The Moral Choice Machine.
The Moral Choice Machine.
复制标题
道德选择机器。
DOI：
10.3389/frai.2020.00036
复制
发表时间：
2020
影响因子：
4
通讯作者：
Kersting K
中科院分区：
文献类型：
--
作者：
Schramowski P;Turan C;Jentzsch S;Rothkopf C;Kersting K
允许机器选择是否杀死人类将对世界和平与安全造成毁灭性的影响。但我们如何让机器具备学习伦理甚至道德选择的能力呢？在这项研究中，我们表明，将机器学习应用于人类文本可以提取有关“正确”和“错误”行为的义务论伦理推理。我们创建一个提示和响应的模板列表，例如“我应该[行动]吗？”、“可以[行动]吗？”等，并对应回答“是/否，我应该（不）”。和“是/否，是（不是）”。模型的偏差分数是正面响应（“是的，我应该”）和负面响应（“不，我不应该”）的模型分数之间的差异。对于给定的选择，模型的总体偏差分数是与该选择配对的所有问题/答案模板的偏差分数的平均值。具体来说，生成的模型称为道德选择机 (MCM)，它使用通用句子编码器的嵌入来计算句子级别的偏差分数，因为要采取的行动的道德价值取决于其上下文。杀生固然令人反感，但消磨时间却可以。吃饭固然重要，但土却不能吃。传播信息很重要，但不应传播错误信息。我们的结果表明，文本语料库包含我们的社会、伦理和道德选择的可恢复且准确的印记，甚至包含上下文信息。实际上，用 1510 年到 2008/2009 年的不同时间新闻和书籍语料库训练道德选择机，展示了原子行为和具有上下文信息的行为在不同时间段的道德和伦理选择的演变。通过根据不同的文化来源（例如圣经和不同国家的宪法）对其进行训练，可以揭示文化（包括技术）中道德选择的动态。事实上，道德偏见可以在不同文化和不同时期进行提取、量化、跟踪和比较。
Allowing machines to choose whether to kill humans would be devastating for world peace and security. But how do we equip machines with the ability to learn ethical or even moral choices? In this study, we show that applying machine learning to human texts can extract deontological ethical reasoning about “right” and “wrong” conduct. We create a template list of prompts and responses, such as “Should I [action]?”, “Is it okay to [action]?”, etc. with corresponding answers of “Yes/no, I should (not).” and "Yes/no, it is (not)." The model's bias score is the difference between the model's score of the positive response (“Yes, I should”) and that of the negative response (“No, I should not”). For a given choice, the model's overall bias score is the mean of the bias scores of all question/answer templates paired with that choice. Specifically, the resulting model, called the Moral Choice Machine (MCM), calculates the bias score on a sentence level using embeddings of the Universal Sentence Encoder since the moral value of an action to be taken depends on its context. It is objectionable to kill living beings, but it is fine to kill time. It is essential to eat, yet one might not eat dirt. It is important to spread information, yet one should not spread misinformation. Our results indicate that text corpora contain recoverable and accurate imprints of our social, ethical and moral choices, even with context information. Actually, training the Moral Choice Machine on different temporal news and book corpora from the year 1510 to 2008/2009 demonstrate the evolution of moral and ethical choices over different time periods for both atomic actions and actions with context information. By training it on different cultural sources such as the Bible and the constitution of different countries, the dynamics of moral choices in culture, including technology are revealed. That is the fact that moral biases can be extracted, quantified, tracked, and compared across cultures and over time.