《麻省理工学院科技评论》报导，经典的人工智能是一个铸造知性模型来认识世界的架构，这个架构继而用来推理和推敲世界的细节。其二，经典人工智能是以规则导向的编程所铸造的。然而，在当下火红的深度学习时代，人们却想避开具规则导向的编程，只用人工神经网络来建构所有的东西，完全不想要再触碰经典的编程。吊诡的是，许多我们惯常面对的难题都是靠传统编程程序来解决的，譬如通过谷歌地图来寻早路径。
纽约大学教授暨前优步人工智能实验室总监Gary Marcus指出，我们其实同时需要这两种开发人工智能的技术。机器学习擅于从数据中学习，但它却非常拙于如电脑程序般进行有效的抽象性概念描述。经典的人工智能十分精于以抽象概念进行描述，但它却需要倚赖人手编译，而这个世界有不胜枚举的知识，不是纯碎以人手编程所能概括的。由是观之，我们需要一种崭新的综合方式来结合这两种开发人工智能的手法。
目前，人们正尝试以“千篇一律”的人工智能技术来解决本质上完全不同的问题。例如说，理解句子与识别物体根本不一样。但是人们尝试使用深度学习来处理这两种截然不同的任务。从认知的角度来看，这两者在本质上是完全不同的问题，因此Gary Marcus教授对深度学习社群轻视规则性导向的人工智能开发技术感到不知所措。他表示不理解为什么大家会期望以一个银色子弹式的套路来解决所有问题？他认为这样的态度是不现实的，也没有显示出专家们对AI挑战的深刻理解。
[...]
他一步阐述，通用人工智能就是要让人工智能系统能够即时思考，并自行解决新问题。相对于围棋，那是一种2000年来几乎没有改变过规则的游戏。
他提醒道，我们现在还远远没有做到能即时自行思考、分析问题和解决问题的通用人工智能。 他说：“AlphaGo可以在19x19的正方形棋盘上”得心应手“的下棋，但如果棋盘被换成长方形的设置，AlphaGo需要被重新培训才能下棋。”
他也指出，如果你采用一般的深度学习系统，只要在光线充足和大象的纹理清楚的情况下，人工智能系统就可以识别大象。但是，如果你将大象摆成影像轮廓，人工智能系统很可能就不再认识它了。
[Excerpt] Classical AI actually is a framework for building cognitive models of the world that you can then make inferences over. The second thing is, classical AI is perfectly comfortable with rules. It’s a strange sociology right now in deep learning where people want to avoid rules. They want to do everything with neural networks, and do nothing with anything that looks like classical programming. But there are problems that are routinely solved this way that nobody pays attention to, like making your route on Google maps.
We actually need both approaches. The machine-learning stuff is pretty good at learning from data, but it’s very poor at representing the kind of abstraction that computer programs represent. Classical AI is pretty good at abstraction, but it all has to be hand-coded, and there is too much knowledge in the world to manually input everything. So it seems evident that what we want is some kind of synthesis that blends these approaches.
Right now people are trying to use kind of one-size-fits-all technologies to tackle things that are really fundamentally different. Understanding a sentence is fundamentally different from recognizing an object. But people are trying to use deep learning to do both. These are qualitatively different problems from a cognitive perspective, and I’m just sort of flabbergasted at how little appreciation the deep-learning community in general has for that. Why expect that one silver bullet is going to work for all of that? It’s not realistic, and it doesn’t reveal a sophisticated understanding of what the challenge of AI even is.
[...]
General AI is about having AI be able to think on the fly and resolve new problems on its own. This is as opposed to, let’s say, Go, where the problem hasn’t changed in 2,000 years.
But you know, we’re pretty far from that right now. AlphaGo can play very well on a 19x19 board but actually has to be retrained to play on a rectangular board. Or you take your average deep-learning system, and it can recognize an elephant as long as the elephant is well lit and you can see the texture of the elephant. But if you put the elephant in silhouette, it might well not be able to recognize it anymore.[Excerpt]
Reference source/资料出处:
https://www.technologyreview.com/s/614443/we-cant-trust-ai-systems-built-on-deep-learning-alone/
Gary Marcus教授接受麻省理工学院人工智能科研人员Lex Fridman的访问：
Posts
0 Comments