Cloudera首席执行官谈人工智能 | 《财富》专访

一个月前，数据技术公司Cloudera完成了和竞争对手Hortonworks价值52亿美元的合并。

这两家公司现在以Cloudera的名义运营，之前的业务都围绕免费的开源式Hadoop数据处理软件开展，雅虎和Facebook都使用该软件管理和存储数据。此后，两家公司都已经推出了新的开源数据处理技术，进军了大热的机器学习领域。

但是，尽管Cloudera和Hortonworks扎根于热门领域，却仍然无利可图。以免费开源式软件为基础的业务想要赚钱十分困难，很多采用同样策略的公司已经发现了这一点。

Cloudera和Hortonworks合并后仍然面临不少竞争对手，包括资金充裕的数据技术创业公司Snowflake和Databricks，以及亚马逊网络服务和微软等巨头，这两家公司都在推出自己的数据分析和机器学习技术。

Cloudera的首席执行官汤姆·赖利接受《财富》杂志采访时谈了与Hortonworks的合并，谈到了IBM收购开源式企业服务公司Red Hat，以及他为什么更喜欢讨论机器学习而不是人工智能。出于篇幅和清晰表达的考虑，以下内容已编辑：

《财富》：有没有其他同行业公司合并的例子？

赖利：半导体行业的微捷码（Magma）和新思科技（Synopsis）合并了（交易完成于2012年2月）。我之所以知道这事，是因为我之前说过要从其他人的错误中吸取教训。我遇到了微捷码的首席执行官，问他都发生了什么事——什么做得不好，什么做得好。

我希望能重视速度，也就是相对于过去其他任何合并的速度。我了解到，如果推迟决策、迟迟不定不是个好做法。

另外，什么都不能留两份。所以我们已经建立了一个领导团队、一份产品路线图、一个客户支持组织、一个销售团队和一个工程组织。

你们需要裁员吗？

任何合并中都存在重复。为了保证所有的东西都只有一份，会有人失去他们的位置。

幸运的是，因为我们都是高增长公司，很多协同效应都出现在原本打算在以后进行的招聘中，我们都在招聘一些重复的岗位。虽然我们确实有裁员，但相对于其他合并，我们的数量相对较少，因为我们是高增长公司。

你们如何创造一种能将昔日对手团结在一起的文化？

首先，当两家公司合并时，人们总是说会产生文化冲突。我却有不同的看法。

合并并不是将两种文化并在一起，因为我们的文化非常相似。合并是把忠诚的人聚在一起，他们忠诚于各自的老板，忠诚于各自的团队。

公司合并时必须要打破忠诚，这是难点所在。我不得不选一个新的领导团队。我热爱我的旧领导团队，但我必须加入一些新的领导角色。如果双方都后退一步，打破这种忠诚，专注于如何建立一个结合两家公司优点的新企业，会发现不同公司的文化都非常相似。

Hortonworks过去以咨询闻名，而Cloudera更集中于软件销售。合并后新企业的业务重点是什么？

你在竞争时会刻意制造差异。很多时候都是人为的差异。因此，当我们把两家企业合在一起时，天啊，他们的相似点看起来比以前多多了。

我们[Cloudera]主营授权软件，Hortonworks主要提供软件支持服务。抛开那些细微差别——Hortonworks大约82％的业务是软件支持服务创造的营收，我们的业务有82%来自于授权软件的经常性收入，我们也同时为许多软件提供支持。

所以我们的业务几乎完全相同。我们的客户续订率几乎相同。我们的目标市场客户也相同。真正要考虑的是我们重叠的核心业务。他们会说他们更开放[为开源社区服务]，我们会说我们更专注企业级用户。然而，他们也有企业级的客户，我们也是开放的。我们93%的代码都是开放的。

那么由于你们拥有开源技术及付费软件业务，你们公司未来的发展路径是什么？

我们打算成为一家100%开放源代码的公司。你可以说我们用的是Hortonworks的理念，但我们认为这非常重要。数据管理和分析中几乎所有的创新——无论是数据收集还是机器学习——都是开源式的。

你为什么说IBM收购Red Hat是一笔好交易？

IBM为了能在云世界中具有竞争地位，他们可以选择说，“好吧，我们来搭建自己的公共云基础架构，试着在全世界建立数据中心”，这么做会很难。他们也可以收购Red Hat，成为能支持多云[能够同时管理内部数据中心和多个公共云基础架构的能力]的软件抽象层。我认为这种做法很明智。

340亿美元值么？

我认为这比在全世界每一个时区建立20个数据中心要便宜得多，因为数据中心还包括备份、基础设施和冷却设备。

我想我们可以说IBM不是公共云计算的领跑者，对吧？我认为这对他们来说是一个明智之举，这样基本上可以利用所有已经建好的公共云。

谷歌和微软等公司正在推出更方便使用的AI工具，我很好奇这类工具在Cloudera的战略中发挥什么作用。

我们的价值主张是公司可以使用这些工具，但我们希望帮助企业使用它们，把这些工具融入到企业的环境中。

我们假设你想做面部识别。你可以把图片发给谷歌，得到反馈结果。同样的道理，如果你是一家保险公司，你想要推出新的保险产品，就没有和谷歌AI工具类似的黑盒AI系统可供使用。

但是你不认为那些云AI工具取代了公司的内部开发吗？

确实如此。我们想做的是让客户能够开发出把他们和竞争对手区分开的产品或服务。我们要教的是“渔”——给他们提供工具，让他们更加高效。这和我们给他们构建算法的想法不同。我们的工作是帮助他们建立AI工厂——让这个构建新认识的过程自动化。

AI被过度炒作了么？

所以我们的机器学习总经理说，沙特阿拉伯的一个人工智能机器人已成为第一个公民，还有其他一些疯狂的事情。有一点闹剧的气氛在里面，我认为这只会让人们迷惑。我们现在更倾向于讨论机器学习，因为更实用。我们有使用案例——我们知道如何利用机器学习，它们也已经提供了商业价值。

AI和未来的许多理论事物关系更大。我们可以说：“嘿，我们做的是AI，但实际上我们只是做很多和机器学习有关的工作，他们有实际效用。”（财富中文网）

译者；Agatha

A month ago, data technology firm Cloudera finalized a $5.2 billion merger with rival Hortonworks.

The two companies, now operating under the Cloudera name, originally focused on the free open source Hadoop data crunching software, which Yahoo and Facebook used to better manage and store their data. Since then, the two companies have pushed into newer kinds of open-source data processing technologies as well as more buzzy machine-learning.

But despite their roots in a hot space, Cloudera and Hortonworks are still unprofitable. Making money with a business that is based on free open-source software is challenging, as many companies with a similar strategy have found out.

Even with the recent merger, Cloudera and Hortonworks will still compete with rivals including well-funded data technology startups Snowflake and Databricks, along with huge incumbents such as Amazon Web Services and Microsoft, both of which are debuting their own data analytics and machine-learning technology.

In an interview with Fortune, Cloudera CEO Tom Reilly talks about the Hortonworks merger, IBM buying open-source enterprise company Red Hat, and why he prefers discussing machine learning instead of artificial intelligence. The following has been edited for length and clarity:

Fortune: What are some other examples of rival companies merging?

Reilly: Two players in the semiconductor space, Magma and Synopsis, came together [that deal closed in Feb. 2012]. The reason I know that is that early on I said I’ve got to learn from other people’s mistakes. I met the CEO of Magma and asked him what happened—what didn’t do well, what did well.

One of the things I want us to focus on is speed, and speed relative to any other merger in the past. What I’ve learned is delayed decisions and certainty is bad.

Also, don’t keep two of anything. So we already have one leadership team, one product road map, one customer support organization, one sales force, and one engineering organization.

Did you have to do layoffs?

In any merger there are going to be duplicates. To get down to one of everything, there are going to be people that are going to lose their positions.

Fortunately, because we were both high-growth companies, a lot of our synergies were in future hires because we were hiring many of the duplicate things. While we do have layoffs, we’ve had a small number relative to other mergers because we’re such a high-growth company.

How do you create a culture that unifies people who were once rivals?

First off, you read about culture clash when you bring two companies together. I actually have a different view.

It’s not bringing two cultures together, because our cultures turned out to be very similar. It’s about bringing people together who have loyalties—loyalties to their boss and loyalties to the people on their team.

In a merger, you have to break loyalties, and that is the hard thing to do. I had to pick a new leadership team. I love my old leadership team, but I had to put in some new leaders in place. When we step back and break those kind of loyalties and focus on how to build a new business that’s the best of both, it turns out that the cultures are very similar.

Hortonworks used to be known for consulting whereas Cloudera was centered more on selling software. What’s the focus of the new entity?

When you compete, you intentionally create differences. Many times those are artificial differences. So when we brought the businesses together, holy smoke, these businesses look a lot more similar.

So we [Cloudera] license software, and Hortonworks basically licensed support to software. Put aside that little nuance—roughly 82% of their business were revenues from that license support. Eighty-two percent of our business was recurring revenues from licensed software, and we gave support for a lot of software as well.

So it turns out that our businesses are nearly identical. Our customer renewal rates are nearly identical. Our target market customers competed head-to-head. The real way to think of us is in our core overlapping businesses. They would say they’re more open [catering to the open source community], we would say we’re more enterprise grade. And yet they have enterprise-grade customers, and we are open. We have 93% of our code in open source.

So what defines your company going forward as you have open source technology but also a paid software business?

So it is our intent to be a 100% open source company. Now you could say we’re adopting that philosophy from Hortonworks, but we think it’s very important. Nearly all the innovation that’s happening in data management and analytics, from gathering data to doing machine learning is in open source.

Why do you think IBM’s acquisition of Red Hat was a good deal?

For IBM to compete in a cloud world, they could either say, “Well, we’ll build out our own public cloud infrastructure and try to build data centers all over the world,” which would be very difficult. Or they could acquire Red Hat and be that abstracted software layer that enables the multi-cloud [ability to manage both internal data centers and multiple public cloud infrastructures]. I think that is brilliant.

Is $34 billion a fair price?

Well, I think it’s a lot cheaper than trying to build 20 data centers around the world in every time zone, with all of the backup, infrastructure, and cooling.

I think we can all say that IBM wasn’t the front-runner in public cloud computing, right? I think this is a brilliant way for them to basically leverage all the public cloud that’s out there and already built.

Companies like Google and Microsoft are debuting easier-to-use A.I. tools, and I’m curious how those play into Cloudera’s strategy.

Our value proposition is that companies could use those tools, but we want to help enterprises get access to them and bring them into their environment.

So lets say you want to do facial recognition. You could send an image to Google and get some results back. In the same token, if you’re an insurance company and you want to come up with new insurance products, there’s no black-box AI system akin to Google’s AI tools out there to do that.

But you don’t see those cloud AI tools replacing a company’s internal development?

Right. What we want to do is enable our customers to build products or services that distinguish them from their competitors. We are here to teach them how to fish—to give them the tools to make them more efficient. This is contrary to the idea that we’ve built the algorithms for you. Our job is to help them build that A.I. factory—to automate that process that’s building these new insights.

Is A.I. over hyped right now?

So our general manager of machine learning was just saying that in Saudi Arabia the first AI robot has become a citizen or some crazy thing. There’s a bit of a circus atmosphere to it, which I think just confuses the world. We tend to talk about machine learning today, because it’s much more pragmatic. We have use cases—we know how to deliver them and they already deliver business value.

A.I. is more associated with a lot of these theoretical things in the future. We could say, ‘Hey, we’re doing A.I., but pragmatically we’re just doing a lot of really impactful machine learning.”