Vitalik Buterin: The correct development of AI is to become the "mechanical armor" of human thinking to achieve super intelligent civilization.

Ethereum co-founder Vitalik Buterin today commented on the development of AI, stating that the wrong development of AI could create independent and self-replicating intelligent life forms, while the correct direction is to become a 'mecha suit' for human intelligence, in order to achieve a prosperous superintelligent human civilization. (Background: V fears AI development: humans should create convenient tools, not intelligent life) This week, Nvidia CEO Huang Renxun mentioned in his keynote speech at CES 2025 that 'AI agents have the potential to become a multi-trillion dollar industry.' This has fueled the enthusiasm of the encryption community for AI agents. According to Kaito data, AI's attention this week has reached close to 70% for the first time. Source: @sandraaleow V: Correct development of AI equips human intelligence with 'mecha suits' Today, Ethereum co-founder Vitalik Buterin also commented on the development of AI, stating that the wrong development of AI could create independent and self-replicating intelligent life forms. The correct development of AI is to become a 'mecha suit' for human thinking. If we only achieve the former and ignore the latter, humanity may face the risk of permanent disempowerment; but if we achieve the latter, we may realize a prosperous superintelligent human civilization. In terms of achieving the latter, Vitalik further explained: Distilling human judgment is one way to achieve the latter at the collective level. Many AI video generation tools today have shown signs of this trend: it is not simply 'inputting commands and waiting for results,' but rather an immediate process of precise control and adjustment. Vitalik also included a screenshot of an article he wrote in November last year, explaining a method of 'distilling' human judgment through 'information finance,' with the aim of enabling the complex, expensive, and efficient human judgment mechanism to operate in a low-cost, fast, and efficient manner. He quoted Robin Hanson's idea, suggesting the use of 'prediction markets.' Through market mechanisms, we can obtain a 'distilled version' of the original expensive mechanism in a reliable, neutral, fast, and inexpensive manner (here, the use of the term 'distillation' is similar to the process of LLM distillation). Over time, this distillation mechanism will roughly mimic the behavior of the original mechanism, because only participants who help achieve the correct results can make money, while others will lose. From his sharing of this content, he may be exploring 'how to efficiently integrate human values and wisdom into the decision-making of AI systems,' and emphasizing the importance of economic incentives in this process. AI done wrong is making new forms of independent self-replicating intelligent life AI done right is mecha suits for the human mind If we do the former without the latter, we risk permanent human disempowerment. If we do the latter, flourishing superintelligent human civilization - vitalik.eth (@VitalikButerin) January 10, 2025 Should AI agents be created? On the other hand, in response to a question from netizens about 'whether the design choice should be agent-based AI or core-enhancing human capabilities,' Vitalik said: 'Agents' is an interesting term. Sometimes it refers to 'AI that can run autonomously and formulate complex plans, continue for days without human intervention,' but more often it simply refers to 'chatbots replacing GUI as the main way of interaction for applications.' Vitalik frankly stated that the latter is great. This is consistent with an article he published in December last year about the ideal Wallet. In that article, he discussed the profound revolution in interaction methods that AI progress could bring, pointing out that in the future, it may transition from traditional click operations and input modes to a new mode of 'speak your needs, and the robot helps you complete.' He mentioned: Through natural language input, eye tracking, and even future more direct brain-machine interfaces (BCI), as well as a deep understanding of user history (such as text message records, provided that all data is processed locally), 'Wallet' can more intuitively understand the user's intentions. AI can translate these intentions into specific 'action plans,' a series of on-chain and off-chain operations to meet user needs. As a result, the demand for third-party user interfaces will be greatly reduced. If users need to interact with third-party applications or other users, AI should detect potential risks in advance and provide solution suggestions based on user interests. Vitalik added that ideally, these AIs should be developed by different teams to form a diverse open ecosystem to meet different preferences and needs. Advocating the development of tools that enhance human intelligence At the end of last December, when discussing AGI (Artificial General Intelligence) and ASI (Artificial Superintelligence), Vitalik proposed his definition of AGI: AGI is a powerful enough artificial intelligence that, if all humans suddenly disappeared one day and this AI was uploaded into the body of a robot, it could independently maintain the operation of civilization. He also admitted that this is an extremely difficult definition to quantify, but this is the potential intuitive difference in the minds of many people between the 'AI we are used to' and 'AGI.' Evolving from a tool that constantly relies on human input to a self-sufficient life form is the core of this difference. As for ASI, Vitalik sees it as a higher-level concept: when human involvement (human-in-the-loop) no longer brings any value to the system, it has reached the standard of ASI. He pointed out that even in the game of chess, AI has only reached this level in the last ten years. He candidly admitted that ASI makes him afraid, and even the AGI he defined makes him uneasy, because these technologies may bring significant risks of human loss of control. Therefore, Vitalik is more inclined to focus on developing tools that enhance human intelligence, rather than building some form of super intelligent life. His position is consistent with the viewpoint he mentioned today: the correct development of AI should equip human intelligence with 'mecha suits,' rather than creating new independent intelligent life forms. AGI: Refers to artificial intelligence with human-like intelligence, able to perform tasks in multiple different fields, not limited to specific applications. ASI: Refers to artificial intelligence that surpasses all levels of human intelligence, having outstanding capabilities in specific fields and comprehensive superiority in every aspect surpassing humans. Related reports: Vitalik: In the AI era, Deepfake...