Anthropic Calls for a Pause in the Development of Advanced AI

JAKARTA - Anthropic warns the world needs to have the option to slow down, even temporarily stop, the development of advanced AI. The reason is that the pace of AI is getting faster, while humans are not necessarily able to catch up.

Quoted from the Anadolu Agency report, Sunday, June 7, the California-based AI company called for global coordination to open up opportunities for a leading AI development pause. This step, according to Anthropic, can only be taken if other major AI developers participate and their compliance can be checked.

"We believe it would be good for the world to have the option of slowing down or temporarily stopping the development of leading-edge AI, so that social and research structures can keep up with the progress of this technology," Anthropic wrote in its report.

Leading AI is the most advanced AI system that is at the forefront of technology development. Meanwhile, AI alignment means efforts to ensure that AI continues to work according to the goals and values that humans want.

Anthropic assessed that the development pause could not be done alone. There must be a number of large laboratories in several countries that agree to stop under the same conditions. They must also be able to ensure that other participants actually stop, not continue to develop in secret.

The company said the role of humans in the development of AI is narrowing. AI is now increasingly doing tasks that were previously done by engineers and researchers.

"The evidence suggests that the human role is narrowing at every stage of the AI development process," the report said.

Anthropic also estimates that when AI-generated code is equivalent to human-generated code, humans can stop writing code and only act as reviewers.

According to an Anadolu Agency report, more than 80 percent of the code that goes into Anthropic's production code base is now written by Claude, the company's AI assistant. Anthropic engineers are also said to produce code eight times more per quarter than the period from 2021 to 2025.

Anthropic warns that human oversight could be a new weak point. Because humans are not necessarily able to check the code as fast as AI makes it.

"Simply put: the work of doing something - that is, writing code, running experiments, producing results - now requires almost no human time," the company said.

This trend, according to Anthropic, can lead to recursive self-improvement. This term means a condition when AI is able to design and develop its own successor version.

On the one hand, this ability can accelerate the progress of science and technology. But on the other hand, Anthropic warns of a more serious risk. Namely, humans can lose control over the AI system.

"The window to investigate these questions together is open now," said Anthropic. The company urged policymakers, researchers, and civil society groups to discuss AI governance.