OECD經合組織：探索2030年可能的人工智能軌跡（英文版）.pdf

OECD經合組織：探索2030年可能的人工智能軌跡（英文版）。Artificial intelligence (AI) has advanced rapidly in recent years, with systems becoming increasingly capable. This paper presents expert- and evidenceinformed scenarios for how AI could progress by 2030. It considers recent trends in AI and key uncertainties for AI progress through 2030. Current evidence suggests that four different broad scenario classes are all plausible through to 2030: progress stalling, progress slowing, progress continuing, and progress accelerating. This suggests that AI progress by 2030 has a plausible range that includes both a plateau at approximately today’s level of capabilities and rapid improvement that leads to AI systems which broadly surpass human capabilities. This paper decomposes plausible AI capability progress in each scenario in line with the OECD’s beta AI capability indicators, exploring plausible capability trajectories for AI system’s abilities in language; social interaction; problem solving; creativity; metacognition and critical thinking; knowledge, learning and memory; vision; physical manipulation; and robotic intelligence.