虽然科幻电影描绘人工智能已经到机器能够独立思考的程度，但在现实生活中，受限于硬件设备的处理能力和编程逻辑的复杂性，我们身边的的人工智能仍然显得比较幼稚和容易理解——毕竟人类创造了所谓的人工智能，它们不可能超乎人类所能理解的范畴发展。
但谷歌的“深度学习（deep learning）”系统却颠覆了这一常识，谷歌的工程师都表示这套原本只是用来做实验的决策计算机系统表现超乎想象，它通过照片识别事物的能力早已超乎了谷歌工程师的预料，谷歌工程师甚至已经不知道计算机究竟在“想”些什么了。
识别系统的基础
谷歌软件工程师Quac V. Le在上周五旧金山的机器学习大会（Machine Learning Conference）上谈到了这套深度学习系统。这是一套包含了大量服务器群，能够收集并自动对数据作归类的系统，谷歌打算用它来深度研究AI技术。在谷歌手中，此系统的服务应用包括了Android的语音控制搜索、图片识别和谷歌翻译。
深度学习系统曾在去年6月份引起过不少讨论，当时纽约时报刊文称谷歌的DistBelief技术（一个采用普通服务器的深度学习并行计算平台）在获取数百万YouTube视频数据后，能够精准地识别出这些视频的关键元素：猫。未来，这套系统或是能够准确识别谷歌街景照片中门牌号码、网站中人脸图片等的技术依托。
深度学习技术理论上也是分层结构。其神经网络的最底层可检测到图片像素在色彩上的变化，上层随后可了解图片中出现特定事物的边缘部分。位于再上层的几个连续的分析层可通过系统的不同分支学会人脸、摇椅、计算机等各种类型事物的检测方法。
它们真的在“独立思考”
Quoc V. Le说，令他最为震惊的事情是，深度学习系统能够轻易地学习总结出类似碎纸机等物体的特性，这些甚至是普通人类难以轻易做到的。“怎样在系统设计中让软件能够具备识别碎纸机的能力，这是相当复杂的。我在这方面花了很多的时间，但就是难以完成。”
实际上Quoc也曾给身边的好多朋友看了碎纸机的照片，但在随后的识别过程中，具有高等智慧的人类却遭遇了麻烦。而谷歌的深度学习计算机系统则在这方面具有极高的识别成功率，可关键问题是，Quoc自己也不知道他所写的程序是如何做到这一点的。
也就是说，谷歌的工程师们已经无法解释这套系统识别事物的方法和逻辑，它们更像是脱离了其创建者的控制在独立思考，这种复杂的认知方式更是令人不可思议。虽然这种层面的“独立思考”范围还非常有限，但在实际应用中却真实有效，能够解决实际问题。谷歌负责AI研究的主管Peter Norvig认为这种能够实现大量数据统计的模型对于解决如语音识别与理解一类的复杂问题具有非常积极的意义。
结论
Quoc说，对谷歌而言，深度学习系统能够解决人类所不能解决的问题，自然也就是节约人力成本的好东西。将其更多的潜力挖掘出来，总好过雇佣一批每年拿着无数酬劳的高级专家。“机器学习非常复杂，我们需要花大量的时间在数据处理和特性更新上。甚至为了解决一个独立的问题，我们就需要聘请这一领域的专家。以后我们期望能够跳脱这样的模式，我们没法解决的问题，就让机器去完成。”
而且谷歌实际也在开发其他类似的决策选择系统，如Borg与Omega，这些系统在分配工作负荷时，行为方式也更像是活物。将来，机器的“独立思考”或将真正成为可能。至少现在，我们让计算机与人类达成了这样的协作关系。
If this doesn't terrify you... Google's computers OUTWIT their humans
'Deep learning' clusters crack coding problems their top engineers can't
Analysis Google no longer understands how its "deep learning" decision-making computer systems have made themselves so good at recognizing things in photos.
This means the internet giant may need fewer experts in future as it can instead rely on its semi-autonomous, semi-smart machines to solve problems all on their own.
The claims were made at the Machine Learning Conference in San Francisco on Friday by Google software engineer Quoc V. Le in a talk in which he outlined some of the ways the content-slurper is putting "deep learning" systems to work. (You find out more about machine learning, a computer science research topic, here [PDF].)
"Deep learning" involves large clusters of computers ingesting and automatically classifying data, such as things in pictures. Google uses the technology for services such as Android's voice-controlled search, image recognition, and Google translate.
The ad-slinger's deep learning experiments caused a stir in June 2012 when a front-pageNew York Times article revealed that after Google fed its "DistBelief" technology with millions of YouTube videos, the software had learned to recognize the key features of cats.
A feline detector may sound trivial, but it's the sort of digital brain-power needed to identify house numbers for Street View photos, individual faces on websites, or, say, <SKYNET DISCLAIMER> if Google ever needs to identify rebel human forces creeping through the smoking ruins of a bombed-out Silicon Valley </SKYNET DISCLAIMER>.
Google's deep-learning tech works in a hierarchical way, so the bottom-most layer of the neural network can detect changes in color in an image's pixels, and then the layer above may be able to use that to recognize certain types of edges. After adding successive analysis layers, different branches of the system can develop detection methods for faces, rocking chairs, computers, and so on.
What stunned Quoc V. Le is that the software has learned to pick out features in things like paper shredders that people can't easily spot – you've seen one shredder, you've seen them all, practically. But not so for Google's monster.
Learning "how to engineer features to recognize that that's a shredder – that's very complicated," he explained. "I spent a lot of thoughts on it and couldn't do it."
It started with a GIF: Image recognition paves way for greater things
Many of Quoc's pals had trouble identifying paper shredders when he showed them pictures of the machines, he said. The computer system has a greater success rate, and he isn't quite sure how he could write program to do this.
At this point in the presentation another Googler who was sitting next to our humble El Reghack burst out laughing, gasping: "Wow."
"We had to rely on data to engineer the features for us, rather than engineer the features ourselves," Quoc explained.
This means that for some things, Google researchers can no longer explain exactly how the system has learned to spot certain objects, because the programming appears to thinkindependently from its creators, and its complex cognitive processes are inscrutable. This "thinking" is within an extremely narrow remit, but it is demonstrably effective and independently verifiable.
Google doesn't expect its deep-learning systems to ever evolve into a full-blown emergent artificial intelligence, though. "[AI] just happens on its own? I'm too practical – we have to make it happen," the company's research chief Alfred Spector told us earlier this year.
Google's AI chief Peter Norvig believes the kinds of statistical data-heavy models used by Google represent the world's best hope to crack tough problems such as reliable speech recognition and understanding – a contentious opinion, and one that clashes with Noam Chomsky's view.
Deep learning is attractive to Google because it can solve problems the company's own researchers can't, and it can let the company hire fewer inefficient meatsacks human experts. And Google is known for hiring the best of the best.
By ceding advanced capabilities to its machines, Google can save on human headcount, better grow its systems to deal with a data deluge, and develop capabilities that have – so far – befuddled engineers.
The advertising giant has pioneered a similar approach of delegating certain decisions and decision-making selection systems with its Borg and Omega cluster managers, which seem to behave like "living things" in how they allocate workloads.
Given Google's ambition to "organize the world's information", the fewer people it needs to employ, the better. By developing these "deep learning" systems Google needs to employ fewer human experts, Quoc, said.
"Machine learning can be difficult because it turns out that even though in theory you could use logistic regression and so on, but in practice what happens is we spend a lot of time on data processing inventing features and so on. For every single problem we have to hire domain experts," he added.
"We want to move beyond that ... there are certainly problems we can't engineer features of and we want machines to do that."
By working hard to give its machines greater capabilities, and local, limited intelligence, Google can crack classification problems that its human experts can't solve. Skynet? No. Rise of the savant-like machines? Yes. But for now the relationship is, thankfully, cooperative. ®
扫一扫
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
