Artificial intelligence company Anthropic has warned that the increasing ability of the Claude model to write its own code indicates that systems could soon slip out of human control, highlighting the risk of “recursive self-improvement.”
“Elchi” reports that Anthropic, a leading artificial intelligence company, has warned that next-generation AI systems have the potential to develop beyond human control.
According to recent evaluations of the company’s Claude model, the system is now capable of writing more than 80 percent of its own code. This ratio was below 10 percent at the beginning of last year.
Anthropic officials stated that this rapid change points to a process called “recursive self-improvement,” which could reach a level where artificial intelligence might create its own successors in the future.
In a blog post published by company researchers Marina Favaro and Jack Clark, it was stated that while this trend could lead to significant scientific and technological breakthroughs in the long term, it could also pose serious control risks.
The statement emphasized that the AI development process is already partially delegated to AI systems, paving the way for the emergence of models that can improve themselves without human intervention.
The company warned that if this process continues unchecked, humans could lose control over AI systems, and suggested that the option to “slow down or stop if necessary” should be considered to ensure controlled development.
Anthropic also noted that implementing such a global slowdown would require the joint action of many countries and major AI laboratories, making it a highly complex process.
Calls to halt AI development have been raised before, with thousands of technology experts demanding a similar moratorium in 2023. However, these calls did not result in a concrete pause.
Şayəstə