瞭望 | 盯紧AI失控风险

Core Viewpoint - The article emphasizes the urgent need to establish a resilient and inclusive intelligent society amidst the explosive growth of computing power and the inherent risks associated with AI, particularly the potential for AI to become uncontrollable [1][2]. Group 1: AI Control Risks - Experts, including Geoffrey Hinton, estimate the probability of AI becoming completely uncontrollable to be between 10% and 20% [2]. - The rapid evolution of AI systems, driven by intense competition among companies and nations, often lacks adequate consideration of potential consequences [2]. - There is a consensus among many professionals that the risk of AI losing control is a real concern, necessitating serious attention [2]. Group 2: Regulatory Challenges - The article identifies three main challenges contributing to the risk of AI losing control: lagging regulatory mechanisms, deficits in collaborative governance, and insufficient safety measures [3]. - Regulatory policies struggle to keep pace with the rapid technological advancements, as seen with the swift release of competing AI models following OpenAI's GPT-4 [3]. - The lack of international consensus on AI governance, highlighted by the refusal of some countries to sign collaborative agreements, exacerbates the regulatory challenges [3][4]. Group 3: Safety and Governance Improvements - Experts advocate for a shift towards agile governance that supports the healthy and sustainable development of AI [6]. - Recommendations include updating governance frameworks, enhancing communication between regulators and stakeholders, and adopting flexible regulatory measures [6][7]. - There is a call for improved risk assessment and management mechanisms for large AI models, as well as clearer definitions of rights and responsibilities for AI developers and users [7][8]. Group 4: Global Collaboration - Addressing the risks of AI control requires global cooperation, yet there is currently a lack of effective communication among leading AI companies [8]. - Strengthening bilateral dialogues, particularly between the US and China, and implementing existing international agreements on AI governance are essential steps [8].