Minimal Turing

People believe other people have minds, but they also perceive minds in other entities, from animals to God to robots. Previous work on mind perception has relied mostly on participant ratings of pre-selected attributes. We instead use a task, which we call the Minimal Turing Test, in which participants must convince a judge that they are human, rather than an artificial intelligence, using a single word of their choosing. By embedding the production data from a large-scale Minimal Turing Test in a high-dimensional semantic vector space, and constructing an ordering over pairwise evaluations from judges, we provide additional evidence for the agency and experience dimensions discovered in previous work. We identify rich semantic structure within and across these two dimensions, and obtain quantitative measures of the importance that people place on different attributes.

人们普遍认为他人拥有心智，然而，他们亦能在众多实体中感知到心智的存在，这些实体从动物到神明，乃至机器人。以往关于心智感知的研究主要依赖于参与者对预选属性的评分。我们另辟蹊径，采用一种被称为最小图灵测试的任务，在该任务中，参与者需通过选择一词来说服评判者其为人而非人工智能。通过将大规模最小图灵测试的生产数据嵌入高维语义向量空间，并构建评判者间评价的排序，我们为先前研究中发现的代理性和经验维度提供了额外的证据。我们在这两个维度内及其之间识别出丰富的语义结构，并获得了人们对不同属性所赋予重要性的量化指标。