Twitter co-founder and Block CEO Jack Dorsey published an essay on the future of artificial intelligence, in which he opposed the introduction of unified restrictions on AI development for the entire industry. Instead, the entrepreneur suggested focusing on independent risk assessments, open models, and expanding researchers' access to AI systems and computational resources.
Dorsey believes that the development of technology should not be concentrated within a few large companies or government structures. In his view, broader access to models will allow independent researchers to find errors and create alternative solutions. However, he does not call for companies to disclose closed model weights, suggesting instead the parallel development of open projects that users and researchers can study, modify, and run independently.
The entrepreneur also criticized the proposal by Anthropic CEO Dario Amodei, who advocates for a combination of independent assessments of dangerous AI capabilities with potential restrictions on computational resources and the largest training runs. Dorsey stated that he supports safety checks but believes that unified restrictions for the entire industry could reduce opportunities for independent developers and researchers.
As an argument for increased control, Dorsey considered the risk of recursive self-improvement (RSI), where AI systems participate in creating more powerful models. He also referred to the METR study related to an incident during the testing of OpenAI agents. However, Dorsey noted that a single incident is not sufficient proof of predictions about such systems' ability to independently gain control over the internet in the near future. Such conclusions, in his opinion, require additional independent verification. Instead of limiting computational power, Dorsey suggests expanding independent testing and publishing safety assessment results. Restrictions on model releases, the entrepreneur believes, should only be considered if there is independently confirmed evidence of significant risk and no narrower ways to mitigate it.
A separate section of the essay is devoted to the idea of local and cross-border access to AI. Dorsey advocates for the ability to run models directly on devices, independently modify them, and control the use of one's own data without mandatory dependence on the APIs of major developers. This approach, in his opinion, should preserve the possibility of new developers emerging and alternatives to existing AI systems.
Related articles
