麻省理工学院媒体实验室（MIT Media Lab）Pattie Maes 团队的一项研究揭示了“AI 依赖悖论”现象 1。研究发现，当参与者使用 AI 聊天机器人辅助识别假新闻时，其初期的准确率提高了 21% 1。然而，在实验进行到第四周且脱离 AI 辅助后，这些用户识别假新闻的能力比实验前下降了 15% 1。尽管客观能力出现下滑，仍有约四分之一的参与者自我感觉能力有所提升 1。
研究还发现，直接提供答案的 AI 更容易导致依赖，而通过苏格拉底式提问的 AI 更能帮助用户学会独立辨别真相 1。针对这一现象，Anku Rani 表示：“用户对那些‘神奇的’大语言模型感到兴奋，却忘记了它们只是预测序列中下一个‘词元’的统计模型。” 1 同时，Valdemar Danry 表示：“但这在很大程度上是速度与努力之间的权衡。” 1
A study conducted by Pattie Maes's team at the MIT Media Lab has uncovered an "AI dependency paradox" associated with using chatbots to identify fake news 1. When assisted by AI chatbots, participants initially experienced a 21% increase in accuracy in identifying fake news 1. However, by the fourth week, their ability to identify fake news without AI fell by 15% compared to their pre-experiment baseline 1. Additionally, about a quarter of the participants subjectively felt that their skills had improved 1.
The findings also show that AI providing direct answers is more likely to lead to dependency, while AI utilizing Socratic questioning better helps users learn to independently discern the truth 1. Valdemar Danry stated, "But it’s very much a trade-off between speed and effort" 1. Anku Rani stated, "Users get excited about these ‘magical’ LLMs but forget that they’re just statistical models that predict the next ‘token’ in a sequence" 1.
评论
还没有评论，欢迎留下第一条。