The American company Anthropic is urging artificial intelligence developers to agree in advance on mechanisms to slow the development of AI systems if their progress outpaces humanity’s ability to maintain control.
In its appeal, Anthropic emphasized that the emergence of AI systems capable of independently improving their models could be a significant technological breakthrough. However, there is a risk of losing control over such systems, necessitating discussions on the possibility of their coordination for safety.
The company considers it necessary to establish conditions and criteria for a temporary halt in the development of artificial intelligence, as well as methods for control and monitoring of their implementation. Anthropic stressed that unilateral actions by one developer would be ineffective.
To confirm the necessity of such measures, the company cited the example of its model Claude, which independently wrote over 80% of the company’s code base in May this year. Anthropic plans to hold consultations involving politicians, scientists, public organizations, and other developers to discuss the risks of self-improvement in artificial intelligence.
Anthropic, founded by former OpenAI researchers in 2021, has already made significant strides in the development of artificial intelligence. Among their achievements are the chatbot Claude and the new direction AI & Rule of Law. The company also announced plans for public access to Mythos-class models. Meanwhile, Anthropic is under the scrutiny of the US National Security Agency, although the Pentagon has included it on the list of risky suppliers.
Developers believe that without coordinated efforts in the regulation and control of artificial intelligence, the world may face unpredictable consequences, the impact of which will require not only technological but also social adaptations.
