25.婴儿人工智能（1124）

Passage Twenty-Five

Baby AI

婴儿人工智能

【1】An Al trained on the experiences of an infant learned a handful(few) of basic words.For decades linguists have argue over how children learn language.

经过婴儿经验训练的人工智能学会了一些少量但基本的单词。数十年来，语言学家们一直在争论儿童是如何学习语言的。

【2】Some think that babies are born as“blank slates”who pick up language simply from experience-hearing, seeing and playing with the world. Others argue that experience is not enough and that babies' brains must be hardwired to make acquiring language easy.

一些人认为，婴儿生来就是“白板”，他们只是通过经验，也就是听、看和与世界玩耍来学习语言。另一些人则认为，光有经验是不够的，婴儿的大脑必须天生就有某些机制，使得学习语言变得更容易。

【3】AI models such as GPT-4 have done little to settle(solve) the debate(arguement). The way these models learn language-by trawling through reams of text data from millions of web pages-is vastly different to the experiences of babbling babies.

GPT-4等人工智能模型在解决这一争论方面收效甚微。这些模型学习语言的方式是从数百万个网页中搜索大量文本数据，与咿呀学语的婴儿的经历迥然不同。

【4】A team of scientists at New York University examined the question by training an AI model on the experiences of a single infant. Between the ages of six and 25 months, a toddler called Sam wore a head-mounted camera for an hour a week-around 1of his waking hours.The camera recorded everything he saw and heard while he played with toys,enjoyed days at the park and interacted with his pet cats.

纽约大学的一个科学家团队通过对一位婴儿的经验进行AI模型训练，试图解答这个问题。在蹒跚学步的小孩山姆6到25个月大的时候，他每周都会在醒着的时候戴一个小时（大约是他醒着的时间的1%）的头戴式摄像机，摄像机会记录下他玩玩具、在公园玩耍以及与宠物猫互动时的所见所闻。

【5】The recordings and transcribed audio were fed into an Al,which was set up to know that images and words that appeared at the same time were related, but was otherwise left to make sense of the mess of colours and speech that Sam experienced.

研究人员将这些录音和转录的音频输入一个人工智能模型，该模型被设置为能够识别同时出现的图像和单词之间的关系，但在其他方面，它可以自由理解山姆所经历的一片混乱的颜色和言语。

【6】Despite the limited(finite) training data, the AI was able to pick out objects and learn the matching(suited) words. The researchers tested the model by asking it to identify objects that Sam had seen before, such as a chair from his home or one of his toy balls.

尽管训练数据有限，人工智能还是能够识别出物体并学习匹配的单词。研究人员通过让模型识别山姆以前见过的物体来对其进行测试，如他家的一把椅子或他的一个玩具球。

【7】Given a list of four options the model picked(selected) the correct word 62of the time, far above the chance level of 25%.To the researchers' surprise, the model could also identify chairs and balls that Sam had never seen.

给定一个包含四个选项的列表，该模型选择正确单词的概率达62%，远远高于25%的概率水平。研究人员惊讶地发现，该模型还能识别山姆从未见过的椅子和球。

【8】The AI learnt at least 40 different words, but it was far from matching Sam's vocabulary and language abilities(capacities) by the end of the experiment(test).

人工智能至少学会了40个不同的单词，但在实验结束时，它的词汇量和语言能力还远远达不到山姆的水平。

【9】The researchers, published recently in the journal Science, argue that, to match words to objects, learning from experience may well be enough.

研究人员最近在《科学》杂志上发表文章称，要将单词与物体匹配起来，从经验中学习可能就足够了。

【10】Sceptics,however, doubt that the AI would be able to learn abstract nouns or verbs, and question how similar the learning processes really are. The mystery of language acquisition lives on.

然而，怀疑论者怀疑人工智能是否能学习抽象名词或动词，并质疑学习过程的相似性。语言习得的谜题依然未解。