AI may not only support human development of technology but also participate in building the next generation of AI. This is a noteworthy assessment just made by Anthropic, one of the world's leading AI development enterprises, in a new report on the future of artificial intelligence.
Anthropic believes that the rapid progress of AI models could lead to a new stage, where artificial intelligence systems play an increasingly large role in designing, improving and developing their own successor versions.
This concept is called "Recursive Self-Improvement" (RSI). Simply put, it is the ability of an AI system to automatically design and develop a better new version without too much human intervention.
According to Anthropic, the world has not yet reached this stage. However, the necessary conditions for that scenario are gradually forming. Advanced AI models today have shown increasingly effective programming, software debugging, research and problem solving capabilities.
This could create a loop in which AI supports building stronger AI systems, thereby further promoting the development process.
Anthropic says that if this trend continues and comes with sufficient computing capacity, AI may move closer to self-improvement on a large scale. Some researchers call this prospect a "intelligence explosion", which is the time when AI systems can continuously upgrade their capabilities at speeds far beyond human research capabilities.
In an interview with Axios (a media company and digital news site in the US), co-founder of Anthropic - Jack Clark said that the pace of AI progress in the coming years may continue to accelerate instead of slowing down. He believes that this has the potential to bring great breakthroughs in science, medicine and many other fields.
However, Anthropic also emphasizes that the greater the benefits, the more noteworthy the challenges. The biggest concern lies in the ability to control. If AI can improve itself quickly and continuously, humans may have difficulty understanding how the system works as well as the changes taking place inside it.
According to this company, future AI systems may exhibit unpredictable behaviors, exploit safety vulnerabilities, or pursue goals that are not entirely consistent with human desires. Even without the intention to harm, those actions can still lead to unintended consequences.
Another challenge is the development speed. If AI progresses too quickly, management agencies and society may not be able to adapt in time. Therefore, Anthropic believes that AI companies need to strengthen international cooperation, and be ready to slow down or temporarily stop the development process when serious risk signs appear.
However, technology experts also recognize that the scenario of AI completely building its own successor version has not yet occurred. The development of advanced AI systems today still requires a large amount of data, computing infrastructure, specialized hardware and close supervision from humans.
