In March, Yoshua Bengio 收到 a share of the Turing Award, the highest accolade in computer science, for contributions to the development of deep learning—the technique that triggered a renaissance in artificial intelligence, leading to advances in self-driving cars, real-time speech translation,和 facial recognition.

现在,bengio表示深深的学习需要修复。他认为这不会实现其全部潜力,而且不会提供真正的人工智能革命,直到它能够超越模式识别和了解因果关系。换句话说,他说,深学习需求开始问 为什么 things happen.

55岁的老教授在蒙特利尔,谁体育浓密的花白头发和眉毛的大学,说深学习工作以及在理想化的情况,但不会接近复制人类的智慧,而不能推理因果关系。 “这是[因果]融入AI大的事情,” bengio说。 “目前的各种机器学习假设训练的AI系统将在同一类型的数据作为训练数据的应用。在现实生活中,经常不是这样的。”

机器学习系统包括深度学习是非常具体的,训练有素的特定任务,比如认识到音频图像猫,或语音命令。因为破土而出2012年左右,深度学习已经证明识别数据中的模式一个特别令人印象深刻的能力;它已经投入很多实际应用,从点滴的癌症医学扫描标志,在财务数据揭示欺诈行为。

但深学习是根本无视因果关系。不像一个真正的医生,深刻学习算法无法解释为什么一个特定的图像可能表明疾病。这意味着深学习时必须慎重,在紧急情况下使用。

理解因果将使现有的AI系统更智能,更高效。那了解到,掉落的东西使他们打破不需要折腾几十个花瓶的地板上,看看有什么发生在他们身上的机器人。

bengio说这个比喻延伸到自驾汽车。 “人类不需要度过事故的例子很多开车谨慎,”他说。他们可以想像的事故,“为了思想准备,如果没有实际发生。”

The question is how to give AI systems this ability.

学到更多

The WIRED Guide to Artificial Intelligence

At his research lab, Bengio is working on a version of deep learning capable of recognizing simple cause-and-effect relationships. He 和 colleagues recently posted a research paper outlining the approach. They used a dataset that maps causal relationships between real-world phenomena, such as smoking 和 lung cancer, in terms of probabilities. They also generated synthetic datasets of causal relationships.

报纸上的算法基本上形成一个假设哪些变量是因果关系,然后测试改变不同的变量如何适应理论。事实上,吸烟不仅与癌症有关,但实际上导致它,例如,应该仍然是显而易见的,即使癌症与其他因素,如医院探访相关。

A robot might eventually use this approach to form a hypothesis about what happens when it drops something,和 then confirm its hunch when it sees several things smash to the floor.

500 Internal Server Error

Internal Server Error

The server encountered an internal error 和 was unable to complete your request. Either the server is overloaded or there is an error in the application.

Geoffrey Hinton, of the University of Toronto 和 Google,和 亚·莱卡, who works at NYU 和 Facebook的.

深度学习使用人工神经网络在数学上近似的方式人类神经元和突触的学习,通过形成和加强联系。训练数据,诸如图像或音频,被馈送到一神经网络,其被逐步调整,直到它以正确的方式进行响应。深刻学习计划可以通过训练来识别在高精度照片的对象,提供它看到大量的训练图像,并给予足够的计算能力。

广告ment

但深学习算法不善于归纳,或采取什么他们已经从一个方面了解到并将其应用到另一个。他们还捕获是相关的,像雄鸡报晓,太阳上来,而不考虑这会导致其他现象。

因果关系长期以来一直研究在其他领域,和数学技术已经出现了近几十年来探索因果关系,帮助变革领域的研究,包括社会科学,经济学和流行病学。一小群研究人员正在努力因果关系和机器学习相结合。

继续阅读

The latest on artificial intelligence, from machine learning to computer vision 和 more

Judea Pearl, who won the Turing Award in 2011 for his work on causal reasoning, says he is impressed with Bengio’s ideas, although he has not studied them closely. A recent book co-authored by Pearl, The Book of Why: The New 科学 of Cause 和 Effect, makes the case that AI will be fundamentally limited without some sort of causal reasoning ability.

Cognitive science experiments also show that understanding cause and effect is fundamental to human development 和 intelligence, although it isn’t clear how humans form this knowledge.

bengio对因果关系的工作可能对回答这个问题的一小步,但它也反映了各地的深度学习更多的现实主义。即使该技术的应用已经成倍增加,越来越多的专家已经指出,在诸如关键领域的技术限制 language underst和ing.

在采访中,bengio也表示有些无奈与企业如何夸大ai和深度学习的能力。 “我认为这将是一个很好的事情,如果有一个在在商业界的修正,因为这是在炒作,”他说。

Others believe the focus on deep learning may be part of the problem. 加里马库斯, a professor emeritus at NYU 和 the author of a recent book that highlights the limits of deep learning, Rebooting AI: Building Artificial Intelligence We Can Trust, says Bengio’s interest in causal reasoning signals a welcome shift in thinking.

“太多深度学习的重点放在关系没有因果关系,并经常在当他们上都不太一样,他们进行了培训上的那些条件下进行测试亏损留下深刻的学习系统,”他说。

Marcus adds that the lesson from human experience is obvious. “When children ask ‘why?’ they are asking about causality,” he says. “When machines start asking 为什么, they will be a lot smarter.”


更多 Great WIRED Stories

将骑士 is a senior writer for WIRED, covering artificial intelligence. He was previously a senior editor at MIT Technology Review, where he wrote about fundamental advances in AI and China’s AI boom. Before that, he was an editor 和 writer at New Scientist. He studied anthropology 和 journalism in... 阅读更多