评论
打印
Programming arguments
机器也能“舌战群儒”！
For two centuries, the famous University of Cambridge debating society has hosted many prominent figures, from world leaders to scientists to comedians.
两个世纪以来，著名的剑桥大学辩论协会邀请了无数杰出人士前来辩论，如世界领导人、科学家、喜剧演员等等。
On Nov 21, the Cambridge Union hosted its first-ever non-human guest. Designed by IBM, Project Debater is a machine that uses Artificial Intelligence (AI) to form a spoken argument. It can even listen and answer to opposing statements, much like a real person.
11月21日，剑桥辩论学会迎来了首位非人类嘉宾。由美国国际商用机器公司（以下简称IBM）设计的“辩论家计划”是个能够使用人工智能进行辩论的机器人。它甚至能够聆听并提出反驳，就像真人一样。
Speaking with a voice similar to Siri, Project Debater uses AI to search journals, newspapers and other information sources. It then filters out useless information, picks the strongest arguments, and arranges them into a persuasive argument.
“辩论家计划”的声音与苹果智能语音助手类似，它用人工智能搜索期刊、报纸以及其他信息来源。然后，它会过滤掉无用信息，选出最有力的论据，并将其组成一段有说服力的论点。
In the Cambridge test debate, it argued the topic: AI will ultimately do more harm than good. According to IBM, it is the first machine that can debate complicated topics with humans.
在剑桥的这场测试辩论赛中，该机器人就辩题“人工智能是否最终弊大于利”进行辩论。IBM表示，“辩论家计划”是首个能够就如此复杂的话题与人类进行辩论的机器人。
Before the debate, more than 1,100 arguments for the positives and negatives of AI were submitted to the IBM website. The machine then analyzed these sources and formed the basis of its arguments in minutes. Project Debater then showed off its AI by arguing for both sides of the debate.
辩论开始前，IBM官网上收到了1100多条关于人工智能利弊的论据。这款机器人随后对这些来源进行了分析，并在数分钟内形成了其论点的基础。然后，“辩论家计划”通过为正反两方辩论展示了其人工智能的技术。
Over a four-minute speech, it argued first that, "AI can only make decisions that it has been programmed to solve, while humans can be programmed for all scenarios."
在一段四分钟的演讲中，机器人首先提出了这个论点：“人工智能只能解决程序中设置好的问题，但人类却能应对各种情况。”
In support of AI, it then argued that AI would create new jobs and "bring a lot more efficiency" to the workplace.
而在支持人工智能的论点中，它又表示，人工智能可以创造新的就业机会，并为职场“大幅提高效率”。
This isn't the first test for Project Debater. In June 2018, the machine went back-and-forth for 20 minutes against champion debater Harish Natarajan. The topic was whether the government should subsidize US preschools.
这并非“辩论家计划”首次接受测试。2018年6月，该机器人与辩论冠军哈里什·纳塔拉扬进行了一场20分钟的唇枪舌战。当时的辩题是美国学前教育是否应该得到政府补贴。
According to New Scientist, on both occasions, Project Debater repeated points and didn't always sufficiently support them.
《新科学家》杂志认为，在这两次辩论中，“辩论家计划”的论点都有重复，其论据也并非总是充分有力。
Speaking to Cambridge Independent, Noam Slonim, the project's lead engineer, said, "The AI is not perfect, but it's going in the right direction." The team hopes to improve the AI by researching why humans find certain arguments more persuasive than others. They also hope the test might shine a light on fake or one-sided narratives and provide well-informed viewpoints for society, governments and private companies.
该项目的首席工程师诺姆·斯洛尼姆在接受《剑桥独立报》采访时表示：“人工智能并不完美，但它的方向是正确的。”为改进人工智能，团队对人类为何能找出更有说服力的论点进行了研究。他们也希望，这样的测试能够显露虚假陈述或片面之词的问题，并为社会、政府以及私人企业提供及时全面的观点。
