论文标题
人工智能的定义和测试
A Definition and a Test for Human-Level Artificial Intelligence
论文作者
论文摘要
尽管最近在许多特定应用领域进行了AI研究的进步，但我们不知道如何建立人类水平的人工智能（HLAI）。我们猜想，从他人对语言的经验中学习是将人类智慧与其他人区分开来的基本特征。人类可以用口头描述更新动作值函数，就像他们亲身体验状态，动作和相应的奖励序列。在本文中，我们根据各个代理商如何学习和提出了Hlai的定义和测试，对智力进行了分类。主要思想是，如果没有明确奖励的语言获取可以是Hlai的充分测试。
Despite recent advances of AI research in many application-specific domains, we do not know how to build a human-level artificial intelligence (HLAI). We conjecture that learning from others' experience with the language is the essential characteristic that distinguishes human intelligence from the rest. Humans can update the action-value function with the verbal description as if they experience states, actions, and corresponding rewards sequences firsthand. In this paper, we present a classification of intelligence according to how individual agents learn and propose a definition and a test for HLAI. The main idea is that language acquisition without explicit rewards can be a sufficient test for HLAI.
