微软的未来在于语音控制

未来，人们想通过应用订披萨时，再也不必浪费时间摆弄他们的智能手机了。他们只需要对着设备说出要求，一款小软件助手便可以帮他们搞定一切。

至少在微软CEO萨提亚•纳德拉的眼中，这就是未来的发展趋势。在旧金山召开的微软年度开发者大会上，纳德拉介绍了他眼中人机交互的未来愿景。

纳德拉认为，与互联网的发展，以及在移动设备上使用触摸屏一样，利用“人类语言的力量”与机器沟通，将产生深远的影响。

在纳德拉眼中的未来，微软个人助手软件Cortana居于核心地位。微软一直致力于将Cortana打造成消费者的便利助手，例如用它来询问电影院的方向，提醒工作任务，甚至发送电子邮件等。

与苹果Siri和Google Now等数字助手类似，Cortana基于机器学习算法，通过消化用户每次提问题时输入的数据，训练服务能力。虽然数字助手不能始终正常运行，但它们正在变得越来越好。

纳德拉断言，Cortana最终将更加擅长回答问题，人们不必移动手指，只要张开嘴说出要求，便可以在设备上执行日常任务。

除了Cortana，纳德拉还认为，所谓的聊天机器人也是人机交互的一个重要元素。这些软件程序被越来越多的零售商和投资公司用于网站上，鼓励投资者提问基本问题，由聊天机器人进行回答。

与数字助手类似，由于程序员赋予了聊天机器人机器学习能力，它们正在变得越来越擅长理解人类语言。随着人工智能的发展，如自然语言处理等，机器人现在可以执行更加高级的任务，比如为那些通过职场通信服务Slack提出请求的用户预定Uber打车服务。

微软高管在舞台上演示了如何完全通过软件订甜品外卖。

例如，微软Skype Internet呼叫服务高管莉莉安•里肯演示了一番她如何与Skype内置的Cortana对话，让Cortana与一家面包店创建的软件机器人交流，协调蛋糕外卖。此外，里肯还演示了，未来Cortana如何从用户之前的聊天信息中学习，预测用户的需求。

例如，她演示了Cortana如何推测出她将出差到外地，参加一个会议。之后，Cortana决定提供帮助，通过与威斯汀酒店的聊天机器人交流，为她提供在出差日期可用的房间。

一切都非常顺利。当然，这毕竟只是演示。在现实中，人工智能技术和聊天软件的效果尚不尽如人意。

纳德拉说道：“我们希望创建的技术能够学会人性最好的一面，而不是最恶的一面。”这引起了台下程序员们的一片笑声。

他们之所以发笑，是因为上周，一个微软研究项目推出了一款试验性聊天机器人Tay，最终却失去了控制，因为爱恶作剧的网民令这款机器人吸收了大量冒犯性的语言，让她发表了许多不当言论。对于这个问题，纳德拉表示，微软“很快意识到这款机器人远达不到上市的要求”，随即让Tay下线。

纳德拉表示：“现在我们要从头开始设计。”

将这些小问题放在一边，很显然纳德拉将聊天和语音命令，视为扩大微软影响力的途径之一。另外值得注意的是，他是在开发者大会上描绘语音交互的未来的。

微软不断向程序员们传递一个信念：他们可以信任公司，利用微软的技术开发软件。这已经不是前任CEO史蒂夫•鲍尔默领导下的微软。鲍尔默倾向于把微软打造成一家更封闭的公司，批评者认为，这导致微软错失新技术带来的良机。2014年，鲍尔默自己也承认，在他的任期内，微软未能预见到移动计算的崛起。

纳德拉希望程序员们相信，如今的微软已是最新技术潮流的引领者。

虽然微软在大肆夸耀Cortana有多么智能，但要想实现纳德拉的未来愿景，微软需要其他公司开发出先进的聊天机器人。如果没有外部程序员开发出各种可支持Cortana订酒店或订糕点的聊天机器人，纳德拉的语音交互未来将很难成功。

译者：刘进龙/汪皓

审校：任文科

In the future, people won’t have to waste time fidgeting with their smartphones as they try to order pizza from an app. They’ll just have to talk to the device, and little software helpers will do all the work for them.

At least, that’s the way Microsoft CEO Satya Nadella sees it. Nadella outlined his vision of the future of human and computer interaction during Microsoft’s annual developer conference in San Francisco.

For Nadella, using “the power of human language” to communicate with machines will be as profound as the development of the Internet and the use of touchscreens on mobile devices.

At the center of Nadella’s glimpse into the future is Microsoft’s MSFT 0.62% Cortana personal assistant software. Microsoft has been pushing Cortana as an easy way for consumers to ask for directions to movie theaters, remind them of tasks, and even send emails.

Similar to other digital assistants like Apple’s AAPL 1.69% Siri and Google Now, Cortana is powered by machine-learning algorithms that train the service as it ingests data from users every time they ask it a question. Although these digital assistants don’t always work correctly, they are getting better.

Nadella is betting that Cortana will eventually get so good at answering questions that people will stop moving their fingers and start moving their mouths to perform daily tasks on their devices.

Besides Cortana, Nadella also believes that so-called chat bots will be an important element for computer interaction. These software programs, which are increasingly used by retailers and investment companies on their websites, prompt visitors to ask basic questions and then answer them.

Like digital assistants, however, these chat bots are getting better at understanding human language as coders wire them up with machine-learning capabilities. With more advances in artificial intelligence like natural language processing, bots are now capable of performing more advanced tasks like ordering a ride from Uber for people who ask for one through workplace messaging service Slack.

Microsoft executives demonstrated on stage how placing orders for a dessert delivery would be almost completely handled by software.

For example, Lilian Rincon, an executive for Microsoft’s Skype Internet calling service, showed that she could talk to Cortana within Skype, and have Cortana interact with a software bot created by a bakery to coordinate a cake delivery. Rincon also showed off how Cortana would one day be able to learn from a user’s previous chat messages and anticipate what that person will want.

In one example, she showed how Cortana had deduced that she would be traveling out of town for a conference. It then decided to help her by looping in a chat bot from the Westin Hotels to show her some available rooms during the dates she was traveling.

Everything worked smoothly. But this was merely a demonstration, after all. In real life, artificial intelligence technologies and chatting software don’t always work as intended.

“We want to build more technology so that it gets the best of humanity and not the worst,” Nadella said as the audience of coders laughed.

They were chuckling because last week, a Microsoft research project involving an experimental chat bot named Tay went haywire after Internet pranksters fed the bot offensive language, which trained it to say hostile things. Nadella addressed the problem by saying that Microsoft “quickly realized that it was not up to the market” and took Tay offline.

“And so we are back to the drawing board,” said Nadella.

Hiccups aside, it’s clear Nadella views chatting and voice commands as a way to spread Microsoft’s influence. It’s also noteworthy that he laid out the future of voice interactions at a developer conference.

Microsoft is continuing to push the idea to coders that they can trust the company and build software using Microsoft technology. This is not the same company under former CEO Steve Ballmer, who preferred a more closed business that critics say missed the boat on new technologies. Ballmer evenadmitted in 2014 that Microsoft failed to anticipate the rise of mobile computing during his tenure.

Nadella wants coders to believe that Microsoft is a new company that is on top of the latest technological trends.

And while Microsoft may brag about how smart Cortana is, it will need companies to build the advanced chat bots that will help Nadella’s vision of the future to become reality. Without outside coders building the versatile chat bots that enable Cortana to book hotels and order pastries, Nadella’s future of voice interactions will fail to pan out.