eDen  伊顿书屋

The following is a 2015 Fortune magazine article. More than two years later, it is still an interesting piece of reading.  

Novermber 1, 2015, Fortune, pp. 102-112. 

THE 21ST – CENTURY CORPORATION: Every Aspect of Your Business is about to change. 

Imagine an economy without friction – a new world in which labor, information, and money move easily, cheaply, and almost instantly.

Psst – it’s here. Is your company ready? 

By Geoff Colvin

Cars bursting into flames are never a good thing. So when a Tesla Model S ran over a metal object in Kent, Wash., in October 2013 and burst into flames, owners, potential customers, investors, and company executives got worried. When the same thing happened a few weeks later in Smyrna, Tenn., federal regulators opened an investigation. We all know what happens next: a massive recall, costly repairs at dealerships nationwide, and a painful financial hit to the carmaker. 

Yet none of that occurred. The problem was that the Model S could lower its chassis at highway speed to be more aerodynamic, and if debris hit the car’s battery pack in just the wrong way, it could catch fire. So Tesla beamed a software update to the affected cars, raising ground clearance at highway speed by one inch. The problem went away. Just four months after opening their investigation, the regulators closed it. 

Using software and the mobile-phone network, Tesla avoided any need for a recall. It doesn’t have any dealerships; customers can configure and order a car online, and they can test-drive cars at company-owned showrooms. Tesla’s advanced electric technology is simpler than gas or diesel technology, so cars can be built with fewer employees and less capital. Combining those factors and here’s what happens: General Motors creates about $1.85 of market value per dollar of physical assets, while Tesla creates about $11. GM creates $240,000 of market value per employee, while Tesla creates $2.9 million. You don’t get differences like that just by being more efficient. Tesla, though in the same business as GM, is a fundamentally different idea. 

GM is changing, but for now it’s still a 20th-century corporation. Tesla is a 21st-century corporation, built for sweeping new realities that change the rules of success. The big theme is the arrival of the long-heralded friction-free economy, a new world in which labor, information, and money move easily, cheaply, and almost instantly. Companies are forming starkly new, more fluid relationships with customers, workers, and owners; are rethinking the role of capital (as traditionally defined), finding they can thrive while owning less and less of it; are creating value in new ways as they reinvent R&D and marketing; and are measuring their performance by new metrics because traditional gauges no longer capture what counts. 

Not all 21st-century corporations are glamorous Silicon Valley startups. They can be of any age and in any industry (even cars). Nike is a 21st-century corporation, aggressively reinventing manufacturing with 3D printing and cannily using social media for marketing. General Electric is becoming one, if partly as a result of shareholder frustration and outside pressure. Every company needs to be one. 

The new realities begin at capitalism’s foundation, capital. In a friction-free economy, a company doesn’t need nearly as much as it used to. Consider the world’s most valuable company, Apple. Unlike Google and Microsoft, the second and third most valuable firms, Apple gets most of its revenue from selling physical products. Yet the company says “substantially all” of its products are made by others. Because it can coordinate vastly complex global supply chains, it can pay those firms, mostly Foxconn, to make its products and get them where they need to be on time. Apple has even rented other companies’ servers to host its iCloud service so that it can add or remove capacity easily, paying only for what it needs. 

The U.S. government classifies Apple as a manufacturer, and with some 500 brick-and-mortar stores worldwide, its total capital - $172 billion of it, according to the EVA Dimensions consulting firm – is immense. But in traditional models it would need much more. Its achievement is using that capital to stunning effect, creating a market value of $639 billion. By comparison, Exxon Mobil uses far more capital, $304 billion, to create a market value, $330 billion, that’s barely half as much as Apple’s. 

Those are companies that make and sell physical stuff. A friction-free economy also enables companies with virtually no physical capital to compete powerfully with capital-heavy incumbents. It’s often observed with wonder that Alibaba is the world’s most valuable retailer but holds no inventory, that Airbnb is the world’s largest provider of accommodations but owns no real estate, and that Uber is the world’s largest car service but owns no cars. Each has found ingenious ways to take friction out of its industry, connecting buyers and sellers directly and conveniently, enabling new, nearly capital-free business models. 

But hold on – actually, those and all 21st-century corporations own tons of capital. Accounting rules just don’t always call it that. There is intellectual capital in the form of software, patents, copyrights, brands, and other knowledge; customer capital in the form of relationships with buyers; and especially human capital. The 21st-century corporation, even if it makes or sells physical products, is above all a human-capital enterprise, which raises a profound question: Who really owns it?

IT WAS OBVIOUS long ago that law firms consist almost entirely of human capital, so it’s illegal for them to sell stock to the public; outside stockholders couldn’t own anything of value. Are consulting firms and ad agencies any different? Even companies that own valuable patents or brands may still get most of their value form human capital. What if the hundred smartest people left Starbucks or Johnson & Johnson or Walt Disney, or what if a crazed CEO tried to destroy each company’s titanium-strength culture? In the 21st-century corporation, whether it’s acknowledged or not, employees own most of the assets because they are most of the assets. 

That reality is affecting corporate structure. The number of U.S. corporations increased only modestly and their revenues rose 150% from 1990 to 2008, says the IRS (using the most recent available data), while the number of proprietorships and partnerships, which are owned by their managers, increased far more, and their revenues rose 394%. The 21st-century corporation isn’t always a corporation. 

Most businesses will have to create value in new ways or lose out to competitors that do so, often with Internet-enabled business models. The trend is as old as the Internet’s early days, when a slew of web insurance upstarts forced term-life premiums to plunge 50% or more – and when user-friendly hotel – and airline-booking sites put some 18,000 travel agents out of business almost overnight. Now entrepreneurs are extending the trend into physical products in sophisticated ways. Warby Parker sells high-quality eyeglasses for a small fraction of what traditional retailers charge by using a low-friction online model; private investors recently value the firm at $1.2 billion. Even an industry that seems highly resistant to online disruption, consumer packaged goods, is threatened. Harry’s and Dollar Shave Club, which make and sell men’s grooming products online, are forcing Gillette (owned by Procter & Gamble) to promote its wares on value, not just quality, for the first time. 

The trend is especially frightening for even established category leaders because even if they switch to new, low-friction business models, they could still end up smaller and less profitable than they were. That’s because “some tech and tech-enabled firms destroy more value for incumbents than they create for themselves, and many gains are competed away in the form of consumer surplus,” says the McKinsey Global Institute. For example, Microsoft’s Skype service brought in some $2 billion in 2013, yet McKinsey calculates that in that year Skype transferred $37 billion away from old-guard telecom firms to consumers by giving them free or low-cost calls. 

Other new business models have similar stories. San Francisco’s taxi regulator reported that the number of fares per licensed cab fell 65% from March 2012 to July 2014 as Uber, Lyft, and others entered the market. Uber – see our companion story in this package – is currently valued at $ 51 billion by its investor. Meanwhile, the cumulative market value of every New York City taxi medallion is less than $13 billion, as Fortune reported in September. 

When Airbnb entered Austin, hotel revenue dropped 8% to 10%, say Boston University researchers, and “affected hotels have responded by reducing prices, an impact hat benefits all consumers, not just participants in the sharing economy.” Yet the new companies causing the disruptions collect only fraction of what the incumbent firms lose. 

The 21st-century corporation will increasingly be an idea-based business, operating not just in infotech but also in media, finance, pharmaceuticals, and other industries that consume lots of brainpower. McKinsey finds that while “asset-light, idea-intensive sectors” generated 17% of Western companies profits in 1999, they generate 31% today. The losers in that shift are capital- and labor-intensive sectors like construction, transportation, utilities, and mining. That doesn’t mean companies in those industries are doomed. As Tesla shows, they may be able to prosper if they’re reimagined. 

Or they can succeed if they redefine success. An intensifying source of pressure on companies of all kinds is the rise of competitors willing to sacrifice profits for growth. Frequently they are family-owned or state-owned companies that have achieved massive scale in emerging markets. For example, Alcoa’s recent decision to split into two companies, a high-tech materials business and a commodity aluminum producer, was prompted in part by the cost advantage achieved by giant Chinese aluminum smelter; forced to compete with them, Alcoa’s commodity business was dragging down the whole company. As emerging-market companies increase their share of global business – they’re now about 30% of the Fortune Global 500 – the profit pressure will increase. 

Further pressure will come from another category of 21st-century corporations that sacrifice profits for growth, those that see vast territories to be grabbed in new-model businesses. Exhibit A is Amazon, which famously reports little or no profit quarter after quarter. Investor agree with CEO Jeff Bezos that the money is better invested in expansion; future profits will be that much greater as a result. The stock recently hit an all-time high. 

HOW TO VALUE THE NEW BREED

Few hard assets, no products to speak of, but intellectual capital aplenty. What’s an investor supposed to do with that? 

THE SO-CALLED UNICORNS, the fast-growing breed of 21st-century startups worth more than a billion dollars, have been so celebrated that it almost seems beside the point to ask, “Are they worth anywhere near their valuations?”

Fledging student-loan provider SoFi recently raised $1 billion in equity at a valuation of more than $3 billion. Facebook paid $19 billion for privately held WhatsApp, a communications tool, before it had meaningful revenue. Uber, which has yet to report a profit, is worth $51 billion, per its latest private-market funding round. Disruptive or not, is there something that their backers are seeing that those in the rational investing world – they who assess companies on the present value of their future earnings – aren’t? 

It honestly doesn’t matter. Private-market valuations are just that, the assessment of a small number of investor operating outside public scrutiny. Public-company valuations are tested every day that markets are open. The valuations of private company shift only when they raise more money. In fact, says venture capitalist Keith Rabois of Khosla Ventures, “private-market valuations are more of an art than a science. They are a negotiation, with the venture capitalist asking, “At what price will somebody who doesn’t need my money take my money?” 

Several factor account for the abundance of venture-backed companies that have remained private well past the time they historically would have gone public. Low interest rates have made capital cheap and have encouraged institutional investor to seek superior returns in riskier bets. In turn, with so much available cash, private companies have been in no hurry to seek the glare of public markets. As a result, they are able to raise ever greater amounts at ever higher valuations.

Sometimes the eye-popping valuations aren’t exactly what they appear to be. For example, in an effort to achieve billion-dollar status – resulting in positive publicity, bragging rights for recruiting talent, and ego stroking for entrepreneurs – some companies accept onerous restrictions in return for higher paper values. Payments-software provider Square, for example. Raised money at a valuation of $6 billion while promising investors that an IPO price below an agreed-upon level would trigger the issuance of additional shares. This would dilute other shareholders – and possibly hurt the valuation of the company. 

All this behind-the-scene maneuvering eventually ends, of course, when companies finally go public. Already, tech companies like Box and Hortonworks, have been unable to hold public values that exceed their previous private valuations. A unicorn, after all, is a mythic animal whose true value is in never being seen. Public companies have nowhere to hide. The new cohort may continue to innovate, but eventually they, too, will be worth only what money they can make for their shareholders over time. – Adam Lashinsky 

SOME OF THE DEEPEST RETHINKING to be done by 21st-century employers will follow from this question: What happens when the labor market becomes friction-free? It’s clearly headed that way, as the rise of the gig economy shows. Companies still employ full-time workers who aren’t really needed full time, but keeping them on staff is easier than constantly hiring and firing. At least it used to be. Now employers are hiring millions of workers worldwide to do information-based work through online marketplaces such as Upwork; each worker is rated by previous employers, and you don’t pay unless you’re satisfied with the work. While much of the work is routine, like language translation, a market place called HourlyNerd rents out former consultants and top business-school graduates to help with strategic planning, financial analysis, and other high-level tasks; customers are mostly small and medium-size businesses but have also included giants like General Electric and Microsoft. 

Project the trend a few steps further, and the whole model of employment could change fundamentally. Employee-owned businesses are likely to increase, but they’re just one option among many, which may eventually include a far more radical structure. Former Cisco CEO John Chambers said in June that “soon you’ll see huge companies with just two employees – the CEO and the CIO.” It’s crazy, except that Chambers has a record of making crazy predictions (like opening your hotel-room door with your smartphone) that eventually come true. 

Even employers that continue with plenty of employees will probably change the relationship. “It’s possible to measure the outcome in almost any job now, so you can reward people accordingly,” an executive of a performance-evaluation software maker told Fortune recently. As a result, top performers are being paid more, and the rest are getting less. Aon Hewitt reports that virtually all large employers now offer bonuses to regular salaried employees, often for achieving specific periodic goals like collecting more receivables or other performance metrics that are now easy to track. When individual performance was cumbersome to measure, pay was less differentiated and underperformers could keep their jobs. No more.

What’s true for workers is true for the 21st-century corporation itself. As friction disappears and ambitious new competitors arise in emerging markets, underperforming companies can’t hide. Winners will win bigger, and the rest will fight harder for what’s left. Idea-intensive sectors “are developing a winner-take-all dynamic, with a wide gap between the most profitable firms and everyone else,” says new research from the McKinsey Global Institute. More generally, competition is simply getting tougher. Global corporate profits recently totaled about 10% of world GDP, says MGI, a number we may someday recall with envy; the profit share could shrink to 8% by 2025, MGI predicts, barely more than it was in 1980. Result: “As profit growth slows, there will be more companies fighting for a smaller slice of the pie.” 

IT’S A WORLD in which corporations, though fighting ever harder, keep less of the global economy’s output – seemingly the recipe for a declining role in world affairs. Yet for many 21st-century corporations it will be just the opposite. Some are achieving the scale of nations, a new phenomenon. Conducting billions of searches a day, Google possesses better real-time knowledge of what’s going in the world than any government does; research shows it can predict disease outbreaks, stock market movements, and much else, and could influence elections if it wanted to. With 1.5 billion users, Facebook has a bigger population than China does and can accurately describe its users’ personalities and predict their success in work and romance. On any given day, Apple probably has more cash on hand than the U.S. Treasury. Bharti Airtel, an Indian telecom company, has about as many customers as the U.S. has residents. With 2.2 million workers, Walmart employs more people than any other organization on earth except the U.S. and Chinese defence departments. 

And now one more mind-bending concept for the 21st century： Corporations, even as some achieve colossal stature, will on average live shorter lives than they used to. The trend is striking: The average life span of companies in the S&P 500 has declined from 61 years in 1958 to about 20 years now, says Yale’s Richard Foster, who predicts further steady declines. Well before the 21st-century’s end, the concept of companies as continuing institutions could even cease to be the norm. 

After all, why do companies exist? The English economist Ronald Coase won a Nobel Prize in economics for answering that question. In the theoretical world, the global economy spins like a top based on price signals between individual operators, with no apparent need for big companies. But in the real world, as Coase pointed out, “there are negotiations to be undertaken, contracts have to be drawn up, inspections have to be made, arrangements have to be made to settle disputes, and so on.” That is, there are transaction costs - friction - and consolidating transactions inside companies is the most efficient way of handling them. Now, as technology shrinks those costs, many companies are unbundling themselves, outsourcing functions to others, crowdsourcing R&D, and exchanging employees for contractors. A continual Hollywood model, in which people and resources come together to achieve a goal and then disperse to other projects, may become common across the economy. It’s happening already.

The good news is that accelerating change, creative destruction, and new business models are all opportunities for the venturesome. A unifying theme as the economy transforms is that in almost every business, barriers to entry are coming down. Opportunity is more widely available than ever. Every person and every organization can possess the 21st century’s most valuable assets: openness to new ideas, ingenuity, and imagination. 

2018.09.12 

AlphaGo Zero Mirrors Limitation of Human Experience

Author: Bin Chen
Last modified: 2017-10-30 16:05:03
Source: Southern Weekly Editorial (The article was originally published in Southern Weekly on October 26, 2017)
Translated by Qun Rene Chen (陈群）

On October 19, 2017, Google's Deepmind team published a thesis in Nature magazine, "Mastering Go Games Without Human Knowledge". A new version of AI Go game software, AlphaGo Zero, started learning the game from scratch and played with itself without reference to any Go movement sets or human experience. It developed some simple movement sets by itself in 10 hours, some complex ones in 15 hours, and some movement sets that had never been used by any top human players in 55 hours. In 72 hours, the training was completed. Coming out of the reinforcement learning, it beat the AlphaGo Lee version (which had defeated Lee Seedol) 100 to 0, and the more powerful version, AlphaGo Master, 89 to 11.

After the AlphaGo series defeated top human players, the battle in Go games is no longer among the human masters, or between the AI software  and the human players, but among the AI software versions. It's amazing to see that AlphaGo Zero's three days self-learning overpowered not only accumulated human Go game knowledge over thousands of years, but also performance of AlphaGo Lee and AlphaGo Master that both rely on human experience and big data training. This reveals, to a full extent, the limitation of human experience.

Some conclude that the triumphant ascent of AlphaGo Zero in the Go game world has demonstrated that "Algorithm is more important than big data." Of course, this is correct. The essential power of Go game players is algorithm and calculation capacity. 

First of all, the triumph of AI over top human players is a triumph of its algorithm and calculation power. Due to restricted deliberation capacity, human players use local optimal algorithm, considering the current and the follow-up impact of every move on the local circumstances.  AI, with superior computing capacity, may theoretically use the global optimal algorithm, considering the current and the follow-up impact of every move on all circumstances. In most cases, the local optimum is identical to the global optimum. But in a few rare cases, the local optimum is not the global optimum, which is when AI surmounts top human players. 

Secondly, the fact that the self-learning and -training AI with a neural network defeated the AIs relying on human experience and big data training is also a triumph of algorithm. We may say the former is the real global optimal algorithm that avoids being biased by the human experience as it does not refer to any movement maps; whereas the later is simply a sub-global optimal algorithm that is "polluted" by the local optimal algorithm of those movement maps created by humans. That is to say, human experience has put reins on the enhancement of AI's playing power. This is really an embarrassment to the human beings. 

Therefore, it seems the "end of Go game" is not far away. The Go game is played with complete disclosure of all information and the rules are fixed. Theoretically speaking, there exists an "ultimate game", in which every move made by both sides is an optimal move taking into consideration of all circumstances. In other words, any move by any side is considered unwise otherwise. Hence, an "ultimate game" in the end.

The most striking significance of AlphaGo Zero is to serve as enlightment  to our reflections on human experience limitation. It enables us to see clearly the boundary of the validity of experience so that we won't think our own experience is the absolute truth. Human experience is, in fact, essentially local optimal solution. One's personal experience is a local optimum from one's own perspective.

If the reason why human players must adopt local optimum in Go games is due to their limited calculation capacity, people adopt local optimum in daily decision-making mainly because of incomplete and inaccurate information. Games in human society is mostly played on conditions of information asymmetry. Based on limited information, one makes decisions after discerning the refined from the crude and the true from the false. These decisions are the optima on conditions of restrictions, but not necessarily the real optima or the most beneficial ones. However, if one takes into consideration as complete and accurate information as possible, the decision made will be closer to the global optimal solution from an omniscient perspective. 

In addition, one most common mistake people make in decision-making is setting their eyes on the short-term returns at the expense of the long-term returns, on the  "tangible" effects rather than the "intangible" effects, and on the direct "primary impact" instead of the "secondary" and other higher-level impact. This looks very much like resulting from a "limited calculation capacity", but is in fact an "algorithm defect", or an intelligent lethargy. AlphaGo Zero won't make such mistakes. It will definitely take into account all current and follow-up influences of every move. 

Yes, AlphaGo Zero is a mirror reflecting the limitation of human experience.

下面是《财富》（Fortune）杂志2015年发表的一篇文章。两年之后读来，还可以发现许多有趣的观点。 

 《财富》2015年11月1日刊，第102-112页。 

二十一世纪企业：商业的方方面面均将改变 

想象一种没有摩擦的经济 ——   一个劳动力，信息和货币可以轻易地、低成本地、并几乎瞬间地移动的新世界。

嗖 —— 它来了。你的企业准备好了吗？   

（原文）吉奥夫·克文 （Geoff Colvin）

（译文和校对） 陈群（Q. Rene Chen）

汽车突然起火从来不会是件好事。所以当2013年10月，一辆特斯拉 S 型汽车在华盛顿州的肯特撞上金属物体后突然起火，老板、潜在顾客，投资人和公司总管们都忧心忡忡。几个星期之后同样的事件在田纳西州的斯莫纳再次发生，联邦监管部门立案调查。我们都知道接下来会发生什么：大规模的召回，全国各代理商处昂贵的修理，和令汽车制造商痛苦的财政打击。

然而上诉一切均未发生。原来问题出在 S 型车在高速行驶状态下会降低底盘以充分利用空气动力。这时一旦碎物意外击中汽车的电瓶箱，便会着火。所以特斯拉向受影响的车辆传送了一个软件更新，将其在高速行驶状态下的地面高度提高了一英寸。问题消失了。立案调查仅仅四个月后，监管部门就结了案。

使用软件和移动电话网络，特斯拉避免了所需的召回。它也没有任何代理商；顾客可在网上自行配置和订购汽车，并在公司的展示厅内试驾。特斯拉先进的电动车技术比汽油或者柴油车技术更简单，因此生产汽车所需的员工和资本更少。综合这些因素，结果是这样的：通用汽车每一美元物资资产能创造1.85美元的市场价值，而特斯拉则能创造大约11美元。通用每位员工能创造24万美元的市场价值，而特斯拉一位员工能创造290万美元。这样的差异不是仅仅依靠提高效率就可以获得的。特斯拉尽管与通用在同一行业，却是根本不同的概念。

通用在变革，但是眼下它仍然是一家二十世纪企业。特斯拉则被建成一家二十一世纪企业，它所体现的横扫一切的新现实改变了成功的规则。主导的主题便是早被宣布要来到的无摩擦经济的到来。在这个新经济世界里，劳动力、信息和货币可以轻易地、廉价地、几乎瞬间地移动。企业与顾客、雇员和所有者之间正在形成一种全新的、更加变动不定的关系；企业发现它们在拥有的资本量越来越少的同时仍然能够发展，因而正在重新思考（传统定义上）资本的角色；通过创新研发和市场宣传，企业正以新的方式创造价值； 同时它们正用新的尺度衡量业绩，因为传统的标准已经无法反映什么会影响业绩。

不是所有的二十一世纪企业都是金光闪闪的硅谷创业者。它们可以属于任何年龄段，可以从事任何行业（甚至汽车行业）。耐克是一家二十一世纪企业，因为它正在积极地运用3D打印技术发明新式制造方法，并聪明地运用社交媒体进行市场宣传。通用电气正在转型，即使部分原因来自股东们的沮丧情绪和外界的压力。每个企业都必须成为这样的企业。

新的现实从资本主义的基础资本开始。在无摩擦经济中，一家企业远不像过去那样需要很多。看看世界上最有价值的企业，苹果。与价值排名第二和第三的企业谷歌和微软不同，苹果从销售实体产品中获得大部分收入。然而苹果说它 “绝大部分” 的产品是由别人制造的。因为可以协调分布广泛而复杂的全球供应链，苹果可以付钱让其他企业，主要是富士康，来生产自己的产品并按时将它们运到需要去的地方。苹果甚至曾租用其他企业的服务器来建立自己的云服务，以便可以轻易地加减容量，只需为所需要的容量付费。

美国政府将苹果归类为制造商，它在全球拥有大约500家一砖一瓦建造出来的零售店。根据伊娃维度（EVA Dimensions ）咨询公司的数据，它的资本总量数额巨大，为1720亿美元。但是按照传统模式，它需要的资本应该更多。苹果的成就在于将其资本发挥出令人惊奇的效用，以此创造出6390亿美元的市场价值。与之相比，美孚石油使用更多的资本，3040亿美元，才创造了3300亿美元的市场价值，几乎不到苹果的一半。

这些都是生产和销售实体产品的企业。无摩擦经济也赋予实际上没有任何物资资本的企业强大的竞争力，使它们可以与资本雄厚的传统老大们一争高低。人们惊奇地注意到阿里巴巴是全球最有价值的零售商却没有任何自己的库存，爱彼迎是世界最大的旅馆餐饮供应商自己却不拥有任何房地产，而优步是世界最大的出租车服务商却不拥有任何汽车。每家企业都找到了天才的方式消除行业内的摩擦，将买家和卖家直接而方便地联系在一起，使新型的、近乎无需资本的商业模式成为可能。

但是，等一下 —— 实际上，这些以及所有二十一世纪企业都拥有巨额资本。只不过会计规则不这样称呼它罢了。那便是以软件、专利、 版权、品牌和知识形式存在的知识产权资本；以买家关系存在的客户资本；特别是人才资本。即使是制造或销售实体产品的二十一世纪企业也首先是一家人才资本型的企业。这便提出了一个深刻的议题：谁是它真正的拥有者？

许久以前的情况很清楚，法律事务所几乎完全由人才资本构成，因此它们向公众发行股票是违法的；事务所之外的股票持有人无法拥有任何价值。咨询公司和广告公司有什么不同吗？即便是拥有价值很高的发明专利或者品牌的企业同样仍然要依赖人才资本获得大部分的价值。如果数以百计的最聪明的人离开星巴克或者强生或者迪斯尼，如果一位疯狂的CEO试图摧毁每个公司如钛合金般强大的文化，情况会怎样？不论承认与否，在一家二十一世纪企业，雇员拥有大部分的资产因为他们本身就是企业资产的大部分。

这个事实正在影响企业的结构。美国国税局（使用最新公布的数据）指出美国的法人制公司的数量只有中度增长而它们的收入在 1990 到 2008 年间增加了150%。与此同时，经理人拥有的所有权制和合伙人制公司的数量增长远远超出，它们的收入增长了394% 。二十一世纪企业不仅仅包括法人公司。

大部分商家将必须以新的方式创造价值，通常是依赖互联网的商业模式。否则将会输给那些这样做的竞争者。这一趋势从互联网诞生之初便已经出现，当时许多网上保险暴发户迫使定期人寿的预付金一下子降低了50%还多；还有方便顾客的旅馆和航空公司订票网站几乎一夜之间令一万八千多家旅行社关门停业。现在，企业家们正在将这一趋势以复杂的方式扩大到实体产品行业。Warby Parker 用一种低摩擦网上模式以仅为传统零售商收费一小部分的费用销售高质量的眼镜；私人投资者给这家公司的最新估值为十二亿美元。甚至看上去对网上侵入具有高度绝缘性的行业，大众日常消费品，也受到了威胁。哈里和一元剃须俱乐部在网上制造和出售男子美容产品，正迫使宝洁拥有的吉列品牌第一次以价值而不是质量来宣传自己的产品。

对那些稳居商品目录领先地位的企业来说这一趋势甚至尤为可怕，因为即便他们也转而启用新的低摩擦商业模式，结果仍然会规模较前缩水，利润较前减少。麦肯锡全球研究所（McKinsey Global Institute, MGI）指出，那是因为 “一些技术和依赖技术的公司为被挑战者们摧毁的价值高于他们为自己创造的价值，许多收益在竞争中以消费者剩余的形式流失。” 例如，微软的 Skype 服务2013年带来了二十亿美元的收入，然而麦肯锡计算出同一年  Skype 通过向消费者提供免费和低成本的通话而从老式通讯公司向消费者转移了370亿美元的利益。

其他的新型商业模式情况类似。旧金山的出租车监管部门报告随着优步、来福车和其他公司进入市场，每辆有执照出租车的载客次数在2012年3月到2014年7月之间减少了65%。优步 —— 参看我们本期的相关报道 —— 目前的投资人估值为510亿美元。同时，据《财富》杂志九月的报道，每块纽约市出租车注册牌照的累积市场价值已低于130亿美元。

波士顿大学的研究者说，当爱彼迎进入奥斯汀时，旅馆收入降低了8%到10%。“受影响的旅馆以降价来回应，结果对所有的消费者有利，而不只是分享经济的参与人。” 然而，引发混乱的新企业收获的只是被挑战企业失去的利益的一部分。

二十一世纪企业将愈来愈趋向于一种基于创意的商业，不仅在信息技术行业如此，同时在媒体、金融、制药和其他消耗大量脑力的行业亦如此。麦肯锡发现 “轻资产重创意的领域” 在1999年创造了西方企业17%的利润，而如今它们创造的利润占31%。这一转变过程中的输家是资本和劳动力密集型的行业，比如建筑、交通、水电和采矿业。这并不是说这些行业中的企业注定灭亡。正如特斯拉所显示的那样，一旦重新发挥想象力，他们也能成功。

或者说，如果他们重新定义成功的话，也能成功。各类企业面临的压力越来越多地来自于那些为了发展宁愿牺牲利润的竞争者们的发展。它们通常是在新兴市场中已取得较大规模的家族所有或者国有制企业。例如，美国铝业公司（Alcoa）最近决定分裂为两家企业，一家高科技材料公司和一家铝产品制造公司。导致这一决定的部分原因在于巨型中国铝冶炼厂所获得的价格优势。被迫与之竞争的美铝商品产业拖累了整个公司。随着新兴市场企业全球商业份额的增加 —— 目前它们已占了财富全球500 强的30%左右 —— 利润压力还将增加。

进一步的压力还会来自另外一类牺牲利润换取发展的二十一世纪企业，就是那些看到新模式商业中广阔天地有待开发的企业。可供展示的头号范例便是亚马逊，它因为一个季度接一个季度地报告很少甚至零利润而著名。投资人与首席行政官杰夫·贝佐斯（Jeff Bezos）意见一致地认为钱投资于扩展会更好，带来的未来利润将会高出许多。亚马逊的股票最近达到了历史最高水平。

如何为新品种估值

少量的硬件资产，说不上任何产品，大量的知识产权资本。投资者该如何对待这样的企业？

所谓的独角兽企业，就是那种成长迅速、价值超过十亿美元的初创企业如此受追捧，以至于似乎没有必要问一问，“它们的真正价值与它们的估值沾边吗？”

仍处于发展阶段的学生贷款供应平台 SoFi 最近以超过三十亿美元的估值募集到十亿美元的股权资金。脸书在获得真正有意义的收入之前曾支付190亿美元收购私募股权持有的通信工具瓦次普（WhatsApp）。还没有申报一分钱利润的优步在最近一轮私募市场融资中的估值为510亿美元。无论算不算扰乱，是否它们的支持者所看到的，而那些理性投资世界中以未来收益的当前价值评估企业的人没有看到？

说实在的，这无关紧要。私募市场估值向来如此，是由很少的投资者在公众监督之外进行的评估。公募企业的估值在市场开放的每一天都在被检验。私募企业的估值只有在它们筹集更多资金的时候才会改变。科斯拉风险投资公司（Khosla Ventures）的风险投资人基斯·拉博伊斯（Keith Rabois）说，实际上“私募市场的估值更像一门艺术而不是一门科学。” 它们是一种谈判，风险投资人会问，“在什么价位一个不需要我的钱的人会接受我的钱？”

风投支持型企业是指那些按历史惯例本该已经转到公募市场的时间点过去之后仍然保持私募的公司。它们的繁荣有几大因素。低利率使得资本成本低廉，并鼓励机构投资人在风险更高的博弈中寻求最高回报。如此大量的现金供应又转而令私募性质企业不急于寻求公募市场的光环。结果就是它们可以在更高的估值基础上募集到更多的资金。

有时，这些令人目瞪口呆的估值并不像它们表面上看起来的那样。例如，为了取得亿万美元的地位 —— 这可以带来积极的公众宣传，招募人才时夸夸其谈的权利， 和企业家的自我满足 —— 一些公司会为了更高的书面价值而接受一些苛刻的约束条件。比如，支付软件供应商 Square 以六十亿美元的估值募集到资金，但同时向投资人承诺如果首次公开募股价格低于约定的水平将导致额外股份的发行。那样将稀释其他持股人的利益 —— 并可能损害企业的估值。

当然，一旦公司最终转为公募，所有这些幕后操作终将结束。曾经有过一些技术公司，比如 Box 和Hortonworks，便无法维持超过先前私募估值的公募价值。毕竟，独角兽作为一种神话动物，真正的价值在于从来没有人见过。而公募企业则无处可藏。这个新群体也许会继续创新，但最终它们的价值同样不过是长期内能为股票持有人赚到的钱。—— 亚当·拉辛斯基（Adam Lashinsky）

二十一世纪企业的雇员需要重新深入思考的话题应该围绕这样一个问题：劳动力市场在变得无摩擦之后会发生什么？正如零工经济的崛起所显示的那样，很明显情况正朝着无摩擦方向演变。企业仍然雇佣全职员工，虽然并不全职地需要他们，但维持员工比不断地雇佣和解雇员工更容易。至少过去是这样。现在雇主们正在通过类似 Upwork 的网上劳力市场在全世界范围内雇佣成百万的工人来做以信息为基础的工作；以前的雇主们会对每一位工人进行评价，并且在你对工作满意之前不需要支付任何报酬。许多都是日常化的工作，比如语言翻译；一个名为书呆子小时工（HourlyNerd）的劳力市场出租前咨询顾问和顶级商科毕业生来帮助完成策略规划、金融分析和其他高等级任务；顾客大多是中小企业，但也包括像通用电气和微软这样的巨型企业。

将这一趋势再前瞻几步的话，会发现整个雇佣模式有可能发生根本性改变。雇员拥有的企业很可能数量增加，但这只是诸多选项之一。最终，这些选项也许会包括一种比雇员拥有激进许多的企业结构。前思科（Cisco）首席行政官约翰·钱伯斯（John Chambers）六月份时曾说 “很快你们将看到只有两名雇员的巨型公司 —— 一位首席行政官（CEO）和一位首席信息官（CIO）。” 听起来很疯狂，不过钱伯斯曾有发表的疯狂预测（比如用智能手机开旅馆房门）最终变为现实的先例。

那些继续雇用大量员工的雇主也可能改变雇佣双方的关系。一位业绩评估软件制作商的行政官最近告诉《财富》杂志，“现在，几乎任何工作的结果都可以测量，所以你可以依此回报人们。” 结果是表现优异者获得的酬劳更多，其余的酬劳变少。恰安·翰威特的报告中说实际上现在所有大主雇都在常规薪酬之外给员工发奖金，通常是因为达到了某些特定的阶段目标，比如收取了更多的欠款或者依据其他那些如今很容易追踪的业绩标准。当个人表现测量起来很麻烦的时候，薪酬的差别就小，表现不佳者仍能保留职位。以后不会了。

员工的真实便是二十一世纪企业自身的真实。随着新兴市场摩擦的消失和野心勃勃的新竞争对手的出现，表现不佳的企业将无处藏身。赢家会赢得更多，其余的将为获得剩余的利益而更加激烈地竞争。麦肯锡全球研究所一项新的研究表明，重视想法的领域 “正在形成一种赢家获得一切的风气，利润最高的公司与其余公司之间的利润差别巨大。” 更笼统地说，竞争只会更加激烈。麦肯锡的报告指出跨国公司的利润总量最近达到了世界各国国内生产总值的10%。有一天我们会带着嫉妒心理回顾这个数字; 麦肯锡预测利润占比到2025年将缩小至8%，几乎不到1980的数字。结果：“随着利润增长的减缓，更多的企业将为分得更小的一块蛋糕而战。”

在这个世界中，企业虽然更加努力地奋斗，获得的全球经济产出份额却在减少 —— 看起来像一个在世界事务中影响力不断减弱的配方。然而对于二十一世纪企业来说，情况将恰恰相反。一个新现象是一些企业的规模正在超越国家。谷歌每天进行数以十亿计的搜索，它所拥有的对世界正在发生什么的实时了解比任何政府都多；研究表明，它能够预测疾病的爆发，股票市场的运动，还有其他许多。如果它愿意的话，甚至可以影响选举。脸书拥有比中国人口还多的十五亿用户，还可以准确地描述其用户的性格特征，并预测他们在工作和恋爱中的成败。苹果每一天经手的现金量大概要超过美国财政部。印度一家通讯公司巴提·爱尔特尔（Bharti Airtel）的顾客数量大概与美国居民人数相当。沃尔玛雇佣220万员工。在地球上它的员工人数仅次于美国和中国的国防部门。

现在再说说另一个令人费解的二十一世纪概念：虽然有些企业取得了巨大无比的规模，但平均而言企业的寿命将比过去缩短。这一趋势令人震惊：耶鲁大学的理查德·福斯特（Richard Foster）指出，标普500指数中企业的平均寿命从1958年的61年下降到现在的20年。他预测下降趋势将继续稳定持续下去。甚至远在二十一世纪结束之前，企业作为一种持续存在的机构这一概念本身都可能不再是常规。

毕竟，企业为什么会存在？英国经济学家罗纳德·科斯（Ronald Coase）因为回答了这个问题而获得过诺贝尔经济学奖。在理论的世界里，全球经济像一个依赖个体运作者之间的价格信号而运转的顶盖，对大型企业没有明显的需求。但是在真实的世界里，正如科斯指出的那样，“为了解决争端会有谈判发生，会起草合同，监督检查必须进行，协定安排必须达成，等等。” 也就是说，交易有成本 —— 摩擦 —— 而在企业内部整合交易是处理它们的最有效方式。现在，随着技术使成本降低，许多企业在分解自身，外包一些功能给他人，众包研发功能，用合同工代替员工。好莱坞一直以来的模式，即人员和资源为取得一个目标走到一起，之后又四散到其他项目当中，也许会成为整个经济的普遍模式。这已经在发生。

好消息是加速的变化、创造性的摧毁和新的商业模式对于具有冒险精神的人来说都是机会。随着经济的转型，一个统一的主题是，在几乎所有的商业活动中，进入的门槛都在降低。机会的大门前所未有地敞开。每个人和每个机构都可以拥有二十一世纪最有价值的资产：对新创意、天才和想象力的开放和拥抱。 

2018.09.12

新阿尔法狗照见了人类经验的局限性

作者：陈斌
最后更新：2017-10-30 16:05:34
来源：南周评论（本文首发于2017年10月26日《南方周末》）

2017年10月19日，谷歌Deepmind团队在《自然》刊发论文《精通围棋对弈无需借助人类知识》。一款新的人工智能AlphaGo Zero，从零开始学习，自己与自己对弈，不借助任何围棋棋谱与人类经验，10小时左右发现简单定式；15小时发现复杂定式；55小时左右发现人类顶尖高手从未用过的新定式；72小时出关。出关后以100∶0完败AlphaGo Lee（曾战胜李世石），以89∶11大胜棋力更强的AlphaGo Master。

在AlphaGo系列版本连败人类顶尖高手之后，围棋对弈的看头已不在人类顶尖高手之间，也不在人工智能与人类顶尖高手之间，只在人工智能与人工智能之间了。AlphaGo Zero三天的自我学习，超越了人类几千年对围棋知识的积累，也超越了借助于人类经验与大数据训练的AlphaGo Lee及Master，让人叹为观止，充分揭示了人类经验的局限性。

有人说，AlphaGo Zero横空出世，说明“算法比大数据更重要”。这当然是对的。棋力的本质就是算法与算力。

首先，人工智能战胜人类顶尖高手，是算法与算力的胜利。由于算力的限制，人类棋手采用的是局部最优化算法，每一落子，主要考虑对周边局面当下与后续的影响。算力优裕的人工智能，原则上可采用全局最优化算法，每一落子，考虑对全局当下与后续的全部影响。在绝大多数情况下，局部最优与全局最优的落子是一致的。但极少数情况下，局部最优就不是全局最优，人工智能就碾压人类顶尖高手了。

其次，用一张神经网络自我学习与训练的人工智能，战胜借助于人类经验与大数据训练的人工智能，也是算法的胜利。看来，前者才是真全局最优化算法，因不看棋谱而免于被人类经验带偏；后者只是准全局最优化算法，算法受到了人类棋谱中的局部最优化算法的“污染”，也就是说，人类经验束缚了人工智能棋力的提升。这对人类来说真是一个尴尬的事实。

由此，“围棋的终结”那一天似乎也不远了。围棋对弈是在完全信息下进行的，规则也很明确，理论上存在着一盘“终极棋局”：对弈双方下的每一步，都是全局意义上的最优落子，也就是说，任何一方的任何一步不这样落子是不明智的，最后就形成了“终极棋局”。

AlphaGo Zero的最大意义，在于有助于我们反思人类经验的局限性，让我们清醒认识到经验有效性的边界在哪里，不至于以为自己的经验是绝对真理。人类经验的本质其实就是局部最优解。你的个人经验就是你个人视角下的局部最优解。

如果说在围棋对弈中人类棋手不得不采用局部最优化算法的原因是算力限制，那么在日常决策中人们采用局部最优化算法的主要原因在于信息不充分、不准确。人类社会中的博弈基本都是信息不对称下的博弈。你在对自己掌握的有限信息去粗取精、去伪存真之后做出的决策，是局限条件下最优的决策，但并不一定等于真正对你最优、最有利的决策。不过，要是你占有的信息尽可能全面、尽可能精确，那你的决策就会更接近于上帝全知视角下的全局最优解。

此外，人们在决策中常犯的错误是只考虑短期效应，而忽视长期效应；只看到“看得见的”，而忽视“看不见的”；只看到直接的“一阶效应”，而忽视“二阶效应”等高阶效应。这看起来也挺像“算力限制”导致的，但其实是一种“算法缺陷”，是智力上的懒惰。AlphaGo Zero就不会犯这样的错误，一定会考虑落子对当下与后续的全部影响。

是的，AlphaGo Zero就是一面镜子，照见了人类经验的局限性。