这些下一代超算运行时太热，得放进冷库

虽然媒体一直紧盯哪个国家能开发出最先进人工智能系统，但另一项重要的竞赛也在不断升温，就是谁能第一个造出量子计算机。

该领域近年来突飞猛进，2018年可能会出现重大突破。这是所谓的“量子霸权”竞赛，研发完成后量子计算机不管处理任何任务性能都会明显超过传统超级计算机。

谷歌和IBM都是量子计算领域的领跑者，也都制定了实现目标的计划。英特尔也加入竞争，上周拉斯维加斯国际消费电子展上发布了为量子计算研究设计的新型49量子比特神经形态芯片。

这项技术牵涉甚广。量子计算机有可能为目前一些最难的数学和计算问题提供新角度，比如人体健康方面分析多组基因的相互作用，为化学品能态建模以及预测原子粒子的行为。互联网的安全性可能会遭到削弱，因为量子计算机可以迅速破解用于保护IT基础设施和网络的密码系统。

有一件事确定无疑，即量子计算时代很快就会到来，而且世界会出现极大变化。

简单来说，量子计算机使用的技术单位是量子比特。普通半导体用一串1和0来表达信息，而量子比特代表的是量子性，计算时可同时作为1和0。也就是说，两个量子比特就可以代表1-0、1-1、0-1和0-0四个数字序列，而且每增加一个量子比特，计算能力就会呈指数级增长。理论上，一台量子计算机只要有50个量子比特的运算能力，就能超过当今性能最强大的超级计算机。

量子计算机来得也正是时候。摩尔定律预计，单位计算能力每18个月就会翻一番，而单位计算能力的价格会下降一半。虽然定律本身没什么问题，但如今实现进步所需的资金已经远远超过以往。换句话说，每次为了提高速度，半导体公司和研究机构增加的研发开支越来越多。而另一方面，量子计算正在迅速崛起。

量子计算公司D-Wave Systems称正在销售的量子计算机有2000个量子比特。不过争议一直存在。有些研究者发现该公司的计算机作用很大，但性能上一直未能超越普通计算机，而且只能用于解决某些问题，即优化方面的问题，优化是指在所有可行解决方案中找出最好选项。举个例子，如果某个复杂模拟问题有多个可能的结果，D-Wave计算机可能就没那么容易解决。此外，外界认为该公司的量子计算方式打败超级计算机的希望并不大。

另一方面，谷歌、IBM以及众多初创公司正努力打造的量子计算机可能会更灵活也更强大，因为能处理各种各样的问题。几年前，新型计算机通常拥有2-4个量子比特。过去一年中，各公司接连发布性能更强大的量子计算机。2017年11月IBM实现关键突破，宣布已造出有50个量子比特的计算机，也是科学家认为量子计算机超越传统超算的水平。

坏消息呢？IBM这款计算机每次保持量子计算状态的时间只有90微妙。实际上，量子计算时普遍存在于不稳定情况，必须采用强力冷却手段才能维持工作；还得另行计算一遍，以便纠正早期量子计算机由于不稳定而出现的计算错误。不过，科学家正迅速改善不稳定的问题，希望五年内能造出可在室温下正常工作的量子计算机。

下面解释一下为什么量子计算与AI结合前景光明。随着大规模应用AI初步产生重大影响，人们意识到基于半导体的传统计算机限制了通过AI解决重大问题的能力。科学家预计，量子计算机不用脱离冷库就能协助一些非常重要的运算。

量子计算有可能填补空缺，通过强大的运算能力解决重大挑战。精准医疗、廉价能源以及超强新型材料都是量子计算可能实现突破的领域，因为量子计算机可以将体积压缩至极小，却能完成数以十亿计的运算。谷歌的研究人员曾描绘过前景——他们用量子计算机模拟了氢分子的电子结构，这是化工设计从凭经验测量和有根据的猜测变成准确的工程和模拟过程中关键步骤（也可用于研发新药）。

虽然应用前景广阔，但量子计算的危害也不容小觑。量子计算机可以轻松破解目前大多数加密模式（虽然安全专家已经在制作量子比特无法破解的加密代码）。举例来说，如果俄罗斯或中国等国家在量子计算领域占据优势，完全有可能从事更隐秘的黑客活动以及破解加密通信等。

在政府、大公司、创业企业和大学实验室里，聪明的工程师们正为量子霸权争分夺秒地研发。最后的胜负结果也确实可能改变全球势力格局。（财富中文网）

维维克·瓦德哈是卡耐基梅隆大学工程学院杰出研究员。亚历克斯·索克埃尔是一位作家、演说家，曾在Mozilla担任营销副总裁。两人合作撰写了《无人驾驶汽车中的司机：技术选择怎样塑造未来》（The Driver in the Driverless Car: How Our Technology Choices Will Create the Future）一书。

译者：Charlie

审校：夏林

While much of the media attention has been focused on the race among nations to develop the most powerful artificial intelligence systems, an equally crucial race has been heating up: the race to build the first working quantum computers.

As progress in the field accelerates at an exponential rate, 2018 should see an avalanche of breakthroughs. It is a race for so-called “quantum supremacy,” when a quantum computer demonstrably and markedly outperforms a classical supercomputer for any class of problems.

Booth Google (GOOG, +0.96%) and IBM (IBM, +2.82%), two leaders in quantum computing, have laid out plans to achieve this goal. Intel (INTC, +2.92%) also has a horse in the race, announcing a new 49 qubit neuromorphic chip designed for quantum computing research at the annual Consumer Electronics Show in Las Vegas last week.

The stakes are enormous. Quantum computers promise to set a new paradigm for solving some of the hardest math and computing problems today—problems such as analyzing the interactions of multiple genes in health outcomes, modeling the energy states of chemicals, and predicting the behavior of atomic particles. They also might make the Internet inherently insecure by quickly cracking modern cryptography used to lock our IT infrastructure and the web.

One thing is for sure: The era of quantum computing is coming on soon, and the world will never be the same.

Put simply, quantum computers use a unit of computing called a qubit. While regular semiconductors represent information as a series of 1s and 0s, qubits exhibit quantum properties and can compute as both a 1 and a 0 simultaneously. That means two qubits could represent the sequence 1-0, 1-1, 0-1, 0-0 at the same moment in time. This compute power increases exponentially with each qubit. A quantum computer with as few as 50 qubits could, in theory, pack more computing power than the most powerful supercomputers on earth today.

This comes at a timely juncture. Moore’s Law dictated that computing power per unit would double every 18 months while the price per computing unit would drop by half. While Moore’s Law has largely held true, the amount of money required to squeeze out these improvements is now significantly greater than it was in the past. In other words, semiconductor companies and researchers must spend more and more money in R&D to achieve each jump in speed. Quantum computing, on the other hand, is in rapid ascent.

One company, D-Wave Systems, is selling a quantum computer that it says has 2,000 qubits. However, D-Wave computers are controversial. While some researchers have found good uses for D-Wave machines, these quantum computers have not beaten classical computers and are only useful for certain classes of problems—optimization problems. Optimization problems involve finding the best possible solution from all feasible solutions. So, for example, complex simulation problems with multiple viable outcomes may not be as easily addressable with a D-Wave machine. The way D-Wave performs quantum computing, as well, is not considered to be the most promising for building a true supercomputer-killer.

Google, IBM, and a number of startups are working on quantum computers that promise to be more flexible and likely more powerful because they will work on a wider variety of problems. A few years ago, these flexible machines of two or four qubits were the norm. During the past year, company after company has announced more powerful quantum computers. In November 2017, IBM announced that it has built such a quantum machine that uses 50 qubits, breaking the critical barrier beyond which scientists believe quantum computers will shoot past traditional supercomputers.

The downside? The IBM machine can only maintain a quantum computing state for 90 microseconds at a time. This instability, in fact, is the general bane of quantum computing. The machines must be super-cooled to work, and a separate set of calculations must be run to correct for errors in calculations due to the general instability of these early systems. That said, scientists are making rapid improvements to the instability problem and hope to have a working quantum computer running at room temperature within five years.

And here’s where the confluence of quantum computing and AI looks so promising. As we are seeing the first major impacts of wide-scale artificial intelligence, we are also realizing that classic semiconductor-based computing limits our ability to solve the biggest problems that we had hoped artificial intelligence could tackle. They expect quantum computers to start performing very useful calculations well before they’re ready to leave the freezer.

Quantum computing promises to step into that breach and provide the rocket fuel needed to solve these grand challenges. Precisely targeted medical treatments, radically cheaper energy production, and new types of super-strong materials are all breakthroughs that quantum computing could make possible by performing billions and billions of calculations simultaneously in a relatively small package. Google researchers demonstrated the promise when they used quantum computing to simulate the electron structure of a hydrogen molecule, a key step toward moving chemical design from empirical measurement and educated guesses to more proper engineering and simulation. (This will also work for drug discovery.)

The perils of quantum computing are also real. Quantum computers will be able to easily crack most forms of encryption in use today (although security experts are already at work on creating codes that are not crackable by qubit attack). Should Russia or China, for example, gain quantum computing dominance—which is entirely possible—they could use their advantage for even more sophisticated hacking and decrypting of encoded communications.

Between governments, big companies, startups, and university labs, some of the brightest engineering minds are rushing toward quantum supremacy. This literally could shift the global balance of power.

Vivek Wadhwa is a distinguished fellow at Carnegie Mellon University’s College of Engineering and Alex Salkever is an author, public speaker, and former vice president of marketing at Mozilla. Together they authored The Driver in the Driverless Car: How Our Technology Choices Will Create the Future.