AI Models Could Help Decode Animal Communication
Earth Species Project researchers have used machine learning to identify individual calls from little crows and determine which bird produced each call. The nonprofit is developing NatureLM, a model trained on human language, music, and animal sounds to identify patterns across different sounds. Researchers suggest that more advanced AI, including models such as Claude and ChatGPT, may eventually help translate animal communication, but true semantic translation remains theoretical. Researchers warn that models may over-rely on acoustic similarity, classifying calls that sound similar but have different meanings. They also note that many animals communicate through visual, tactile, chemical, and combined signals, so sound analysis alone may be insufficient.
The work demonstrates an existing application of AI to animal-call analysis while distinguishing it from the still-unproven goal of translating animal semantics.
(来源:麻省理工科技评论)
人类远不是唯一拥有交流天赋的物种。那些长着皮毛、羽毛或鱼鳍的动物朋友们,或许也有许多话想对我们说。破译动物之间的交流方式,不仅能让我们窥见其他物种的内心世界,也能帮助人类认识到,“动物是有感知能力的生命,它们会思考,也有自己的希望和梦想”。致力于开发犬类与人类语言交流技术的公司 Zoolingua 创始人康·斯洛博奇科夫(Con Slobodchikoff)说道。
要弄清动物究竟在表达什么,并非易事。为了破译动物的嚎叫、尖鸣和咔嗒声,研究人员必须将它们发出的声音与周围发生的事情联系起来。例如,当郊狼靠近时,草原犬鼠发出的警报叫声,与附近出现鹰时的叫声就有所不同。这是一项极其耗费精力的工作,对于深潜鲸类或生活在地下的鼹鼠等难以观察的动物来说,尤其如此。
#NatureLM#Earth Species Project#Claude#ChatGPT#Zoolingua#Tel Aviv University