Meta CEO Mark Zuckerberg released an essay defending artificial intelligence as an innovation capable of generating prosperity and empowering individuals. However, he warns that the primary danger lies in the concentration of this technology within a single organization.
Zuckerberg expresses his opposition to regulations that could compromise the advancement of AI in the United States, arguing that such measures could pave the way for other nations to assume technological leadership.
In his manifesto, the executive assesses that the most significant risk to AI is when only one entity, whether a government or a corporation, has access to the most powerful tools. For this reason, he opposes norms that would put the United States at a competitive disadvantage against other countries.
The main points raised by Zuckerberg include the view that AI agents can become super-personal, assisting in areas such as health, finance, career, and relationships, promoting human empowerment. While acknowledging that AI may reduce the number of employees in companies, he also foresees the emergence of new small businesses and the facilitation of entrepreneurship. The true potential of AI, according to him, lies in its ability to facilitate new inventions, not merely in task automation.
He emphasizes that AI development should be guided by principles of freedom, open research, free enterprise, and equal opportunity. The greatest risk pointed out is centralization, as it can lead to totalitarianism, mass unemployment, or threats to humanity itself. Despite this, Zuckerberg believes that technology will make humanity more prosperous, healthy, and free, following the pattern of past innovations.
In addition to theoretical principles, Zuckerberg's text details concrete plans for Meta itself. The company intends to resume the release of open-weight models, aiming to 'put power in people's hands'.
Other initiatives include making AI available for free or at reduced costs, while paid versions would be determined through a dynamic auction of computational capacity. Additionally, a billion-dollar fund will be created to support the communities where Meta operates its data centers. Finally, the company will grant greater autonomy to the board of directors regarding the safety criteria applicable to AI models.
Meta CEO Mark Zuckerberg expressed severe criticism regarding the artificial intelligence (AI) development methods employed by his competitors, citing OpenAI and Anthropic. In an interview given to The New York Times this Tuesday (28), the executive warned that centralizing the creation of advanced models in a few corporations could restrict society's access to this technology and consequently hinder innovative progress.
Although he did not name OpenAI and Anthropic directly at all times during the conversation, Zuckerberg indicated that his criticisms were directed at companies advocating for stricter control over AI. According to him, this stance diverges from the traditional culture of the technology sector in the United States. The executive commented that much of the debate in other laboratories is excessively pessimistic, emphasizing the need for voices that introduce a sense of realism into the discussion.
These statements reinforce an existing polarization in Silicon Valley regarding how advanced AI systems should be distributed and created. On one side, executives like Dario Amodei of Anthropic and Sam Altman argue that certain advanced models already present excessive risks and require strict control mechanisms before being released or developed. On the other hand, large companies such as Microsoft, Nvidia, Google, and Meta itself maintain that a significant portion of these models should remain widely available to drive new research, businesses, and technological advancements.
This controversy gained more intensity this month after Chinese AI models diminished the technological lead that American companies, such as OpenAI and Anthropic, held in the sector. In reaction, these companies increased dialogue with regulators in Washington and expressed concern about the advancement of open-source models originating from China.
On the previous Friday (24), Jensen Huang, CEO of Nvidia, released a letter advocating for the use of open-source software for AI, a position supported by several other technology companies. Additionally, Huang announced the formation of a coalition focused on developing open cybersecurity and AI safety tools, uniting entities such as SpaceX and IBM. Huang and Altman were among the executives who visited Washington to discuss possible guidelines for the sector with President Donald Trump's government.
During the interview, Zuckerberg reiterated his defense that AI must be accessible to the largest number of people. He contested the idea of creating a single superintelligence fully aligned with human interests, a concept common in AI safety discussions. The executive stated that he was skeptical about the notion that a singular type of AI could be guaranteed to benefit humanity through idealized alignment.
In contrast, he proposed the concept of 'personalized superintelligence,' where AI assistants can be adjusted according to each user's individual preferences, values, and needs. Zuckerberg affirmed that it is practically impossible for a single benevolent superintelligence to satisfy everyone simultaneously.
As of the release of the interview, neither OpenAI nor Anthropic had commented on Zuckerberg's statements. Previously, OpenAI had declared its support for open-source initiatives and mentioned its work with the White House to establish safe methods for AI expansion. As recently as Monday (27), Amodei published a text detailing that Anthropic is focused on preventing advanced chips from reaching authoritarian regimes, avoiding massive replication of AI models, and requiring safety testing for highly capable systems, regardless of whether they are open or closed.
Despite the differences between the companies, researchers from Anthropic, OpenAI, Meta, and Google collectively signed an open letter called 'Pacing the Frontier.' This document advocates for a global effort to create technical tools and governance mechanisms that allow for slowing down AI advancement when safety requires it.
Meta's approach to AI has also undergone recent transformations. Previously, the company was known for making open-source AI models freely available. However, after losing ground to OpenAI and Anthropic, Zuckerberg redefined the strategy. The company invested $14.3 billion (R$ 73.5 billion) in the startup Scale AI and hired its founder, Alexandr Wang, to lead the AI division, which was renamed Meta Superintelligence Labs. Furthermore, billions of dollars were allocated to hiring research specialists, who began developing a new closed AI model instead of a fully open one. This month, Meta began selling access to its Muse Spark model for the first time, while maintaining the development of open-source models.
To conclude, Zuckerberg expressed his conviction that the trend in the technology industry points toward a progressive democratization of AI tools, concluding that the general course of the sector has always sought to make technology more accessible to more people, not fewer.
