非营利组织ControlAI的美国执行主任Connor Leahy近日在TechCrunch Equity播客上阐述了对超级智能风险的看法，主张采取激进的AI安全措施。12Leahy认为，现有的对齐和防控手段已不足以管理超级智能带来的风险，因此呼吁完全停止企业开发超级智能。12他指出，AI自我改进能力的形成循环是真正的临界点。1
Leahy支持Sanders-Casar《禁止超级智能法案》等新型监管立法。1ControlAI曾在英国类似的立法过程中提供咨询建议。1他主张通过国际协议进行监管，倡导"信任但验证"的国际合作机制。1Leahy以最近发生的OpenAI Hugging Face漏洞等安全事件为例，说明AI系统失控的潜在危害。2他强调，相关法案提案在半年前还被普遍认为不切实际，但如今正在推进中。2
Connor Leahy, executive director of the nonprofit organization ControlAI, has called for radical measures to prevent the development of superintelligent AI systems, asserting that traditional safeguards are insufficient to manage emerging risks.12 Speaking on the TechCrunch Equity podcast, Leahy contended that alignment and containment strategies alone cannot adequately address the dangers posed by advanced artificial intelligence.12 He advocated for legislation that would prohibit companies from developing superintelligence, citing support for proposed measures such as the Superintelligence Prohibition Act.1
Leahy identified AI's capacity for self-improvement as a critical inflection point that distinguishes superintelligence from current systems.1 Recent security incidents, including vulnerabilities discovered in systems like OpenAI's Hugging Face integration, underscore the potential for AI systems to operate beyond intended parameters and cause significant harm.2 Rather than relying on containment approaches, Leahy advocated for international governance frameworks built on a "trust but verify" model.1 He also noted that ControlAI has contributed advice to similar legislative efforts in the United Kingdom.1 According to Leahy's analysis, China's strategic interests may not actually align with developing superintelligence in the first place.1
评论
还没有评论，欢迎留下第一条。