SoftBank CEO Masayoshi Son believes artificial intelligence is rapidly approaching a threshold of “superintelligence,” with the next generation of AI models being designed by AI itself. This assertion comes amid growing concerns from AI research labs, like Anthropic, about the potential implications of accelerated AI development and the need for greater oversight.
During an interview, Son shared insights from his discussions with OpenAI CEO Sam Altman and the firm’s engineers, who revealed that an existing AI model is actively involved in designing future iterations. “So that’s going to happen to all the other major models,” Son stated, suggesting that human ingenuity alone may soon be outpaced in the development cycle of cutting-edge AI.
He elaborated on this concept, explaining that as models begin to generate their successors, the resulting intelligence will grow exponentially. “Once that happens, [the] model generates [the] next model… and it’s going to be exponentially smarter than all of us. That’s a superintelligence,” Son told CNBC.
While OpenAI has declined to comment on unreleased models, the company has previously highlighted instances where AI has played a role in its development process. In February, OpenAI noted that its GPT-5.3-Codex model was “instrumental in creating itself.” The team behind Codex, OpenAI’s suite of coding tools, utilized early versions of the model for tasks such as debugging its own training, managing deployment, and diagnosing test results. This demonstrates a nascent form of AI-assisted development that Son believes will escalate.
Son’s vision of “artificial superintelligence” (ASI) aligns with his past predictions. He previously described ASI as AI that could be “10,000 times smarter than humans” and projected its arrival within a decade. However, in his recent conversation, Son suggested that his earlier timeline was a conservative estimate to manage public perception. He now anticipates ASI arriving within the next two years.
The SoftBank CEO revealed his own extensive engagement with AI, using ChatGPT for two to three hours daily, finding it superior to his own knowledge across “most subjects.” He predicts that within a few years, AI will surpass human intelligence in approximately 70% to 80% of domains, potentially exhibiting intelligence levels “10 times smarter than average people” in those areas.
Son has been a vocal proponent of AI’s transformative potential for years, strategically positioning SoftBank at the forefront of the AI revolution. This strategy includes its ownership of chip designer Arm, a significant stake in OpenAI, and investments in emerging fields like robotics and autonomous driving. He has repeatedly emphasized the magnitude of the AI boom, comparing it to being “50 times bigger than the dot-com revolution in the 2000s.”
The intensifying discussion around advanced AI systems and their potential risks was further amplified by a recent blog post from Anthropic. The post focused on “recursive self-improvement” (RSI), a scenario where AI systems can autonomously design and develop their successors. While acknowledging potential benefits, Anthropic cautioned that RSI could escalate the “risks of humans losing control over AI systems.” The company advocated for a coordinated effort among AI labs to slow down the development of such advanced technologies, suggesting that such a slowdown “would likely be a good thing.”
It remains unclear whether Son’s comments specifically referred to RSI. However, an OpenAI research paper from June indicated “early signs” of RSI in current AI systems. The paper warned of increased competitive pressures among developers and nations, leading to governance challenges that existing institutions are ill-equipped to handle. It stressed the need for societies to develop mechanisms to guide AI development and ensure it aligns with human interests as RSI emerges.
Original article, Author: Tobias. If you wish to reprint this article, please indicate the source:https://aicnbc.com/22532.html
