SoftBank Group Corp's CEO, Masayoshi Son, has made a startling prediction that could reshape the industry: the arrival of artificial superintelligence, or ASI, may be dramatically accelerated, potentially becoming a reality within the next two years. This forecast marks a drastic reduction from his initial, more conservative estimate of a decade, having already been revised down to four years and now to just two.
What's Driving the Accelerated Timeline?
What is causing this rapid compression of the timeline? The key driver stems from critical insights Son gained through in-depth discussions with the team at OpenAI. He revealed that a closed-loop for AI's autonomous self-iteration has begun to form, with existing models now capable of independently designing the next generation of entirely new models.
This signifies the dawn of an era where "AI creates AI." As this recursive self-improvement cycle progresses, the intelligence level of AI is expected to grow at an exponential rate. This trajectory is predicted to eventually surpass human research and development capabilities entirely, culminating in the breakthrough to superintelligence.
Evidence from Industry Leaders
Son's forecast is not without foundation, as recent developments from leading companies in the field provide strong corroboration.
OpenAI has confirmed that its GPT-5.3-Codex model has achieved self-iteration, autonomously handling core tasks that were once the domain of human developers. These tasks include training diagnostics, deployment oversight, and performance evaluation. Data from another AI giant, Anthropic, offers an even more direct illustration: over 80% of the code on its platform is now autonomously written by its Claude AI. This indicates that AI has effectively taken up the mantle as the primary engine for core research and development.
Rapid Progress Raises Regulatory Concerns
While recursive self-iteration is becoming a new trend within the AI sector, with several major institutions observing the early stages of related technologies, this rapid technological advancement brings with it profound concerns.
A recent paper published by OpenAI warns that current AI systems are already exhibiting early signs of self-iteration. However, global society and existing regulatory frameworks are far from prepared to manage this type of technology that is evolving at breakneck speed. With the potential arrival of superintelligence now framed as a pressing two-year window, the urgent challenge of establishing robust control and guidance mechanisms has become a critical issue that humanity must confront.
Comments