2014年以来，雅虎、英特尔、Dropbox、领英、Pinterest以及推特都收购了人工智能公司。人工智能领域的民间投资在过去4年里平均每年增长62%。
目前在市场上有两类虚拟助理，其中一类是综合类，比如微软Cortana、苹果Siri、亚马逊Alexa 等，它们会回答各种问题、设定闹钟、给我们讲一些预先设定好的笑话，并且一直在“进步”。另外一类是垂直应用类虚拟助理，比如《创业美国》第5季第1集中的人工智能系统，专注于会议日程安排的X.ai。
X.ai的创始人和首席执行官叫做Dennis，英属哥伦比亚计算机专业毕业的他从丹麦来到纽约。公司的创意来自Dennis的亲身体验。2012年全年，他为自己安排了1019次会议，而其中670次会议在安排之后都有所变动，这样的过程是恼人且没有效率的。于是他设计了一个名为Amy的虚拟助理，她能代替你处理循环往复的会面安排邮件，并学习用户行为，根据用户此前的选择和偏好提供更合适的会面地点。
市场中，与X.ai相似的产品还有Clara Labs、Calendly、Skylar Labs等。而Clara Labs与X.ai高度相似，同样是通过邮件帮助安排日程的虚拟助理机器人。在竞争如此激烈的人工智能领域，X.ai已经融资超过3400万美元，并已成为同类产品中的领头羊。
觉得节目意犹未尽？我们也整理了更多未曾曝光的X.ai创始人中英访谈实录
★人工智能需要攻克的技术难点
★X.ai如此人性化的幕后智囊团
★哪些瓶颈人工智能依旧无法突破
★X.ai的商业模式和营销策略
－《创业美国》第5季第1集完整视频－
采访者：创业美国陈一佳（以下简称C）
受访者：DENNIS R. MORTENSEN（以下简称D）
什么是人工智能秘书
C: Hi Dennis, Hi! I have already met Amy and Andrew through Email. They are working great.
你好Dennis！我已经通过Email和Amy还有Andrew见过面了，他们都很厉害。
D: They obviously work so great that you are here today at the right address at the right time.
很显然就是因为有他们，今天我们才能顺利的在对的时间对的地址见面。
C: Tell us exactly what they do for us?
能跟具体我们讲讲X.ai可以帮我们做什么吗？
D: So we've spent the last three years in creating this intelligent agent that can be your assistant and manage your calendar. So that if you email me and say, hey Dennis, Do you got time to meet up in Manhattan, come first week of March. I can reply back and say, yeah I'm up for that. I have CC in Amy, who will help put something on my calendar. Amy it's not a human being. That's a machine who understands just what I asked her to do. She'll remove me from the conversation, reach out to you or more participants have a very human like negotiation, figure out exactly when we supposed to meet and upon conclusion, send us an invite. It's not that you haven't heard about this before. This exists already. It's called The Human Personal Assistant. He or she is just very expensive. So we trying to figure out how can we take that experience and give it to everybody.
我们过去三年一直在潜心研发这个可以帮你管理日历的智能助手。如果你现在给我发邮件说，Dennis，三月的第一周你有没有空，我们在曼哈顿见个面。我可以回复他说，好啊我可以。我抄送了 Amy, 她就可以帮助我记录在我的日历上。然而，Amy并不是人类，她只是一个能够理解我需求的智能系统。接下来，我不需要继续在参与后续的谈话，她会直接与你和其他的参与者进行人性化的沟通，并根据相关人员的时间安排出准确的日程。这不再是闻所未闻的事情，事实上他早就存在了，那就是私人助理，只不过一般请助理的代价昂贵罢了。所以，今天我们做的事，是希望改变这样的情况，让每个人都有能享受拥有私人助理的体验。
C: Amy has a name, even Amy is not a human being. And Andrew as well. How did you name them?
尽管他们都不是人类，但却依然有都有名字。你是怎么给他们命名的呢？
D: So if you think about it. They are call Amy Ingram and Andrew Ingram. So what are the initials?
你可以想想看，他们全名的首字母缩写是什么？
C:AI.
人工智能（ Articicial Intelligence ）
D: But I think it's probably more important to kind of underline that when you create one of these agents, you have to decide whether you humanize it or not. As in to call it Google assistant or Alexa or Amy or Siri and you need to pick one of those two. You can't really do anything in the middle. So it's either non-humanized or humanize and we just see so much value in humanized the agent.
但我觉得更重要的可能是。当你创造这些程序的时候，你需要决定是否要将它们人性化。是叫它谷歌助手还是类似 Alexa、Amy或者Siri这样的名字。在这个问题上你没办法折中，只有这两个选项, 而我们看到了赋予它人类的属性会更有价值。
C: Talking about humanized, do Amy and Andrew have personality?
说到人性化，Amy和Andrew是否都有自己的性格呢？
D: I think they do!
我认为他们都有！
C:How did you design them?
你是怎么设计他们的呢？
D:So even if we didn't apply a personality just by the fact that we communicate with other human beings, they will apply some persona to whoever they speaking to. So we've certainly created some persona for we design all the output against. So if you think about what we do which is not an app, Not a Web site, not a plugin or extension. I'm just a dude and if you e-mail me I'll just write you back and hopefully I'm kind of polite. and Amy needs to do the same. So we spent a lot of time in trying to figure out how do we decide Amy's so that she is diligent, polite. attentive. Not overly casual, not overly formal, but some version in the middle. So we do have this persona backdrop and we think it's working quite well.
事实上即使我们不为他们设计性格，和他们对话沟通的人类，也会对Amy和Andrew赋予一个人物形象。所以我们确实为我们的产品设计了一些性格上的表现。我们设计的不是应用，不是网站，更不是拓展插件。如果你给我邮件，我会礼貌的回复你，Amy的工作也是如此。 所以我们花费了很多精力致力于将 Amy 变得更勤奋礼貌且细致，言语不要过于随意，也不要过于正式。基于这样的设定，我觉得我们现在做的还算不错。
C: Who are designing Amy's personality? What kind of people? Only programmers or?
是谁在设计Amy的性格呢？是程序员吗还是另有其人？
D: No so that I actually think would be a mistake. And I think many people in them are just allowing their ingenuity however brilliant ingenuity might be to just do random templates here and there. So if you and I did an app for our iPhone or Android device we would hire UI designer and UX person, we know exactly what they're supposed to do with the work the individual little pixels. Who do you hire to design text. So we have to what we call AI interaction designers. So Anna who was a drama major out of Harvard who understands how to create characters.
我不觉得应该让程序员去设计Amy的性格。他们更擅长在他们的领域发挥才华，比如各种各样的模版等等。如果你想设计一个苹果或者安卓的应用，我们还是要雇一个精准了解用户需求和图像处理的UI（用户界面）设计师还有UX（用户体验）人员。所以针对这些文字的设计，我们也另外雇佣了人，我们叫她人工智能互动设计师。我们哈佛戏剧专业毕业的安娜，她知道如何去设定好这些角色。
C: You mentioned Amy and Andrew don't exist on app but though the email, why did you design on that way?
像你提到的，Amy和Andrew不需要借助应用而存在，而是通过邮件工作。你为什么要这么设计呢？
D: The difference between the app and Amy, the Agent is that the app is something which assist me in doing the job. Amy is the one which takes over their job because once I told them what to do, it's not my job anymore. Amy. You can work on seven emails two emails over two weeks over one hour. I don't care. Just do what I told you to do.
应用程序和Amy的区别在于，应用程序只是辅助我工作的软件，而Amy则是那个可以帮我接手工作的帮手，一旦我告诉他们该做什么之后，那就不再是我的工作了。Amy我不管你用多少时间去处理7封邮件也好，2封邮件也好。你只管帮我安排好这些事情。
C: What's the core technology of it?
X.ai的核心技术是什么？
D: The first part is something which you could think of as the natural language understanding. The first part of what we tried to solve here. It's hard because people are so crazy and they'll say things that are ambiguous, untrue, outright lies. we're taking about a year long endeavor to figure this out.
第一部分是自然语言理解，这是我们最先要解决的部分。这也是特别困难的部分因为人类太疯狂了他们会说各种各样的模棱两可、不真实不准确的谎言，我们花费了将近一年的时间去解决这个部分。
D: Then of course if I understood that request what you just asked me to do, what to do with it。 But I need to have some sort of reaction to it or set of actions. And that could be if I ask Amy to set up a new meeting her action should be to understand if there's any constraint. And look at my preferences, assemble e-mail, suggest we meet up next Thursday at 1:00 o'clock at my office so that intelligence part is gonna be the second element to what we do.
在理解了用户需求之后，我们继续做的是对一系列行动做出反应。比如我需要 Amy去安排一个心的会议，她的反应应该先是了解是否有相关的限制，然后考虑我的其他偏好，同时整理相关的邮件内容，并建议我们下周四1点钟在我的办公室见面。智能与理解，是我们第二个技术点。
D: And then the last one is the natural language generation part. I understood it. I knew what action to take. Now I need to communicate to one or more participants, I need to write something back to you which you understand. As in something so easy to understand that just for a second you forget that this is a machine. So those are the three major components.
然后最后一部分是自然语言生成。在理解和反应处理之后，还需要和一个或者多个相关人员进行交流，还需要回复同样他们也可以理解的文字内容。需要是一个让人可以完整理解并忘记她是一个器械的能力。这是我们最主要的三个技术组成。
人工智能真的这么聪明吗
C: Are there any situation where Amy will not be able to answer the question or won't be able to understand ?
有没有一些情况是Amy无法理解无法回复的?
D: So when she does not understand. She'll to say so.
如果有这种情况她会告诉你。
C: Say I don't understand?
她会说她不懂吗？
D: Some slightly nicer version of that. But that's exactly what it is. That happens all the time in real life. So me shooting you an email Sunday night 10:00 p.m. Let's meetup next week. What do I mean? Do I mean in two hours from now or do I mean in seven days and two hours. So that ambiguity happens all the time with real human existence. So she will come back and say you know what. I didn't understand could you please clarify.
会有更好的解释版本，但这种情况的确不可避免。比如周日的晚上10点我邮件给你，说我们下周见面吧。这是什么意思？是说2个小时后的下周，还是7天再2个小时后的下周。这种歧义时常在人类语境中存在。这种情况她就会回来询问说，你能否更明确的一点。
C: Does that happened a lot? That's ambiguous cases?
这种产生歧义的情况时常发生吗？
D: Not a lot. We are on a quest where we want Amy to understand everything, which is impossible. We certainly have that question where we want to push forward than that and see it happen as rarely as possible. But I would like to say that for this particular product, even when we either don't understand or do make a mistake, as in any real human assistant will also make mistakes. As in He's in San Francisco, I gave him times in EST. Yo my bad! Let me suggest some new times we can be. I just didn't know he was at West. Right？ That happens. But Amy can just self-correct. And many people actually don't even notice the mistakes they just see this natural communication.
实际上不是很多。我们总追求希望让Amy了解一切，但这是不可能的。目前我们确实有这样的问题存在而我们也一直在不断地改善，希望尽可能的减少这种情况。我想说的是，即使是真正的助理也会犯类似理解傻姑娘的错误。比如我联系的某个人在旧金山，但我给了他东海岸的时间。我确实给忘了，我们再重新约一下吧。我确实不知道他在西海岸，对吧？这种情况时常发生，不过Amy可以进行自我修正，很多人甚至都不会注意这样的沟通失误。
C: I read something online about one employee left your company because he felt he did a lot of shity work for the robot? so what's your comment of that?
我在网上看到有一些传闻，你的前雇员辞职后表示自己才是人工智能背后擦屁股的人，对此你有什么看法？
D: So today Google have I believe about 65 cars in their self-driving car program. In each of every one of those cars, there's a driver. So the cars aren't self-driving, but does the driver do that. He's obviously not hired to just drive a car around town. He's hired to collect data, so you can call him an AI trainer if you want. As if his job is to help train the machine. Uber have about 30 some odd cars, two people in every car, an engineer and an AI trainer. Why? Because they're in a space where there is no data. You need to collect data. Me though I'm in the same setting.
现在今天谷歌的无人驾驶计划，有65辆无人驾车汽车。在每辆车里其实都有司机，也就是说这些车其实都不是绝对意义上的无人驾驶。但谷歌其实不是雇人去开车，而是雇人去搜集数据。所以你可以叫他们人工智能训练员，这些人的工作的工作就是负责训练机器。优步有30多辆无人驾驶设备，每辆车里都有2个人，一个工程师一个人工智能训练员。这是为什么呢？因为现在他们正处在一个没有数据的领域，人们需要搜集大量的数据，我们公司也是如此。
D: There is no meeting scheduling data set you and me can go download from the Internet tomorrow and start train on. So what we need to do is to schedule a meeting and annotate the hell out of that. As in what was the intend. What are the different entities. And we have three entities, which we need to label in every single piece of data we get, which is time, date obviously, location obviously and people obviously. So those there need to be labelled. The better we label them, the better we can train Amy. We need to do the most training at the beginning of the firm and the least training at the end. Just like, In the very end that self-driving car you might be buying. I don't know what you think, but let's say in 10 years or 20 years doesn't matter. There won't be another human being. It's not like you're saying. So I bought that Tesla. But it came with a person, he's right there, as like of course not, it will just be a car. and all the way there. They would have people help label that data to get to that moment. And we're very much in the same setting. It's just a brutal job. as in label data. But we continue to do that and have dozens and dozens of people who keep labeling data. That's actually how we make Amy better.
现在根本没有基于会议安排的数据库，我们也没办法从网上下载后就开始训练他们，所以我们只能通过会议去标记这些数据。我们需要完成的不同板块的整理，有三个主要板块。这意味着我们需要把每一个获得的数据都进行标注，显然，是时间地点和人。我们把这三个部分标记的越准确，就能越好的训练 Amy。在初期我们的训练越严格，后面的阶段就越轻松。就好像无人驾驶汽车一样，你将来也可能会买，可能是10年20年以后，那时候车里不会有任何人。不是说你买了一辆特斯拉，还同时配备一个人在车里。当然不会了，无人车就是无人车。在这个过程中，一定会有人去搜集并整理数据，我们也刚好处于这个发展的阶段。这样的工作非常残忍，但我们只能继续下去，用数十个人不断的去整理数据。只有这样，我们才能让Amy越来越好。
C: Does That mean AI at current level is not smart enough?
这是不是意味着人工智能现在还不够智能呢？
D: I think one has to distinguish between a human in the loop, and that of annotating data. Some people might just surrendered and say: I don't think this can be a fully autonomous agent. As in this will not be a fully self-driving car. It might just do lane control and I'm happy with that feature. Are you still going to be driving the car yourself, but on the highway, It can kind of keep you in the right lane. Right？The Googles likes right for where we think this could be a fully autonomous car. there won't even be a steering wheel. I don't think AI is not clever enough. I do think some people confused the two. And I think some people might have taken on too ambitious of a goal too early. We spent three years on it so far in a basement, in the dark, in Manhattan to get to this level here and probably have another couple of years before we are exactly where we want to be. But I think we are on a good path.
我觉得这必须要区分人工管理和数据分类注释的定义。有些人可能会让步说，我不觉得她可以成为一个完全自动的系统，就像大家还不完全信赖无人车一样。也许它将来只能自动变道，我就挺满意了。你还是打算自己驾驶，但在高速上，这样的功能可以帮助你控制车道。这也挺好对吧？实际上，这就是人工管理。而有召一日，谷歌可能会发明连方向盘都没有的真正的无人驾驶汽车，实现真的的无人管理。所以我不觉得人工智能还不够聪明，我只是觉得有时候人们混淆了这两个定义。我还觉得有些人过早的制定了一个野心勃勃的目标。我们在这样一个崭新的领域里，花费了3年时间摸索着做到了今天，可能还要再过几年才能真的达到我们的目标，但我确信我们的路走的没错。
C: How many AI trainer do you have?
你现在有多少个人工智能训练员呢？
D: We have Around 70 people there. Does nothing but label data.
目前有70个人，就专门负责数据分类标记。
C: Beside label data, do they do any thing else? like answer email for Amy....
除了数据，他们还有别的工作指责吗？比如为Amy回复邮件？
D: So if you think about these three parts are talked about. So every single word coming out of Amy or Andrew's mouth is machine driven. Every single decision that Amy taken today, whether right or wrong is a machine decision.
就像我们刚刚谈到的三个主要技术。Amy和Andrew所写的每个字都是由机器驱动的。他们所做的任何判断也都是机器的决定。
D: Say I'll give you just a good example here, which we still labeling data for. Human beings are so kind in general that they don't know how to cancel a meeting. So most people will cancel a meeting like this. Hey Dennis. I can't make it today. So sorry. We should really get coffee at some point. What does he say? Is he saying that this meeting is being rescheduled and we will have coffee later? No! he's not. He's just trying to be kind. That though, we need to label a ton of data to figure out what is a cancellation and what is a reschedule. we need human beings who can sense what is what here to label it one or the other. Not once, not twice, not a thousand times, tens of thousands of times because once I've seen that enough.
我来给你举一个特别好的关于我们需要标记的数据的例子。有的时候人会特别客气以至于他们不知道如何取消会议。所以很多人都会找这样的借口。嗨，Dennis，今天不行了，太抱歉了，我们找机会一起喝咖啡吧。他的意思是说要重新安排这次会议，然后等下一起喝咖啡吗？他当然不是，这只是一个委婉的说辞而已。正是因为这种情况，我们需要标记大量的数据，弄清楚什么是取消，什么是重新安排。所以我们才需要人力去分辨具体的情况和标签，这个过程无法一蹴而就，需要成千上万的数据，以匹配各种各样不同的情况。
D: So that is what the trainer do all the time. That Level of detail and the more data we get the smarter the agent becomes. So that at some point I don't even need to train that more. And there's many things which we trained on, for we don't need to human being to training any more.
这些都是训练员在做的事情。在不同层面不同细节上的数据越相近，系统就会越聪明。这些都是为了实现将来有朝一日，我们可以不再继续训练他们。
人工智能是否会侵犯用户隐私
C: Amy has access to people's calendar, people's email. They have lots of information. like how are your going to protect your user's privacy?
Amy有权限登陆别人的日历还有邮件，她掌握了大量的信息。但怎样去保护用户的隐私呢？
D: So here's the thing. It used to be that the more money you pay, the better of the features, or the stronger the features. As in if you want a faster car, you just give me more money. Right？If you want more, it remind of, a better kind of car for the environment. You pay more money. I'm not so sure that if you pay more money for an assistant, the better security. Say you hired Tom, you pay him $60000 a year to sit here in Manhattan and manage your calendar. Tom will make mistakes, He will do a reply all, you've done that, I've done that. I said you shouldn't click reply all here or you wrote Dennis but it's the wrong Dennis. So human is making a ton of all these mistakes.
我觉得是这样的。过去你花的钱越多，获得的产品功能就越强大。如果希望你的车更快，你可以多花钱。对吧？如果还想要其他的，类似更环保的汽车，你也可以多花钱。但我不确定你花更多的钱，你就能请到一个更安全的助理。比如说你在曼哈顿花了6万一年雇了Tom坐在办公室里帮你安排日程。Tom也会犯错误，有时候他发邮件的时候也会不小心回复全部，我们一定都经历过这种情况吧。他不应该点击回复全部，或者他不小心把邮件发给了另外一个Dennis。人无完人。
D: As in Amy will never do a reply all mistake. That concept doesn't even exist in her universe right. Amy will not gossip. As in, she doesn't understand that concept. So I feel quite comfortable that we'll be able to surpass the security of having somebody on staff to do this. Now, on the technical end. One, we only asked for access to your calendar so we can not see your emails. I think that is important! Two, If there's any attachments to the e-mail we actually don't take them in. You are never including information in the PowerPoint which you are assisting me to read. Right. As in, Let's meet up on Wednesday. I wrote it on Slide 7. No you didn't. It doesn't make sense. So we can choose to just discard all the attachments the calendars though. You're absolutely right. We know who meets with who. And we need to protect that at all costs.
但是 Amy不会犯这样的错误，她的设定中没有这样的概念。她也不会八卦，因为她根本不理解那些意思。所以我反而觉得不聘请员工来做这些事更安全。从技术层面来说，我们只会访问客户的日历，所以我们不会看见邮件的其他内容。这是至关重要的！第二，如果电邮中有其他的附件，我们同样不会阅读。就像你不会让我看到PowerPoint里面的信息那样。举个例子，你不会在PowerPoint的第七页写我们周三见面吧。你绝对不会，所以我们除了日历外会过滤其他全部附件。我们只会知道谁要见谁，这也是我们竭力保护的内容。
C: I'm not so sure if you read the news. Amazon echo got involved in a murder case. Amazon is defending, they are system by using the First Amendment. and if that happen to XAI. What are you going to do? like some murder, or some potential got involve in certain meeting time, if FBI ask information from you, What are you going to do?
我不知道你有没有看这样一则新闻，亚马逊的echo智能音箱卷入了一桩命案。而亚马逊通过美国宪法第一修正案的条款来辩护，拒绝泄漏客户隐私。如果这件事发生在X.ai上，比如特定的谋杀，或者可能有关的会议时间和记录，如果美国联邦调查局向你询问信息，你会怎么办？
D: I think if X.ai is ever asked to extract data information from our system, as long as what we're being asked to do is within the framework of the law. we should adhere to that. I think what is being challenged right now is where is that threshold. For where the Amazons, and the apples and others think, that's The FBI and others are taking it too far, and that it's not really for me to decide because we're not the Amazons or the apples. So they will help me set that threshold. whatever we end. And will this play within that.
我认为如果X.ai被要求从系统中提取数据信息，只要对方的要求是基于法律框架内，我们就会坚持遵守。而现在我们面临的挑战在于在这个领域并没有制定好相关的法律门槛。这次无论是苹果还是亚马逊的人都觉得联邦调查局的人有点过头了，这毕竟不是我能左右的。但是他们或许会帮助我们设定规范。不管如何，我们都会根据规范行事。
D: But I'm hoping here is that we end up in a setting for where every software company can protect their customers. And in a setting for when legally required fair and square, law enforcement can extract information when needed. And I know that it's not easy and sometimes it’s a little bit muddled.
我希望最终能有一个体系，让每个软件公司都能保护他们的用户。在法律公平公正的前提下，执法部门也可以在必要的时刻获取信息。这毕竟不是件容易的事情，有时候很容易造成混乱。
关于人类与机械的关系
C: So what kind of jobs will be replaced by Amy?
你觉得Amy会取代掉哪些人的工作呢？
D: I think there's two types of AIs. And again there'll be an infinite amount. but I think you can think about two types of AIs. those that might replace an existing job and those that will democratize access to something that used to be a luxury. Amy won't replace any jobs.
我认为有两种人工智能，当然数量可能不计其数。但我觉得你可以把他们都归位两类。一类可能去取代某些工种，而另一类则会带来便利，让更多人享受到一些原本奢侈的服务。艾米是后者，她不会取代谁的工作。
C: How about secretary?
那类似秘书这样的职业呢？
D: So this a little bit above 10 billion for meetings being set up in the U.S. Every year. Ten billion! That's about two a week on average is by every knowledge worker. so it's probably even a timid number. But two separate studies suggest that that would be the number. Less than 0.1 percent of those are set up by Admin. As in We don't have any Admin. We're on our own. So we're not here to figure out how can I kill the 0.1 percent. That's not even a good business for my investors. Now we are trying to figure out how can I get this to the remaining 99.99%.
每年美国都会开上超过10亿次以上的会，我说的是每年！10亿！要知道，平均每个白领每周都要开会两次，所以这还可能只是个保守的数字。但两项单独的研究都证明了这个数字的可信度。这些人里，只有不到0.1%的是通过行政助理去管理会议的。我们没有行政助理来辅助我们安排会议，大家都还是在依靠自身。我们今天不是要在这里讨论如何去取代这0.1%的用户，投资人们也不会觉得这是一桩好生意。我们更需要研究的如何让剩下的99.9%也享受到我们的服务。
D: So that is democratizing assets. You don't have an assistant though you can be a great recruiter. You don't have an assistant. you can be a good income manager. You don't have an assistant. Or like a company post-acquisition 450 people. Two assistants for seven executives. The rest they are on their own.
它应该要成为一个更大众化的工具。你可能是很厉害的招聘人员，但你没有助理。你也可能是特别优秀的投资收入经理，但你也没有助理。还可能是一家公司收购了450个新雇员，7名高管共享2个助理，其他的人都只能自尽其责。
C: You are not helping of the little percentage of population, who had assistant already, but bigger population who doesn't?
所以你的公司不是要帮助那些已经有助手的人群，而是那些没有助理的工作者？
D: What I would like though is that over the long run Amy becomes so good that even the assistants will using Amy. So that if you go ask any assistant. So you work for your boss do you like manage this calendar. No I fucking hate it. I said that is the least challenging part of the job but I have to do it. But they will much rather do 15 other things.
我所希望的更长远的是，Amy可以发展的越来越好，即便是助理们也愿意使用Amy。你随便去问一个助理，给老板打工的时候你会喜欢帮他们管理日程吗？没人会喜欢这件事，因为这是工作中最没挑战性的一部分了。有这些时间和经理，他们更愿意做其他15件事情。
C: Do you think lots of jobs have been replaced by other robots?
那你觉得会不会有很多工作被其他机器代替呢？
D: Nope, you know I am super optimistic and I'm obviously so optimistic that you can call me naive and that's okay. What we are saying is that a lot of that manual labor, as you suggest might disappear or change to the extent for where all these people could end up in some mass unemployment. I just don't think that is the case.
不会的，我对未来很乐观，我可能太乐观了甚至让你觉得我有点幼稚，不过没关系。大家都说很多人力工作都会消失，或者大量的劳工都会失业，但我并不觉得这是个问题。
D: I do think that first of all whenever something disappear, you and me have the imagination to ask for more. Let's just pick up an example here. So most of my groceries arrive from Fresh Direct. I think they have their warehouse in Long Island, I guess, somewhere not in the city. And somehow it arrives in my apartment building and it arrives downstairs and gets up to my door, and I pay for that and I'm a happy chap. Let's say at some point, in not too distant future that truck will be self-driving. They'll go from long island to my building. The truck won't go into my building so I still need somebody to go in the front door, go up to my front door. On the 19th floor. But then why does it go all the way into my face. Then I just want that! right?! I was work to pay for the service before. so I might actually be willing to pay for tomorrow as well. I need to get more. That's to just an increase in productivity.
但我觉得不管什么东西消失不见，你我都依然会对更多的期待报以想象。让我来举个例子。我平日里大部分的杂货和食物都从是Fresh Direct买的，他们应该在曼哈顿城外类似长岛的某些地方有仓库。我也不知道他们怎么运作的，他们会先抵达我的公寓，然后出现在楼下，然后摆放在我的门口。我是花钱享受这些的，并且我乐在其中。假设在不远的未来，卡车可以自动驾驶了，但他们只能从长岛的仓库开到我家楼下，但仍然需要有人能把货品运上楼放在我门前，这才是我需要的完整的服务。对吧？我以前享受到的，我明天也会继续愿望付钱享受，并且我还愿意享受更多。我觉得这更能体现的是生产力的提高。
D: We've seen that over and over again. Whenever there's an increase in productivity you just ask for more for the same amount. So I just don't see it happen that doomsday scenario. There are a lot of people talk about like truck drivers in the U.S and say millions of people and all the associated jobs around them. What the hell are we going to do here. What the hell we're going to do? We're going to ask for more. And when we asked more we need people to help deliver that.
事实上我们已经司空见惯，历史上一旦生产力提高，总会影响到生产关系的变革。所以我不认可这种悲观的“末日之谈”。大家都在讨论类似美国的卡车司们，讨论可能受到影响的数以百万的基础劳动力们。我们要怎么办呢？人们总会需要更多，这种时候，我们就要靠改变生产关系来实现这一目标。
C: Bill Gates has a very interesting proposal, he proposed to tax robot. What do you think about that?
比尔盖茨有一项特别有意思的提议，她建议跟人类一样，也设立机器人税。你怎么看？
D: He's not crazy and I have a lot of respect for Bill Gates version too and all of the good things that he's been doing of late. So. Taxing machines is difficult. Is going to be difficult to measure and implement. I see where he's coming from and what he's really trying to say here is that if you want to figure out inequality. Replacing a thousand employees with 15 machines is certainly going to make the owner a whole lot richer but a thousand people a whole lot unhappier. And how can we can distribute that and make sure that we take some of the gain from that owner and get it back to that 1000 people. Taxing machines. I don't think he's crazy. I just I can't think through a scenarios for where we can successfully implement that.
我不觉得他疯了，我也很尊重他的说法以及他一直所来的贡献。但向机器纳税太困难了，毕竟现在还没有标准来衡量并实施。我了解他的想法从何而来，他是想尽力解决不平等的问题。用15台机器去取代1000名雇员当然会让老板节约更多成本更有钱，但同时也会让1000个失业的人不高兴。如果进行合理有效的分配，能够把一部分老板的利润返还给那1000个人呢，对机器征税或许可以。我不觉得他的想法不靠谱，我只是无法想象我们可以成功实现这个目标的场景。
商业模式和团队经营
C: What's your business model now》
你现在的商业模式是什么呢？
D:You can get a free edition of Amy that can schedule a few meetings for you. Hopefully you fall in love. Then you buy a professional edition so you can use it in a professional setting. You can schedule unlimited meetings you can choosing the signature and VIPs and do all sorts of good stuff. Hopefully. you then exposure to your colleagues. Then we have a business edition. You can move Amy to your domain you can add all the users, get a single invoice and the company can pay for it. And as the price point is still in the 30-50 bucks per person. So it's not one of those majestic contracts.
你可以先使用免费版本的Amy，她可以为你安排一定数量的会议。如果你觉得好用的话再去购买适用于各种专业场合的高级版本。高级版本可以安排无限次的会议，还可以选择个性签名以及其他有用的高级功能。当然我们也希望你可以跟更多的同事们介绍她，未来我们还会有一个商业版。你可以让Amy的邮件改成企业名称后缀，同时添加更多用户，并用单次支付的方式为公司其他成员申请使用。每人的费用是30-50美元不等，所以也不需要一个复杂的合约。
C: Just like subscription model？
比较类似订阅模式？
D: Yep.
没错
C: Who are using X.ai now?
都有哪些人在使用 X.ai呢？
D: So up until this point we've only had the professional edition. Meaning that single individuals people from Netflix, UBER Salesforce OR the New York Times or what have you will go to the website Sign-Up the pay user and be happy.
现阶段我们还只提供到专业版的服务。意味着我们的用户是一些来自奈飞、优步、Salesforce还有纽约时代杂志的员工们，只要在网上注册，就可以开心的享受我们的服务。
C: How do you reach out to your users and how do you market your product？
你是如何找到客户，把你的产品推向市场的？
D: So we have an Inherently viral product. as in You've tried my product whether you like it or not because that's the only way we can meet up. You try the product, that exposure is a very strong component. That means every paying customer helps us kind of advertise it to the next person. Just because they're thinking hey we went to college together. Why do you have an assistant and I don't have one. You know what. It's because I'm not paying $60000 a year. I pay a lot less. You should get it too.
我觉得我们的产品自带病毒传播的属性。拿你来说，你已经试过我们的产品了，不知道你是不是喜欢她，但这是我们能顺利见面的唯一途径。试验产品就是产品曝光的最佳方式，这意味着每个付费用户都可以帮助我们和其他人推荐产品。因为他们可能在想，我们都一起上的大学，为什么你有助理我就没有。知道原因吗？因为我没有花6万美金一年，我只花了很少的钱，你也应该申请一个。
C: Do you have any competitors?
你有没有类似的竞争者呢？
D: Certainly competitors at the bottom of the startups that are trying to do something similar. All of them up until this point are relying on humans and there's nothing wrong with that. It's just a different business model. And I'm trying to scale on that. But you could make a dropbox sized business out of The humans. It just you can't employ 100000 people and it just doesn't get to that.
现阶段我们的竞争对手都是一些初创类公司，也都在一些类似的事情。在这个阶段很多人工智能团队都非常依赖人力，我不觉得这有什么问题，只是不同的商业模式而已。我也在试着扩大规模，当然即便不纯粹依靠人类，你也有可能会做到一个Dropbox那么大规模的公司。不过你没办法去雇10万个人，这也不完全解决问题。
C: Can you introduce the investment situation now？Which round of fund have you finished?
公司现在的融资情况怎样了？已经第几轮了？
D: So we raised an additional 23 million dollars in April. Last year we've raised about 34 million. We will not really be raising any additional capital this year. What we think is a good 10 million dollars and annual recurring revenue. Once we hit that we think we've done what we're supposed to do within this milestone and then we will go find a good partner to figure out how do we go from 10 to some other interesting number. And that is likely to be the earliest days of 2018.
去年4月我们又融资了2300万美元，去年一年我们总共融资了大概3400万美元。今年我们不会再继续增加融资了。我们都认为1000万美金的稳定年收入是一个目标，一旦我们完成了这个里程碑式的任务我们就会考虑再找其他的合作伙伴，在这个基础上继续更上一层楼。希望有机会能在2018年达成吧。
C: What do you think why investors are interested in you? Like they put a lot of money into you?
创业美国: 你觉得为什么投资人们都对你们这么感兴趣？他们这么愿意给你们投钱？
D:We are nice people. But outside of that I certainly think there's a few things. Including why I'm willing to be investing my own money, my own time and reputation and all that good stuff. One I think there is a paradigm shift happening right now for where some movement from apps to agents is happening. Somebody should grab that. So any new paradigm shift whoever arrives first tends to be able to profit from it. And don't tell me here. Say you staying at the Hilton while you are in New York and it's 11:00 p.m. again and you want diet coke. so your first thought is not one where hey you know what. Let me go to the App Store let me find the Hilton app let me download the app. Let me set up a set of credentials.What you should do though is probably just text the downstairs lobby. Hey guys.can You bring up a Diet Coke. That's an agent. That's a new setting and it's not that the app was wrong. it's just the app wasn't the answer to getting a diet coke. And that actually means that most the time the phone is superior to the app. as in said Room Service. Sometime i don't even want to speak to them, I just want to text them. Right. So think this this paradigm shift is interesting. So investors are certainly investing in this new intelligent agent paradigm. That’s One. Two as I suggested before if there's a real interesting setting for where Amy can speak to herself that suggests there will be a few dominant players. That is their whole business model. They're trying to find Kind of real winners that takes massive markets. I'm not saying we will but at least the opportunity.
因为我们都是非常友好的人。除此之外有几个我为之付出的原因，我觉得科技圈正在酝酿着从手机应用到智能代理的转变。我觉得创业公司应该要把握住机遇。谁最先抓住机遇谁就能最先获得盈利。打个比方，好比如你在纽约的希尔顿酒店，你在房间里想要点一杯健怡可乐。没有人会想到下载希尔顿酒店的手机应用。大多数人想到最方便的服务是发手机短信到前台，这就是智能代理，因为手机应用不是帮助用户获得可乐最便捷的方式。这意味着，对于用户来说，手机之于应用，因为用户有时候只想发信息给前台而不是下载手机应用。所以说种思维的转变是十分有意思的。而投资者也意识到智能代理将会是一个新的爆点。另外，从业内竞争来看，也有不少新的公司加入到这个领域中，所以这是个很大的市场。投资人想要在这个市场中找到最优秀赢家。我不说我们能成为大赢家，但是这个是个很好的机遇。
★
访谈记录依然让你意犹未尽？
还有其他问题想问X.ai创始人？
扫码关注创美并留言
下期《创业美国》为你答疑
带你遇见未来的第一视频账户
微信：chuangyemeiguo
微博：创业美国
知乎：陈一佳
节目视频收看地址
微博：三橙视频
微信：三橙视频
腾讯：一佳创美
今日头条：三橙视频
秒拍：三橙视频
《创业美国》节目音频
凤凰、喜马拉雅、蜻蜓、多听
合作联系
