Bottom-up

Jerry Brito’s new podcast is even better when I’m not the guest. His latest episode features Michael Sawyer, of the Berkeley Center for Law and Technology, who has written about the User-Generated Content Principles that were negotiated between copyright holders and UGC sites a couple of years ago. Sawyer argues that the filtering technologies that the principles require UGC sites to employ give short shrift to fair use, and he proposes reforms whereby users could contest fair use determinations.

I agreed with much of what Sawyer had to say, but one place where I didn’t agree is when he characterized it as a “huge burden” for content companies to police UGC sites for infringing content. I don’t doubt that such policing costs a lot of money, but I also think we need to clearly understand the economics here. Content providers have a virtually unlimited number of options when it comes to the ways they might earn revenue from their works. Because of the particular economic and technological environment of the 20th century, the dominant paradigm in a lot of content industries is to sell copies of the works. But there are other models, including blanket licensing, ad-supported content distribution, and the sale of complementary goods such as concerts or popcorn.

One of the factors any business considers is the costs of enforcing their rights. If you’re opening a diamond shop, you need to consider the costs of locks, surveillance cameras, security guards, bars on the windows, and so forth. If you have a choice between a high-crime neighborhood and a low-crime one, you have to weigh the lower rent in the high-crime neighborhood against the higher costs of securing your store there. True, the police and your landlord should do what they can to help prevent crime, but the bulk of the costs almost always falls on the business owner himself. This is a good thing, because it encourages entrepreneurs to consider these costs when deciding where and how to run their businesses.

The same principle applies in content industries. If Business Model A produces $10 million in revenue but requires $5 million in enforcement activities, while Business Model B produces $8 million in revenue and requires only $1 million to be spent on enforcement, then the company should employ business model B. Yet if the law shifts enforcement costs onto a third party, then the company is going to pursue business model A, even though doing so inflicts larger costs on third parties than the increased revenue enjoyed by the content holders.

So if it’s true that policing user-generated content sites is so expensive that content-creating companies can’t afford to do it, then the right response is for content companies to change their business strategy. The most obvious approach is to sign an agreement with the UGC sites that allows their content on the site in exchange for a cut of ad revenue. The beauty of this approach is that it makes the fair use issue go away entirely. If content-detection software is merely used to determine the size of payments, rather than for filtering, then it doesn’t really matter if the software can detect which uses are fair.

Now, I’m not saying this is the business model content companies should pursue, and I’m certainly not suggesting that Congress impose a compulsory license to this effect. My point is simply that it’s not a bad thing that copyright holders bear most of the costs enforcing their own copyrights. Putting the costs of the copyright system on the same parties who receives the benefits gives those parties the right incentive to choose an economically efficient business model. If they decide that monitoring, threatening, and suing millions of users is the right way to run their business, that’s their choice—but they’re not entitled to any sympathy for the costs this business model imposes on them.

One thing I am reasonably sure of is that if you’re hiring human beings to make fair-use determinations on millions of individual videos, you’re doing it wrong. Our focus shouldn’t be on figuring out who should pay for human review, it should be on finding approaches that don’t require such a pointless waste of labor.

Yesterday I attended a panel here at Princeton featuring three and a half right-of-center thinkers: Daniel Larison, Virginia Postrel, Ross Douthat, and David Frum. I say three and a half because Postrel warmed my heart by beginning her comments with the disclaimer that she was a liberal and wasn’t particularly interested in the vitality of conservatism.

Princeton did a fantastic job of choosing representatives of four different strands of right-of-center thought. Douthat, along with my friend Reihan Salam, penned a book arguing that conservatives should make their peace with the welfare state and focus on building a populist brand of “Sam’s club conservatism.” He reprised these themes in his talk. Frum has a reformist agenda of his own, focused on re-building the conservative movement on the twin pillars of free markets and an activist foreign policy, while sanding off some of the rough edges that have turned off more educated voters.

Larison focused on foreign policy, arguing that the Republican Party’s political failure flowed from its failure to recognize that conservative admonitions about the dangers of big government apply to issues of foreign policy and executive power. Larison writes one of my favorite blogs, and I wish his perspective were more widely shared within the Republican movement. He hails from a faction of the conservative movement that has an illustrious past but declined precipitously during the Cold War. And I wish it were true that a more restrained foreign policy and stronger commitment to limits on executive power were the key to Republican success. But I’m not sure there’s much evidence for this and Larison didn’t entirely persuade me. The last two serious non-interventionist candidates for the Republican nomination were Pat Buchannan and Ron Paul, and neither of them came close to winning their party’s nominations, to say nothing of the presidency.

Buchannan was a target of criticism in Postrel’s first book, The Future and Its Enemies, which is in some ways an intellectual forerunner of this blog. Her vision for what she calls a “dynamist” political movement overlaps in important ways with the bottom-up perspective I’ve been trying to articulate her. In her talk here, she argued that the Republican Party’s decline is linked to its failure to appeal to voters with the dynamist, bottom-up worldview she has championed in her writing. But she expressed doubt that the political coalition she envisions can emerge from a Republican Party that is wedded to the demagoguery of Rush Limbaugh and Glen Beck.

Although I’m always wary of the pundit’s fallacy, I do think one of the keys to Obama’s success in 2008 was his ability connect with voters who have a dynamist, bottom-up worldview. These more educated and affluent voters are a growing share of the electorate, and the GOP would be well advised to take their concerns more seriously. Certainly not all affluent, educated voters share her views, but I think many of them will find her arguments resonate more than those of the other panelists.

There’s an important link between my app store and network management posts from last week: they’re both examples of filtering processes that don’t scale. There’s a fundamental difference between an recommendation process, in which a variety of firms offer competing recommendations to users, and a filtering process in which a single firm unilaterally prevents users from accessing certain content or applications. There’s a sense in which these are both exercises of “editorial discretion,” but as the set of items being reviewed increases, they behave in a divergent fashion. Recommendation processes scale gracefully, because there’s room for lots of competing recommendations, with each supplementing the failures of the others. One firm’s failure to include a particular item on its list doesn’t preclude others from correcting the oversight. The system as a whole works better than any one of the participating firms.

In contrast, top-down filtering scales poorly because a single firm has to make a correct decision about every item. No single firm can possibly understand every item in a complex technological ecosystem, and the constraints of bureaucratic management make it unlikely that they’ll even do an especially good job of it. So top-down filtering will lead to mistakes, and those mistakes will directly harm users, who can’t easily choose a different filter.

Indeed, I suspect that an ISP’s effort to behave as an “editorial filter” in a significant way would fail so swiftly and spectacularly that the experiment wouldn’t be repeated. The app store fiasco hasn’t sunk the iPhone because the iPhone is still a relatively simple platform, and because iPhone users have never known an alternative. The Internet is a vastly more complex ecosystem, and its users are used to being able to run the applications of their choice. If, for example, Verizon or Comcast tried to create a whitelist of approved applications and block the rest, they’d make so many mistakes that a large fraction of their customers would revolt.

Broadband ISPs can’t control content or applications on the Internet any more than Microsoft could impose an “app store”-style review process on Windows developers: the ecosystem has become far too complex for any one firm to control, and trying to exert that kind of control would destroy most of the value embedded in the platform. This isn’t to say that they won’t engage in small-scale filtering (like the Comcast/BitTorrent incident), but efforts at more comprehensive censorship aren’t likely to work very well.

A couple of people left interesting comments in response to mypost about the problems with “app stores.” Here’s my former colleague Adam Thierer:

There needs to be some sense of proportionality here, at least about the iPhone (I can’t speak for the Palm experience). In just a little over a year, there’s been 2 billion downloads of over 85,000 apps from over 125,000 developers.

So, when you talk about Apple’s approval process being “plagued by.. problems” and “rejections for trivial or non-sensical reasons” and “long delays in the review process have become a staple of the tech blogosphere” I think you are giving the impression that this is somehow the norm when it is very much the exception to the rule.

I responded in the comments, but I liked Aaron Massey’s comments better:

Adam, I don’t know that the numbers you’ve given are useful metrics in this context. I’m willing to concede that the iPhone is a wildly successful product and that the percentage of applications being rejected for spurious reasons is tiny, even taking into account the points Tim made about developers having a strong incentive to remain silent. The problem is that these numbers don’t prove that the iPhone wouldn’t be an even better platform if it were more open; it only proves that it is a very successful product as it is.

He goes on to describe one approach Apple could take to address the app store’s flaws:

Apple could try to get (some of) the best of both the open and the closed worlds by having different classifications for applications. I’m thinking of something similar to the “components” system that Ubuntu uses. This means they could have a set of applications that are approved in whatever process they want to use and as many sets of ‘unapproved’ or ‘experimental’ applications.

Apple could even set it up so that users wanting to use unapproved applications would have to disclaim liability for the security of their data or the stability of their system. This seems reasonable to me. If you give users the power to approve of and install their own applications, then they should take personal responsibility for the consequences, if any.

Those are just a couple suggestions. I don’t have a problem with Apple doing whatever they want — it’s their product, but then, it doesn’t look like Tim has a problem with that either. He’s just arguing that the top-down approach to the App Store is not as good as a bottom-up approach would be.

Couldn’t have put it better myself.

Here’s one developer’s perspective on Apple’s treatment of developers, from this summer:

The last session of WWDC ‘09 yesterday was about publishing on the App Store. The content of sessions is under NDA, so I can’t tell you what it was about. So I’ll tell you what wasn’t in it: the audience Q&A session that succeeded nearly every other WWDC session and usually provided invaluable access to Apple employees and useful additional knowledge to attendees. The session itself blew through its lightweight examples quickly, ending 45 minutes early. The majority of the audience was clearly there for the Q&A. As people lined up at the microphones around the room, the presenter abruptly showed a simple slide with only “WWDC” in plain lettering, thanked us for coming, and bolted off the stage. The Apple engineers, usually staying around the stage for one-on-one questions, were gone. The lights came up instantly, and it was the only session that didn’t end in music. The audience was stunned.

It was a giant middle finger to iPhone developers. And that’s the closing impression that Apple gave us for WWDC. Clearly, they had absolutely no interest in fielding even a single question from the topic that we have the most questions about…

In talking to many Apple employees this week, it’s clear that the hostility and inaccessibility is not generally a problem with most individual employees or any sort of universal culture inside the company. In fact, nearly everyone from Apple that I spoke with was helpful, friendly, and — most importantly — human. These were just regular people who deal with hard problems and need to keep a lot of secrets.

The problematic policies and attitudes, clearly, are enforced from higher up in the company.

To some extent, this is comforting: most of an app developer’s risky decisions (mainly rejections) are made by the lower-level people who seem well-intentioned.

I’ve also learned that many of the problems aren’t intentional: many parts of Apple’s internal infrastructure are overloaded or extremely outdated, and they’re scrambling to keep up with their growth, often ineffectively.

One major surprise was when an Apple employee told me that the App Store reviewers are not outsourced: they’re all direct, full-time Apple employees. Many of them work nights and weekends to keep up with the extremely high submission volume, and they’re constantly expanding the staff to keep up with the submission growth.

This seems to me like the unsurprising result of trying to impose bureaucratic control on a large and otherwise decentralized system. Look at any understaffed IT department in a large organization, and you’ll see the same dynamic: users perceive indifference and sloppiness, while administrators feel like they’re rushing to keep up with a firehose of requests.

My colleague Jim Harper has written about the problems that are likely to result if Congress is ever stupid enough to impose electronic employment eligibility rules on private employers: federal workers will be overwhelmed with complaints from employees—many of whom are victims of paperwork errors rather than actual illegal immigrants. From the employees’ perspective, the government bureaucrats will seem lazy and callous, while the bureaucrats will feel they’re struggling heroically to keep up.

The fundamental problem is that top-down systems scale poorly. Decision makers—whether in government or in the private sector—need to be mindful of the risk that what they perceive as reasonable oversight will become a bottleneck that creates an impression of arrogance and indifference to their concerns.

A friend recently emailed me to suggest that this blog may feature longer-than-optimal posts: that when people see an 800-word post they tend to get intimidated and skip on to the next blog. This seems like a reasonable criticism, and I’m curious what other people think. Would you be more likely to read my posts if they were shorter? Or do you prefer the longer-than-average posting style I’ve used thus far? I’d be happy to shift my median post length downward if that’s what readers prefer, so please don’t be shy if you prefer that option.

Until recently, if you bought an iPhone (which I did) or a Pre, your only officially-permitted mechanism for getting applications onto your phones was the iTunes App Store or the Palm App Catalog, respectively. This is strange: most modern software platforms have open executable formats that allow developers to distribute their software directly to end-users. So, for example, you don’t need to get Apple’s approval to release a new application for Mac OS X. But iPhone and Pre users don’t enjoy that freedom. And the result, I think, is a great illustration of the problems that top-down systems can create.

An app store sounds like a good idea in principle. Users want applications, and app stores give them a convenient way to get them. And Apple and Palm are understandably concerned about quality control. But the problems start when the app store is made the sole distribution channel for applications. One developer recently recounted his months-long ordeal seeking approval for a Pre application. This was an application he developed by himself, in his free time, and was seeking to give away for free. Three months and dozens of emails later, the application in question still isn’t available to Palm users. Apple’s approval process is plagued by the same kind of problems: rejections for trivial or non-sensical reasons and long delays in the review process have become a staple of the tech blogosphere.

The striking thing about these incidents is that they don’t seem to be the behavior of a rational, profit-maximizing firm. Some of the rejections could be plausibly described as strategic business decisions, and others may be legitimate quality control, but we’ve seen a number of examples that can only be described as pure stupidity. It seems unlikely that the management of Apple or Palm deliberately set out to make the review process slow, unpredictable, and frustrating for developers. The value of a platform is greatly enhanced by third-party applications. These companies should be bending over backwards to make the review process as simple and painless as possible.

I think the problems with the Pre and iPhone approval processes reflects the inevitable inefficiencies of top-down filtering. Reviewing applications is a complex, labor-intensive process. As these platforms have grown, the number of people required to perform the necessary review has grown as well. And as the number of reviewers grows, it gets harder and harder to make the process both fair and efficient. Different applications will be given to different reviewers, who will interpret the rules (which may not be particularly clear to start with) differently. The overhead required to keep all reviewers on the same page grows faster than the number of reviewers, so as the platform gets more popular the review process itself becomes a major bottleneck for the platform’s continued growth.

This is an old story in the technology industry. In the 1990s, for example, we saw a competition between the open Internet and a variety of closed online services like AOL and CompuServe. The open Internet won because it scaled better. Because no one had to ask permission to offer a new Internet service, the quantity of interesting content and applications on the open Internet grew at a rate no one firm could hope to compete with. Top-down systems scale poorly, and so making a top-down system a bottleneck for your platform is a long-term recipe for stagnation.

This week Palm announced some changes to its approval process that sound like steps in the right direction. Although the details are still sketchy, it appears that application developers will soon have the option to release their software directly to the general public, bypassing Palm’s approval process. If you want to get your application into Palm’s app catalog, you still have to submit to Palm’s review process, but review won’t be a pre-condition for distribution.

Making review optional nicely addresses the scalability problems with the app store concept. Mandatory review of every app means that Palm wastes resources reviewing applications whose developers may not need it or may lack the time and resources to fully participate. In contrast, optional inclusion in the app catalog allows developers to self-select into the program if they think they will benefit. And this, in turn, allows Palm to focus its resources on reviewing and approving the cream of the crop. Most important, it removes the review process as a bottleneck for platform growth. The option to route around the app catalog acts as a sort of safety valve, ensuring that the inadequacies of the review process won’t bring down the whole platform.

The more fundamental point here is that people tend to underestimate the extent to which scalability problems limit the effectiveness of top-down social systems. I’m sure that the app catalog looked like a great idea on some Palm executive’s powerpoint slide. In the abstract, it’s easy to imagine that a review process can be made quick and painless. On a real-world platform with thousands of developers, the practical realities of bureaucratic management become binding constraints. Policies that seem reasonable on paper may turn out to be unworkable in practice because they require too much coordination between different parts of the bureaucracy. But the specific difficulties are hard to predict in advance, since they are precisely the kind of implementation details that an executive’s powerpoint slide is designed to ignore. And so decision-makers are prone to systematically underestimate the difficulties of top-down management.

This is why in many cases, including this one, the right solution is to avoid creating bottlenecks in the first place. Firms should recognize that successful technology platforms are too complex to be controlled effectively by any one firm. If they want their platforms to flourish, they need to decentralize decision-making, giving their users ultimate authority over which applications they use. “App stores” can be a useful way to help users find the best applications, but when app store approval becomes mandatory, it becomes a major impediment to the success of high-tech platforms.

My post last week about immigration as a civil rights issue generated a fair amount of discussion. There were a number of critical comments here and at Will Wilkinson’s blog. One objection was to my use of the term “civil rights.” I think the term fits, but if it makes my argument clearer I’m happy to stick with the more general term “human rights.”

When I argue that immigration is a matter of human rights, my point isn’t that a nation is morally obligated to give immigrants all the rights and privileges that American citizens enjoy. There is a variety of plausible reasons a nation might have for regulating the movement of persons, and there are obviously some benefits, including the right to vote, that can be properly reserved to citizens or to permanent residents. I was making a narrower point: that immigrants, legal or otherwise, have rights that the US government is morally obligated to respect. And that there’s something deeply offensive about the premise, reflected in a lot of mainstream discussion of immigration issues, that the rights and welfare of immigrants is irrelevant, or close to it, in thinking about immigration policy.

Consider this sentiment by self-described liberal Tod Robberson, discussing the DREAM Act, which would provide green cards to kids whose parents brought them here illegally while they were minors:

I want to be fair to kids whose only crime was to accompany their parents to the United States and who, in many cases, know no other country but this one. But at the same time, I’m not sure it’s a smart move to open the way for potentially millions of college-educated illegal immigrants to start competing for precious jobs against college-educated American citizens.

I don’t want to get into a semantic debate about whether this perspective is “racist” or not. But I think it demonstrates a kind of crude parochialism that’s morally objectionable for some of the the same reasons that racism is problematic. The kids whose lives would be transformed by the DREAM Act are, we should remember, Americans in every relevant sense. Here are the perspectives of a few of them:

My name is Blanka. I’m 23 years old and I was born in Croatia and currently live in Illinois. I came here when I was ten on a tourist visa and overstayed it without my knowledge. I have a bachelor of science in finance and am applying to graduate school to study applied statistics. I love sports, running (I just ran my first marathon), volunteering, piano, and many other things. I feel the U.S. is my home and I want to give back to a country that has given so much to me.

My name is Mario and I am a sophomore studying mathematics and economics at UCLA. I moved to the U.S. from Guatemala at the age of nine to reunite with my mother after two years of not seeing her, only to once again be apart from her due to her job as a full-time nanny for another family. Even as a single mother, she has dedicated all of her time and effort to trying to make my life easier and better than how we lived in Guatemala. Since I cannot legally work in order to support us financially, I’ve contributed to our struggle by empowering myself with an education in order to support both of us when she no longer can work. I was valedictorian in my high school and the first one in my family to go to a university, but as an undocumented student, these accomplishments are somewhat meaningless to me as I know I have to fight harder in order to accomplish my long-term goals.

Robberson is “not sure” it’s “smart” to extend basic freedoms to these kids because doing so might inconvenience native-born Americans. I suspect his empirical claim is wrong—that the DREAM Act would not significantly reduce the opportunities available to American college kids. But I also don’t really care. I think it’s wrong to condemn a 23-year-old to a life of poverty and uncertainty in the country he calls home. And I don’t think it matters whether recognizing these kids’ right to earn an honest living will depress the wages of native-born Americans, any more than it mattered whether ending discrimination against blacks made it harder for whites to find jobs.

After reading all the comments, both here and in response to Will Wilkinson’s post, I’m still waiting for a convincing explanation of why it’s more acceptable to deny people basic rights on the basis of national origin than on the basis of gender, race, or sexual orientation. The most common response seems to be that we’re justified in violating the human rights of illegal immigrants because a majority of Americans have voted to do so. But few Americans would accept that as a rationale for discriminating against other disfavored groups. I agree with Thomas Jefferson that all men are created equal and endowed with inalienable rights. If that’s “fanatic, indiscriminate, or dogmatic,” (to quote a friend who commented on my last post) then I think I’m in good company.

In my Cato paper on network neutrality, I criticized Penn law professor Chris Yoo’s claim that network providers need to “exercise editorial control” over their networks. I see that Yoo has now expanded this claim into a whole paper, and although I think he makes some good points, I remain skeptical after reading the paper-length treatment. Here’s the core of his argument:

When the Internet first arose, it served primarily as a medium for person-to-person communications, such as e-mail and file transfers. Although some forms of mass communications, such as newsgroups and electronic bulletin boards, did exist on the early Internet, they represented a relatively small proportion of overall Internet traffic. The nature of the Internet underwent a fundamental change during the mid-1990s. The privatization of the Internet backbone and the concomitant elimination of the commercialization restrictions triggered an explosion of mass media web content.

The emergence of the Internet as an important medium for mass communications effected an equally important shift in the importance of Internet intermediaries, both in terms of helping end users filter out bad content and in helping them identify and obtain access to good content. In addition, the literature on the economics of intermediation underscores how intermediaries can play key roles in helping end users obtain access to the content they desire. Together these insights demonstrate that intermediation should not be regarded as a necessary evil, as some commentators have suggested. On the contrary, intermediation can play a key role in helping end users obtain access to the content and applications they desire.

This passage, and the rest of the paper, elides one of the most fundamental distinctions in the Internet’s architecture: that between routers and servers. To recap, a router is a computer that transmits other computers’ packets from one place on the network to another—it generally is not the destination of packets transmitted by network endpoints. In contrast, servers are network endpoints that provide services to end users. These can include web servers, mail servers, instant messaging servers, gaming servers, and so forth.

Yoo lumps these two very different types of computers together under the category of “intermediaries,” and then proceeds to argue, in essence, that because servers often exercise an editorial role, packet filtering by routers cannot be censorship. This is roughly equivalent to arguing that because Newsweek prints some letters to the editor and declines to print others, it isn’t censorship if the US Postal Service refuses to deliver magazines it doesn’t approve of.

To get more concrete, it’s obviously true that Google exercises a certain amount of editorial discretion when it develops its search algorithms. But it’s a huge leap to equate Google’s editorial function with packet filtering by a large ISP such as Comcast. And the reason goes back to the difference between servers and routers. Routers and servers play fundamentally different roles in the Internet’s architecture. If you read Yoo’s paper carefully, you’ll find that virtually all the examples he cites of “good” content filtering is done by servers, while most of the commonly-cited examples of potentially-harmful behavior by service providers is done by routers. So, for example, Yoo mentions spam. While there are a few ISPs that take limited anti-spam measures at the network level (such as blocking port 25), the vast majority of spam-filtering takes place at network endpoints: either on mail servers or end users’ computers. And this is a good thing! I get my email service from parties other than my ISP, and I already have a spam filter I’m happy with. It would make me angry if my ISP started intercepting my communications with my email provider in a misguided effort to “help” me control my spam.

The reason the distinction between routers and servers matters is because the costs of switching ISPs (routers) are much higher than the costs of switching online applications (servers). Most users will be customers of just one ISP at any given point in time, but they’ll interact with dozens, if not hundreds, of different network endpoints. Right now, for example, I’m a Comcast subscriber. Using my Comcast subscription, I access about 150 RSS feeds, 25 podcasts, 4 email accounts, two instant messaging accounts, a Twitter feed, and dozens of websites on any given day. Because the cost of choosing online services is so low, I’ve been able to assemble a group of online services whose editorial decisions collectively satisfy my own idiosyncratic preferences for online content.

This would be much harder to do if filtering were commonly done at the ISP level. I have only a handful of ISPs to choose from, and I’m only going to want to be a customer of one ISP at a time. If ISPs commonly filtered their customers’ traffic, it’s extremely unlikely that I would be able to find an ISP that had exactly the filtering strategy I preferred. So it’s in the interest of me and every other Comcast customer for Comcast to pursue a lowest-common-denominator routing strategy, delivering all packets and letting end-users decide what to do with them. When an ISP fails to deliver a packet a user has requested, I don’t think it’s at all unreasonable to describe that as “censorship” rather than “editorial discretion.”

Last week I critiqued Tim Wu’s Slate piece on the GBS deal. One of Wu’s key arguments was that other companies are unlikely to scan orphan works because there just isn’t that much money in them. Jerry Brito points out that Wu had a different perspective on the value of “long tail” works in 2006.