据美联社报道，在竞相开发更强大人工智能的同时，科技公司正越来越多地向哲学家和宗教人士寻求建议，希望帮助AI分辨是非、作出符合道德的回应。但一些学者和宗教人士提醒，不同信仰的价值观可能相互冲突，将宗教思想汇总成一套AI道德准则，可能扭曲各个传统的原意。
信理部部长维克托·曼努埃尔·费尔南德斯枢机与神学家安娜·罗兰兹、Anthropic联合创始人克里斯托弗·奥拉进行交谈
Anthropic联合创始人克里斯·奥拉（Chris Olah）今年早些时候在梵蒂冈一场有关AI的活动上表示，这些问题无法由实验室独自回答，宗教传统已经思考了数千年。
0penAI首席执行官山姆·奥特曼
据报道，由于科技公司通常不隶属于某个特定宗教，它们往往希望综合不同传统中的道德观念，为模型提供指导。OpenAI首席执行官萨姆·奥尔特曼去年曾表示，他认为ChatGPT应当反映人类道德观的某种“加权平均”。
然而，这种思路引发了争议。密歇根大学人类学家韦布·基恩（Webb Keane）认为，将宗教视为一套可以单独提取的规则或原则，会忽略信众在日常生活中的实践，以及教义、仪式和人际关系的重要性。
在基恩看来，如果科技公司假定每种宗教都能提供一组可提取的道德原则，再将其用于训练AI，就可能优先呈现某些传统，并扭曲其他传统，尤其是那些并不主要将自身理解为伦理体系的信仰。他指出，将宗教当作“规则手册”的理解，与重视经典的宗教密切相关，尤其接近亚伯拉罕诸教的传统。
参与相关交流的人士也肯定了科技公司的努力。Anthropic在今年5月的一篇博客中表示，公司已经与来自15个以上宗教及跨文化群体的学者、神职人员、哲学家和伦理学家会面，并希望继续扩大参与范围。不过，公司没有逐一列明这些传统。
圣母大学宗教哲学家梅根·沙利文（Meghan Sullivan）今年参加了Anthropic组织的一场会议，与会者包括不同教派的基督徒。她表示，相比自己与其他科技公司的交流，Anthropic对宗教问题的认真探讨令她印象深刻。
据沙利文介绍，会议首先由公司用通俗语言解释模型，随后由宗教人士提供反馈、展开讨论。议题既涉及“人类是否拥有灵魂”，也包括“小学三年级学生是否应当使用聊天机器人学习读写”等具体问题。
美联社报道称，Meta、Google和亚马逊的代表也曾与梵蒂冈会面，微软同样积极开展与宗教界的交流。
争议还涉及科技公司如何理解“宗教”本身。宗教学者、《世界宗教的发明》作者增泽知子（Tomoko Masuzawa）指出，将各种宗教划分为边界清晰、可以相互比较的独立体系，是近代形成的观念。她以佛教和印度教为例称，在欧洲学者到来之前，当地实践者并未按照后来形成的明确分类理解自身传统。
一些学者因此担心，科技公司即便邀请了不同信仰的代表，也可能沿用既有的西方分类方式，忽略不同文化理解信仰的差异。
Anthropic在上述博客中表示，其目标并非让模型遵循某一种传统的世界观。公司希望Claude能够以同等深入、严谨的方式理解宗教、世俗和政治等不同观点，并从这些交流中学习人们长期以来对良好品格如何形成的思考。
不过，Meta独立监督委员会成员、研究天主教社会思想与技术的法律专家保罗·卡罗扎（Paolo Carozza）担心，在汇集各种思想的过程中，每种传统独有的特征可能被抹平，最终形成趋向折中、甚至偏重西方宗教的结果。
卡罗扎表示，成熟的传统能够通过与外界互动吸收新知识，并逐步发展；AI则可能把不同思想直接混合，最终无法真正理解其中任何一种。
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
According to the Associated Press, while competing to develop stronger artificial intelligence, technology companies are increasingly seeking advice from philosophers and religious figures to help AI to distinguish between right and wrong and to respond ethically. However, some scholars and religious figures have cautioned that the values of different faiths may conflict with each other and that the aggregation of religious ideas into an AI set of ethical norms may distort the original intent of the traditions.
Victor Manuel Fernández, Minister of Faith, spoke with Anna Rolands, theologian, and Christopher Orra, co-founder of Anthropic.
The co-founder of Anthropic, Chris Olah, said earlier this year at a Vatican event about AI that these questions cannot be answered by the lab alone and that religious traditions have been thinking for thousands of years.
Sam Otmann, Chief Executive Officer, 0pen AI
It has been reported that as technology companies do not usually belong to a particular religion, they often wish to integrate moral values from different traditions to guide models. CEO Open AI Sam Altmann stated last year that he believed that ChatGPT should reflect some kind of “weighted average” of human ethics.
However, that approach was controversial. Webb Keane, an anthropologist at the University of Michigan, believes that a view of religion as a set of rules or principles that can be extracted individually ignores the practice of believers in daily life and the importance of teaching, rituals and interpersonal relations.
In Keane ' s view, if technology companies assume that each religion provides a set of ethical principles that can be extracted and then used to train AI, they may prioritize some traditions and distort others, especially those that do not primarily interpret themselves as ethical systems. He pointed out that the understanding of religion as a “manual of rules” was closely linked to the importance attached to classic religions, in particular the tradition of Abrahamism.
Those involved in the exchange also acknowledged the efforts of the technology companies. In a blog in May this year, Anthropic stated that the company had met with scholars, clerics, philosophers and ethicalists from more than 15 religious and intercultural communities and hoped to continue to expand its participation. However, the company did not list these traditions individually.
Maghan Sullivan, the religious philosopher of the University of Notre Dame, participated in a conference organized by Anthropic this year, which included Christians of different denominations. She stated that, compared to her interaction with other technology companies, Anthropic had been impressed by the serious discussion of religious issues.
According to Sullivan, the meeting began with a model of interpretation in a popular language by companies, followed by feedback and discussion by religious people. The topic covered both “Does Man Have the Spirit” and specific issues such as “Does the third grade students need to learn to read and write using chat robots” in primary school?
Associated Press reports that representatives of Meta, Google and the Amazon have also met with the Vatican and that Microsoft is also active in communicating with religious communities.
The dispute also concerned how technology companies understood “religious” itself. Tomoko Masuzawa, a religious scholar and author of the Innovations of the World Religions, pointed out that the division of religions into separate systems with clear borders and comparability is a modern concept. She cited Buddhism and Hinduism as examples where local practitioners had not understood their traditions in accordance with the clear classification that had subsequently emerged before the arrival of European scholars.
Some scholars are therefore concerned that technology companies, even if they invite representatives of different faiths, may follow the established Western classification, ignoring differences in the understanding of faith among cultures.
In the above-mentioned blog, Anthropic states that its goal is not to have models follow a traditional view of the world. It is hoped that Claude will understand the different perspectives of religion, secularism and politics in an equally deep and rigorous manner and learn from these exchanges how good character has long been shaped.
However, Paolo Carozza, a member of the Meta Independent Monitoring Committee and a legal expert in Catholic social thought and technology, is concerned that, in bringing together ideas, the unique characteristics of each tradition may be diluted, leading to a tendency towards compromise and even a preference for Western religions.
Karoza expressed the view that mature traditions can absorb new knowledge through interaction with the outside world and evolve over time; AI, for its part, may directly mix different ideas and ultimately fail to truly understand either.
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.