AI安全与滥用风险
Search documents
加剧AI恐慌!微软高管:大多数白领工作将在“未来12-18个月内”完全自动化
硬AI· 2026-02-13 13:25
Core Viewpoint - Microsoft AI's chief executive warns that a majority of white-collar jobs may be automated within the next 12 to 18 months, a timeline that is significantly earlier than the expectations of the business community and policymakers [1][4]. Group 1: AI Impact on Employment - The report from Challenger indicates that in January 2023, 7,624 job losses were attributed to AI, accounting for 7% of total layoffs that month. By 2025, the total number of layoffs linked to AI is projected to reach 54,836 [1][4]. - Since the beginning of 2023, a total of 79,449 planned layoffs have been attributed to AI [1]. Group 2: Training AI with Human Labor - The startup Mercor has employed thousands of white-collar contractors, including professionals from fields such as medicine, law, finance, and engineering, to train AI systems that may eventually replace them. These contractors earn between $45 to $250 per hour [5]. - This model highlights the short-term demand for "data labeling and feedback labor" in the AI industry, while also raising concerns about long-term job stability and salary structures [5]. Group 3: Diverging Opinions on AI's Timeline - Not all analysts agree with the rapid timeline for job replacement. Morgan Stanley suggests that the impact of AI may take longer to manifest in economic data, with significant disruptions potentially occurring in the late 2020s or beyond [7]. Group 4: AI Risks Identified by Industry Leaders - Anthropic's CEO, Dario Amodei, outlines six major risks associated with AI, including large-scale unemployment, the potential for AI to possess state-level power, and the rise of terrorism threats due to advancements in biology [9]. - He expresses concern that AI could empower authoritarian regimes and highlights the risks posed by AI companies themselves, which control significant data and influence over users [9][10].
微软高管:大多数白领工作将在“未来12-18个月内”被AI完全自动化
华尔街见闻· 2026-02-13 11:09
Core Viewpoint - Microsoft AI Chief Mustafa Suleyman warns that most white-collar professional jobs may be automated within 12 to 18 months, a timeline much earlier than expected by the business community and policymakers [1][2][3]. Group 1: AI Impact on Employment - AI is predicted to achieve human-level performance in most professional tasks within 12 to 18 months, particularly in white-collar jobs that require computer use [3]. - The Challenger report indicates that 7,624 job cuts in January 2023 were attributed to AI, accounting for 7% of total layoffs that month. For the entire year of 2025, AI-related layoffs are projected to reach 54,836 [4]. - Since tracking began in 2023, a total of 79,449 planned layoffs have been linked to AI [1]. Group 2: Labor Market Dynamics - A notable case of labor replacement is emerging, where the startup Mercor has quietly hired thousands of white-collar contractors, including professionals from fields like medicine, law, finance, and engineering, to train AI systems that may eventually replace them [5]. - These contractors are typically paid between $45 and $250 per hour to review and revise model outputs, providing training support for companies like OpenAI and Anthropic [6]. Group 3: Diverging Opinions on AI's Timeline - Not all analysts agree with the rapid timeline for job replacement. Morgan Stanley suggests that the impact of AI may take longer to manifest in economic data, with significant effects possibly not appearing until the late 2020s or beyond [9]. - The speed of AI adoption may be faster than past technologies, but current economic data may not yet reflect this [9]. Group 4: AI Risks and Concerns - Anthropic CEO Dario Amodei outlines several risks associated with AI, including large-scale unemployment, the potential for AI to possess state-level power, and the rise of terrorism threats due to advancements in biology [10][11]. - Amodei expresses concern that AI could empower authoritarian regimes and highlights the risks posed by AI companies themselves, which control vast data centers and have significant influence over user behavior [11][12].