导读

人们经常问，你所说的"突破"到底是什么意思？ 这是一个合理的问题——我们精选了一些尚未得到广泛应用的产品，或者可能正处于商业化尖端的科技。我们真正在寻找的是一种技术，或者甚至是一系列技术，它们将对我们的生活产生深远的影响。

今年，人工智能领域一项名为"广域网"(GANs)的新技术正在给机器提供想象力; 3D打印来到了一个里程碑的节点;云智能技术让每一个人都能感受到AI的魅力；魔术一般的遗传占卜能从你出生就知道你未来的身体情况；新型量子计算机的出现将给化学家和生物学家的研究带来质的飞跃.

正文

3-D Metal Printing

3-D  金属打印

3D金属打印

While 3-D printing has been around for decades, it has remained largely in thedomainofhobbyists and designers producing one-offprototypes. And printing objects with anything other than plastics—in particular, metal—has been expensive and painfully slow.

虽然3D打印技术已经存在了几十年，但它仍然主要停留在爱好者和设计师制作一次性原型的领域。用塑料以外的任何东西——尤其是金属——来打印物品，既昂贵又缓慢。

Now, however, it’s becoming cheap and easy enough to bea potentially practical way of manufacturing parts. If widely adopted, it could change the way we mass-produce many products.

然而，现在它变得足够便宜和简单，成为制造零部件的一种潜在的实用方法。如果被广泛采用，它可能会改变我们大批量生产许多产品的方式。

In the short term,manufacturers wouldn’t need to maintain large inventories—they could simply print an object, such as a replacement part for an aging car, whenever someone needs it.

短期来看，制造商不需要保持大量库存ーー只要有人需要，他们就可以打印一样东西，比如老旧汽车的替换零件。

In the longer term, large factories that mass-produce a limited range of parts might be replaced by smaller ones that make a wider variety, adapting to customers’ changing needs.

从长远来看，大规模生产有限零部件的大型工厂可能会被生产更广泛品种的小型工厂取代，以适应客户不断变化的需求。

The technology can create lighter, stronger parts, and complex shapes that aren’t possible withconventional metalfabrication methods. It can also provide more precise control of themicrostructure of metals. In 2017, researchers from the Lawrence Livermore National Laboratory announced they had developed a 3-D-printing method for creating stainless-steel parts twice as strong as traditionally made ones.

这项技术可以创造出更轻、更坚固的零件，以及传统金属加工方法无法实现的复杂形状。 它还可以对金属的微观结构提供更精确的控制。 2017年，劳伦斯利福摩尔国家实验室的研究人员宣布，他们已经开发出一种3-d 打印方法，可以制造出强度是传统制造的两倍的不锈钢部件。

Another Boston-area startup, Desktop Metal, began to ship its first metal prototyping machines in December 2017. It plans to begin selling larger machines, designed for manufacturing, that are 100 times faster than older metal printing methods.

2017年12月，波士顿地区的另一家初创企业 Desktop Metal 开始向市场供应首批金属成型机。 该公司计划开始销售专为制造业设计的大型机器，其速度比以前的金属印刷方法快100倍。

The printing of metal parts is also getting easier. Desktop Metal now offers software thatgenerates designs ready for 3-D printing. Users tell the program the specs of the object they want to print, and the software produces a computer model suitable for printing.

金属部件的印刷也变得越来越容易。 Desktop Metal 现在推出了一款软件，可以随时生成3d打印设计。 用户告诉程序他们想打印的对象的规格，软件就会生成适合打印的计算机模型

AI for Everybody

每个人的AI

云智能

Artificial intelligence has so far been mainly the plaything of big tech companies like Amazon, Baidu, Google, and Microsoft, as well as somestartups. For many other companies and parts of the economy, AI systems are too expensive and too difficult to implement fully.

到目前为止，人工智能主要是大型科技公司，如亚马逊，百度，谷歌和微软，以及一些初创公司的玩物。 对于许多其他公司和经济领域来说，人工智能系统过于昂贵，而且难以全面实施。

What’s the solution? Machine-learning tools based in the cloud are bringing AI to a far broader audience. So far, Amazondominatescloud AI with its AWS subsidiary. Google is challenging that with TensorFlow, an open-source AI library that can be used to build other machine-learning software. Recently Google announced Cloud AutoML,a suite of pre-trained systems that could make AI simpler to use.

解决办法是什么？ 基于云计算的机器学习工具正在把人工智能带给更广泛的受众。 到目前为止，亚马逊凭借 AWS 的子公司主宰了云人工智能。 谷歌对 TensorFlow 提出了挑战，这是一个开源的人工智能库，可以用来构建其他机器学习软件。 最近，谷歌宣布云自动化，一套可以使人工智能使用起来更简单的预先训练的系统。

Microsoft, which has its own AI-powered cloud platform, Azure, is teaming up with Amazon to offer Gluon, an open-source deep-learning library. Gluon is supposed to make buildingneural  nets—a key technology in AI thatcrudely mimicshow the human brain learns—as easy as building a smartphone app.

拥有自己的人工智能云平台Azure的微软(Microsoft)正与亚马逊(Amazon)合作，提供开源的深度学习库Gluon。Gluon被认为可以使构建神经网络——人工智能中粗略模仿人脑学习方式的关键技术——像构建智能手机应用程序一样简单。

It is uncertain which of these companies will become the leader in offering AI cloud services.  But it is a huge business opportunity for the winners.

现在还不能确定这些公司中的哪一家将成为提供人工智能云服务的领导者。 但对于赢家来说，这是一个巨大的商机。

These products will be essential if the AI revolution is going to spread more broadly through different parts of the economy.

如果人工智能革命要在不同的经济领域更广泛地传播，这些产品将是必不可少的。

Currently AI is used mostly in the tech industry, where it has created efficiencies and produced new products and services. But many other businesses and industries have struggled to take advantage of the advances in artificial intelligence.Sectors such as medicine, manufacturing, and energy could also be transformed if they were able to implement the technology more fully, with a hugeboost to economic productivity.

目前人工智能主要应用于科技行业，它提高了效率，生产了新产品和服务。但其他许多企业和行业一直难以利用人工智能的进步。如果医疗、制造业和能源等行业能够更全面地实施这项技术，并极大地提高经济生产率，它们也可能转型。

Most companies, though, still don’t have enough people who know how to use cloud AI. So Amazon and Google are also setting upconsultancy services. Once the cloud puts the technology within the reach of almost everyone, the real AI revolution can begin.

尽管如此，大多数公司仍然没有足够的知道如何使用云人工智能的人手。 因此，亚马逊和谷歌也在建立咨询服务。 一旦云技术让几乎所有人都能接触到，真正的人工智能革命就可以开始了。

Dueling Neural Networks

决斗神经网络

决斗神经网络

Artificial intelligence is getting very good at identifying things: show it a million pictures, and it can tell you with uncanny accuracywhich onesdepict apedestrian crossing a street. But AI is hopeless at generating images of pedestrians by itself. If it could do that, it would be able to create gobs ofrealistic butsynthetic pictures depicting pedestrians in various settings, which a self-driving car could use to train itself without ever going out on the road.

人工智能正变得非常擅长识别事物:向它展示一百万张照片，它就能以不可思议的准确性告诉你哪些照片描绘了一个过街的行人。但人工智能本身无法生成行人图像。如果它能做到这一点，它将能够创造出大量逼真的合成照片，描绘出各种场景下的行人，自动驾驶汽车就可以利用这些照片训练自己，而无需上路。

The problem is, creating something entirely new requires imagination—and until now that has perplexed AIs.

问题在于，创造一种全新的东西需要想象力，而这一直困扰着人工智能。

The solution first occurred to Ian Goodfellow, then a PhD student at the University of Montreal, during an academic argument in a bar in 2014. The approach, known as agenerative adversarial network, or GAN, takes two neural networks—thesimplified mathematical models of the human brainthat underpin most modern machine learning—and pits them against each other in a digital cat-and-mouse game.

2014年，当时还是蒙特利尔大学(University of Montreal)博士生的伊恩·古德费勒(Ian Goodfellow)在一家酒吧进行学术辩论时，首先想到了解决办法。这种方法被称为生成对抗网络(简称GAN)，它采用了两种神经网络——支持大多数现代机器学习的人脑简化数学模型——并在一个数字猫鼠游戏中让它们相互竞争。

Both networks are trained on the same data set. One, known as thegenerator, is tasked with creatingvariations on images it’s already seen—perhaps a picture of a pedestrian with an extra arm. The second, known as thediscriminator, is asked to identify whether the example it sees is like the images it has been trained on or a fake produced by the generator.

这两个网络都是在相同的数据集上训练的。其中一个被称为“发生器”，它的任务是在它已经看到的图像上创建不同的变体——可能是一张多了一只胳膊的行人照片。第二种被称为鉴别器，它被要求识别它所看到的例子是否像它训练过的图像，或者是由发生器产生的假图像。

Over time, the generator can become so good at producing images that the discriminator can’t spot fakes. Essentially, the generator has been taught to recognize, and then create, realistic-looking images of pedestrians.

随着时间的推移，该发生器可以变得非常善于产生图像，以至于鉴别器都无法识别。从本质上说，发生器已经学会识别，然后创造出逼真的行人图像。

The technology has become one of the most promising advances in AI in the past decade, able to help machines produce results that fool even humans.

在过去的十年里，这项技术已经成为人工智能领域最有前景的进步之一，它能够帮助机器产生甚至连人类都能愚弄的结果。

GANs have been put to use creating realistic-sounding speech andphotorealisticfake imagery. In onecompelling example, researchers from chipmaker Nvidia primed a GAN with celebrity photographs to create hundreds of credible faces of people who don’t exist. Another research group made not-unconvincing fake paintings that look like the works of van Gogh. Pushed further, GANs can reimagine images in different ways—making a sunny road appear snowy, or turning horses into zebras.

GANs已经被用来创建逼真的语音和假图像。 在一个引人注目的例子中，芯片制造商英伟达(Nvidia)的研究人员用名人照片为GAN植入了数百张真实面孔，这些面孔都是不存在的人。 另一个研究小组制作了看起来像梵高作品的假画，这些画并非令人难以置信。 再进一步推进，GANS可以以不同的方式重新想象图像——使一条阳光明媚的道路看上去像是雪地，或者把马变成斑马。

The results aren’t always perfect: GANs can conjure up bicycles with two sets of handlebars, say, or faces with eyebrows in the wrong place. But because the images and sounds are often startlingly realistic, some experts believe there’s a sense in which GANs are beginning to understand theunderlying structureof the world they see and hear. And that means AI may gain, along with a sense of imagination, a more independent ability to make sense of what it sees in the world. —Jamie Condliffe

结果并不总是完美的: GANs 可以想象出有两套车把的自行车，或者是眉毛长错地方的脸。 但由于这些图像和声音往往逼真得令人吃惊，一些专家认为，有一种感觉，GANs开始理解他们所见所闻的世界的基本结构。 这意味着人工智能可能会随着想象力的增加，获得更多的独立能力，来理解它在世界上看到的东西。 ー杰米 · 康德利夫

Genetic Fortune-Telling

遗传占卜

遗传占卜

One day, babies will get DNA report cards at birth. These reports will offer predictions about their chances of suffering a heart attack or cancer, ofgetting hooked on tobacco, and of being smarter than average.

总有一天，婴儿出生时会得到DNA报告卡。这些报告将预测他们患心脏病或癌症、染上烟瘾以及比普通人更聪明的几率。

The science making these report cards possible has suddenly arrived, thanks to hugegenetic studies—some involving more than a million people.

使这些报告卡成为可能的科学突然出现了，这要归功于大量的基因研究ーー其中一些研究涉及一百多万人。

It turns out that most common diseases and many behaviors andtraits, including intelligence, are a result of not one or a few genes but many acting in concert. Using the data from large ongoing genetic studies, scientists are creating what they call “polygenic risk scores.”

事实证明，大多数常见疾病以及许多行为和特征，包括智力，都不是一个或几个基因的结果，而是许多基因共同作用的结果。 利用正在进行的大规模基因研究的数据，科学家们正在创建他们所谓的"多基因风险评分"

Though the new DNA tests offerprobabilities, notdiagnoses, they could greatly benefit medicine. For example, if women at high risk forbreast cancergot moremammogramsand those at low risk got fewer, those exams might catch more real cancers and set off fewer false alarms.

虽然新的 DNA 测试提供的是概率，而不是诊断，但它们可能对医学大有裨益。 例如，如果乳腺癌高危女性做了更多的乳房 x 光检查，而低危女性做了更少的检查，那么这些检查可能会发现更多真正的癌症，从而减少虚惊一场。

Pharmaceutical companiescan also use the scores inclinical trialsof preventive drugs for such illnesses as Alzheimer’s or heart disease. By picking volunteers who are more likely to get sick, they can more accurately test how well the drugs work.

制药公司还可以将这些分数用于预防老年痴呆症或心脏病等疾病的药物临床试验。通过挑选更容易生病的志愿者，他们可以更准确地测试药物的效果。

The trouble is, the predictions are far from perfect. Who wants to know theymightdevelop Alzheimer’s? What if someone with a low risk score for cancer puts off being screened, and then develops cancer anyway?

问题是，这些预测还远远达不到完美。 谁想知道他们会不会患上老年痴呆症？ 如果一个癌症风险评分较低的人推迟接受筛查，最后还是发展成了癌症，那该怎么办？

Polygenic scores are alsocontroversial because they can predict anytrait, not only diseases. For instance, they can now forecast about 10 percent of a person’s performance on IQ tests. As the scores improve, it’s likely that DNA IQ predictions will become routinely available. But how will parents and educators use that information?

多基因评分也是有争议的，因为它们可以预测任何性状，而不仅仅是疾病。 例如，他们现在可以预测一个人在智商测试中大约10% 的表现。 随着分数的提高，很可能 DNA 智商预测将变得常规可用。 但家长和教育工作者将如何利用这些信息呢？

To behavioral geneticist Eric Turkheimer, the chance that genetic data will be used for both good and bad is what makes the new technology “simultaneously exciting and alarming.” 

行为遗传学家埃里克 · 特克海默认为，基因数据被使用具有双刃剑的特性使得这项新技术"既令人兴奋又令人担忧"

Materials’Quantum Leap

材料的量子飞跃

材料量子

Theprospect of powerful newquantum computers comes with a puzzle. They’ll be capable of feats of computation inconceivable with today’s machines, but we haven’t yet figured out what we might do with those powers.

强大的新型量子计算机的前景带来了一个难题。 它们将具有当代机器所无法想象的计算能力，但我们还没有想出如何利用这些能力。

One likely andenticing possibility:precisely designing molecules.

一个可能和诱人的可能性: 精确设计分子。

Chemists are already dreaming of newproteinsfor far more effective drugs, novelelectrolytes for better batteries,compounds that could turn sunlight directly into a liquid fuel, and much more efficient solar cells.

化学家们已经开始梦想制造新的且更有效的药物蛋白，更好的新型电池电解质，可以将阳光直接转化为液体燃料的化合物，以及更高效的太阳能电池。

We don’t have these things becausemolecules are ridiculously hard to model on a classical computer. Try simulating the behavior of the electrons in even a relatively simple molecule and you run into complexitiesfar beyond the capabilities of today’s computers.

我们没有这些东西是因为分子在经典计算机上很难建模。试着模拟电子在一个相对简单的分子中的行为，你会遇到远远超出今天计算机能力的复杂性。

But it’s anatural problemfor quantum computers, which instead of digital bits representing1s and0s use “qubits” that are themselves quantum systems. Recently, IBM researchers used a quantum computer with seven qubits to model a small molecule made of three atoms.

但对于量子计算机来说，这是一个自然而然的问题。量子计算机使用的"量子位"本身就是量子系统，而不是代表1和0的数字位。 最近，IBM 的研究人员使用一台有七个量子位的量子计算机来模拟一个由三个原子构成的小分子。

It should become possible to accurately simulate far larger and more interesting molecules as scientists build machines with more qubits and, just as important, better quantum algorithms. —David Rotman

随着科学家制造出更多量子位机器，以及同样重要的、更好的量子算法，精确模拟更大更有趣的分子将成为可能——David Rotman

欢迎关注公众号“程序喵咖啡屋”，带你免费看科技外刊，了解世界尖端技术。