Summary of the Key Points
Zuckerberg's lengthy article appears to be a declaration on AI policy, but in reality, it outlines Meta's future strategy: to promote "personal superintelligence" – where everyone has their own AI assistant, rather than superintelligence being concentrated in the hands of a few companies or governments. Meta aims to distribute AI capabilities from large institutions to ordinary people through open-source models, privacy protection, and infrastructure development, with the ultimate goal of gaining control over the next generation of computing platforms (such as personal intelligent agents combined with smart glasses), thereby breaking free from the dominance of Apple and Google's operating systems.
1. The Core Argument: Superintelligence Should Not Be in the Hands of a Few
Zuckerberg believes that while current discussions about AI security focus on the fear of "superintelligence getting out of control," he is more concerned about another issue: if only a few companies or governments possess superintelligence, it could lead to a new imbalance of power. For example, if only one entity has superintelligent lawyers, it would likely win all legal battles; similarly, if only one entity has AI-powered cybersecurity capabilities, it could become the greatest threat.
His solution is "decentralization": ensuring that no single organization controls superintelligence, but rather everyone has their own AI assistant. This way, everyone would have similar capabilities and be able to check each other's power. For instance, with everyone having a superintelligent lawyer, legal resources would be more equitable; and with everyone using AI for defense against cyberattacks, overall security would be enhanced. He also argues that since humans do not share a unified set of values, centralized AI would inevitably favor certain groups, so it is better to let each person's AI be tailored to their own needs.
2. The Future Form of AI: Your "Long-Term Intelligent Assistant"
Current AI systems (like ChatGPT) are mostly "question-and-answer machines," but Meta aims to create "personal intelligent agents" that act as long-term companions, understanding your work habits, health status, and family schedules. These agents could help you write code, design products, plan tasks, and even complete tasks that used to require a company's resources.
For example, an AI agent could remind you to exercise based on your health data, organize your work emails, or plan a family trip for you. Zuckerberg emphasizes privacy, stating that such AI systems will need access to a lot of personal information. Therefore, Meta plans to implement a "completely private mode" where even Meta itself cannot see your data, transforming AI from a company tool into a personal capability.
Regarding employment, he is not worried about AI taking jobs; historical technological revolutions (such as the Industrial Revolution) have always created new job opportunities. He believes that AI will enable ordinary people to acquire skills previously reserved for large companies (like product design and market research). In the future, companies may become smaller, but individuals' productivity will increase, allowing more ideas to be realized.
3. Competitive Strategy: Using Open-Source Models to Build Alliances
Meta has consistently supported open-source AI models (such as the Llama series), which is different from the closed models developed by OpenAI and Anthropic. Zuckerberg views openness not just as a technical choice but as a fundamental principle related to how superintelligence is distributed. More people using open-source models will help identify and fix vulnerabilities, preventing a few companies from monopolizing the technology.
He also discusses global competition, noting that while the U.S. has an advantage in chip design, it falls behind in energy and data center construction speed. Future AI competition will not only involve algorithms but also the availability of computing power and energy. He urges the U.S. not to overly restrict open-source development, as this could give other countries an opportunity to catch up.
On security, he advocates for enhancing defensive capabilities rather than limiting AI's potential. For example, if everyone can use AI for cybersecurity, attackers will face greater challenges; for biological risks, it is better to regulate dangerous substances rather than prevent AI from helping in drug research and development.
4. Infrastructure: Data Centers That Benefit Communities
AI requires numerous data centers and energy, but their construction is often criticized for being resource-intensive. Zuckerberg proposes that Meta should enter into agreements with communities, ensuring that the establishment of data centers brings local employment, educational investments, and public services. This could include donating AI courses to schools and building parks in the community, turning data centers from resource consumers into drivers of regional economies.
He also discusses corporate governance, stating that the deployment of superintelligence should not be decided by a single CEO. Meta plans to have an independent board oversee security reviews to address concerns about tech companies gaining too much power.
5. The Ultimate Goal: Gaining Control of the Next Generation of Computing Platforms
Meta has thrived in the past two decades through social connections (Facebook, Instagram), but it has always been constrained by Apple and Google's operating systems. Now, AI offers an opportunity to change this. If everyone has an intelligent agent that understands them and can perform tasks for them, this agent could become the new operating system.
Smart glasses are not just hardware; they will enable AI to interact with the real world, providing users with immediate feedback (for example, showing restaurant ratings). Open-source models will further expand Meta's influence, as more developers use its technology, thus bringing more people into its ecosystem.
However, there is a contradiction: although superintelligence is distributed to individuals, the computing power, models, and devices that support it are still controlled by large companies like Meta. Whether Meta's approach will succeed depends on its ability to bypass Apple and Google and establish its own platform. It also raises the question of who will control this AI and the data it processes.
In summary, Zuckerberg's article reflects Meta's "survival strategy for the AI era": using the idea that "superintelligence belongs to everyone" to unite developers and users against competitors with closed models, with the ultimate goal of gaining dominance over the next generation of computing platforms. This is both a technological roadmap and a business ambition—Meta aims to transform from a social media giant into a platform leader in the AI era.