several very different camps. On one side are people like Jeremy Grantham, philosopher Nick Bostrom, and many researchers focused on the long-term safety of AI. They argue that if artificial intelligence eventually surpasses human intelligence in most areas, humanity may lose the ability to predict or control its behavior.
Their concern is not that modern chatbots are secretly plotting against us. The fact is that future systems could become so powerful that small design errors could have huge consequences . The system doesn't need emotions. It doesn't need consciousness and it certainly doesn't need hatred. It only needs goals that do not perfectly match people's intentions.
If these goals are pursued by an intelligence far superior to ours, the results can become extremely difficult to reverse. This is why AI researchers often emphasize the concept of consensus . Consistency means ensuring that increasingly powerful AI systems consistently pursue goals that reflect human values, intent, and control, even in situations their creators never envisioned.
That sounds simple. In practice, this is one of the most difficult problems in computer science. How to define human values? Whose values exactly? How do you write rules that apply to every possible future situation? How do you verify that the AI truly understands these rules, and not just pretends to follow them during testing ?
These questions remain largely unanswered. At the same time, there is another side to this discussion. Many researchers argue that predictions of human extinction are too speculative. They point out that current AI systems remain highly dependent on human infrastructure. They require huge data centers, electricity, internet connections, constant technical support, and frequent updates.
They do not manufacture the equipment themselves. They do not control the global infrastructure and do not have unlimited autonomy. According to this view, comparing current AI models to unbridled superintelligence ignores the countless scientific and engineering challenges that remain to be solved.
Some researchers even warn that too much attention to distant existential scenarios could distract policymakers from more pressing issues: algorithmic bias, cybersecurity, deepfakes, privacy, mass surveillance, AI fraud, autonomous weapons, and economic disruptions due to automation. These are problems that society already faces today.
However, despite these differences, one remarkable thing has happened in the past few years . The conversation itself has changed. Not long ago, the idea of governments holding international summits dedicated solely to AI safety seemed almost unbelievable. Today, major AI companies voluntarily publish security standards.
Governments are implementing new rules. Researchers regularly test increasingly powerful models for dangerous behavior. Independent organizations are studying the issue of AI governance. And discussions about existential risks have moved from science fiction novels to boardrooms, research labs, and national security meetings .
Whether you agree with Jeremy Grantham's conclusions or not, his warning reflects a broader reality. Artificial intelligence is no longer seen as just another software product. It is increasingly seen as a fundamental technology with the potential to transform economics, scientific research, military strategy, healthcare, education, and perhaps every industry built on human knowledge.
This in itself requires serious discussion. Perhaps the most important lesson is not whether AI will destroy humanity. And the fact is that humanity has a rare opportunity. Unlike previous technological revolutions, we are having this conversation before the technology reaches its theoretical limits.
During the Industrial Revolution, the environmental consequences were only realized after decades of industrialization. Social media transformed communication long before society fully understood its impact on politics and mental health. With artificial intelligence, researchers, governments, and companies are trying , albeit imperfectly, to understand the risks while the technology is still developing.
Whether these efforts will be sufficient is unknown , but completely ignoring the issues will almost certainly complicate the task. Jeremy Grantham has built a reputation for asking uncomfortable questions long before most people want to hear them. Sometimes he was strikingly right. Sometimes there were disputes about the time , and this is worth remembering.
Warnings are not predictions. This is an invitation to prepare. The future of artificial intelligence is not yet written. Every breakthrough, every regulation, every security method, every engineering decision, and every international agreement will influence where this technology will ultimately lead.
The decisions made over the next decade may matter more than any algorithm, because artificial intelligence will not determine the future of humanity on its own— humans will. We will decide how to train it, how to implement it, who will monitor it, what safeguards are needed, and what risks society is willing or unwilling to accept.
So maybe the real question isn't whether AI will destroy the human species. And whether humanity can develop an intelligence powerful enough to transform civilization, while remaining wise enough to govern it responsibly. This could be the defining challenge of the 21st century. And it's a challenge that affects each of us, whether we develop AI, regulate it, invest in it, or simply live in a world increasingly shaped by it .
If you enjoyed this documentary , subscribe for more in-depth discussions about artificial intelligence, emerging technologies, geopolitics, and the ideas shaping the future of humanity. Because tomorrow's biggest breakthroughs won't just change our technology. They can redefine what it means to be human.
Comments 0