[论文解读] Thousands of AI Authors on the Future of AI

Katja Grace, H. T. L. Stewart+5 more|arXiv (Cornell University)|Jan 5, 2024

Big Data and Business Intelligence被引用 36

一句话总结

一项对2,778名AI研究人员的大规模调查预测了AI进展的里程碑和风险，发现2023年的时间表相较于2022年更早，且对风险和影响的观点存在广泛差异。

ABSTRACT

In the largest survey of its kind, 2,778 researchers who had published in top-tier artificial intelligence (AI) venues gave predictions on the pace of AI progress and the nature and impacts of advanced AI systems The aggregate forecasts give at least a 50% chance of AI systems achieving several milestones by 2028, including autonomously constructing a payment processing site from scratch, creating a song indistinguishable from a new song by a popular musician, and autonomously downloading and fine-tuning a large language model. If science continues undisrupted, the chance of unaided machines outperforming humans in every possible task was estimated at 10% by 2027, and 50% by 2047. The latter estimate is 13 years earlier than that reached in a similar survey we conducted only one year earlier [Grace et al., 2022]. However, the chance of all human occupations becoming fully automatable was forecast to reach 10% by 2037, and 50% as late as 2116 (compared to 2164 in the 2022 survey). Most respondents expressed substantial uncertainty about the long-term value of AI progress: While 68.3% thought good outcomes from superhuman AI are more likely than bad, of these net optimists 48% gave at least a 5% chance of extremely bad outcomes such as human extinction, and 59% of net pessimists gave 5% or more to extremely good outcomes. Between 38% and 51% of respondents gave at least a 10% chance to advanced AI leading to outcomes as bad as human extinction. More than half suggested that "substantial" or "extreme" concern is warranted about six different AI-related scenarios, including misinformation, authoritarian control, and inequality. There was disagreement about whether faster or slower AI progress would be better for the future of humanity. However, there was broad agreement that research aimed at minimizing potential risks from AI systems ought to be prioritized more.

研究动机与目标

• 评估AI研究人员对AI在各任务、职业以及一般人类水平表现方面进展速度的预测。
• 考察HLMI和FAOL时间表的预期及其框架效应。
• 探讨AI研究人员对风险、社会影响以及AI安全研究优先事项的看法。
• 将2023年的预测与2022及之前的调查进行比较，以识别预期和框架效应的变化。

提出的方法

• 调查了在前一年（NeurIPS、ICML、ICLR、AAAI、IJCAI、JMLR）顶级AI会议上发表论文的2,778名AI研究人员。
• 使用李克特量表、概率估计，以及以固定年份和固定概率框架的年度响应来获得预测。
• 对单个受访者数据拟合伽马分布，以获得里程碑以及HLMI/FAOL的聚合分布。
• 实施框架控制和随机化，以减轻问题框架效应并评估结果的鲁棒性。
• 将2023年的结果与2022年和2016年的调查进行比较，以绘制预测时间线和态度的变化。

Figure 1: Most milestones are predicted to have better than even odds of happening within the next ten years, though with a wide range of plausible dates. The figure shows aggregate distributions over when selected milestones are expected, including 39 tasks, four occupations, and two measures of ge

实验结果

研究问题

• RQ1 AI在一年的时间内并且达到指定概率阈值后，何时能实现一组39项任务及其他里程碑的可行性？
• RQ2 HLMI和FAOL的可行性多久会实现，以及任务与职业框架对这些预测有何影响？
• RQ3 AI研究人员对风险、社会影响以及优先开展安全重点研究的可取性有何看法？
• RQ4 2022年至2023年的预测变化由何原因造成，以及人口统计因素如何影响预测？
• RQ5 研究人员是否认为存在智能爆炸风险，以及在HLMI之后他们如何评估进展的潜在提升？

主要发现

• 在接下来的十年内，大多数这39项任务实现可行的概率超过50%，其中包括自动搭建支付站点和生成与知名艺人作品难以区分的歌曲。
• HLMI的50%里程碑日期从2060（2022）移动到2047（2023），而FAOL的50%日期从2164（2022）移动到2116（2023）。
• 框架效应：固定年份与固定概率框架产生不同的50%时间线，固定概率框架预测的里程碑比固定年份框架更早。
• 大多数人（68.3%）认为超人类AI带来良好结果的可能性高于坏结果，但相当一部分的少数派对极端结果如人类灭绝赋予不容忽视的概率（某些回答高达10%或更多）。
• 超过一半的受访者主张将AI安全研究的优先级提高到高于当前水平，且自早期调查以来支持度有所上升。
• 在50%基础上，HLMI预计比FAOL更早到来约70年，这一差异在不同年份和框架下都存在。
• 受访者预计AI在2043年会具备某些特征（如行为上出现惊讶、具有类似人类的讨论能力）且可能性很高，尽管在某些特征上仍存在分歧，如采取行动以获得权力。
• 对AI驱动的错误信息、不平等以及专制滥用存在相当程度的担忧，若干情景被相当大的少数人评为值得高度关注或极端担忧。

Figure 2: Expected feasibility of many AI milestones moved substantially earlier in the course of one year (between 2022 and 2023). The milestones are sorted (within each scale-adjusted chart) by size of drop from 2022 forecast to 2023 forecast, with the largest change first. The year when the aggre