The Contingent Internet

David D. Clark

Abstract: The Internet is so omnipresent and pervasive that its form may seem an inevitability. It is hard to imagine a “different” Internet, but the character of the Internet as we experience it today is, in fact, contingent on key decisions made in the past by its designers, those who have invested in it, and those who have regulated it. With different choices, we might have a very different Internet today. This paper uses past choices made during the emergence of the early Internet as a lens to look toward its future, which is equally contingent on decisions being made today: by industry, by governments, by users, and by the research community. This paper Downloaded from http://direct.mit.edu/daed/article-pdf/145/1/9/1830691/daed_a_00361.pdf by guest on 01 October 2021 identi½es some of those key choices, and discusses alternative futures for the Internet, including how open, how diverse, how funded, and how protective of the rights of its users it may be.

Is it possible that the Internet might never have hap- pened? Is it possible that, in a parallel universe where the Internet’s inventors had pursued different careers, we could be without a network that links all of our computers together? That we might have “personal computers” that were truly personal, not connected to the larger world unless their contents were copied to disk and mailed? Actually, that alternative outcome is highly improb- able. The Internet was in some respects a creation of its time: in the 1960s, the idea of a global network for computers was “in the air.” A visionary of the time, J. C. R. Licklider, had already predicted teleconfer- encing, information sharing, instant messaging, online DAVID D CLARK tax preparation, offshoring, and the potential for a . , a Fellow of the dig ital divide.1 However, at the time of the Internet’s American Academy since 2002, is Senior Research Scientist at the mit launch, there were competing conceptions for how Computer Science and Arti½cial In- to build a “computer network.” Our alternate universe telligence Laboratory. He has been is not without the Internet, but rather is with a very in volved in the design of the Internet different Internet. since the mid-1970s and is a mem- This possibility may itself seem surprising: the In- ber of the Internet Hall of Fame. His ternet today is so omnipresent, so much a ½xture of recent policy publications include a our lives that it seems almost as if it “had to be that chapter in Trust, Computing, and Soci- ety (ed. Richard H. R. Harper, 2014), way.” What might an alternate Internet have looked and articles in the journals Telecom- like? This is an important question, because to recog- munications Policy and Journal of Infor - nize that there were multiple options for the early In- mation Policy. ternet, and that the Internet as we know it is contingent

© 2016 by David D. Clark doi:10.1162/DAED_a_00361 9 The on decisions that could have led to different not optimal for any particular application. Contingent outcomes, is to recognize that the future of Design for optimal performance and design Internet the Internet is itself contingent. Society for generality are two distinct objectives. will meet forks in the road that will deter- And it may take more effort to design each mine the future of the Internet, and recog- application in a general network than in a nizing these points and discussing the alter- network that was tailored to each applica- natives, rather than later looking back and tion. Over the decades of the Internet’s evo- wondering if we chose the right path, is an lution, there has been a succession of dom- opportunity we cannot forego. inant applications. In the early years of the Internet, the Internet was equated to email, The Internet is a “general purpose” net- and to ask people if they were “on the In - work, designed for a variety of uses. It is suit- ternet” was to ask if they had an email ad- ed to email, watching video, playing a com- dress. Email is an undemanding application Downloaded from http://direct.mit.edu/daed/article-pdf/145/1/9/1830691/daed_a_00361.pdf by guest on 01 October 2021 puter game, looking at Web pages, and myr - to support, and if the Internet had drifted iad other applications. To an Internet engi- too far toward exclusively supporting it (as neer, the Internet is the system that moves was happening to some degree), the Web data, and the applications (like a Web brows- might not have been able to emerge. But the er, which users might lump into the larger Web succeeded, and its presence as a com- concept of “Internet”) run on top of that plement to email reminded engineers of the data-transport service. This modularity, and value of generality. But this cycle repeats, this generality, seem a natural way to struc- and the emergence of streaming audio and ture a network that hooks computers to- video in the early 2000s tested the generality gether: computers are general-purpose de- of an Internet that had drifted toward a pre - vices; since the Internet hooks computers sumption that now the Web, and not email, together, it too ought to be general. But this was the application. Today, streaming, high- idea was quite alien to the communications quality video drives the constant reengi- engineers of the early-Internet era, who neering of the Internet, and it is tempting largely worked for telephone companies. once again to assume that we know now They asked what was to them an obvious what the Internet is best suited for, and op- question: how can you design something if timize it accordingly. The past teaches us you don’t know what it is for? The tele- that we should always be alert to protect the phone system was designed for a known generality of the Internet, and allow for the purpose: to carry telephone calls. The re- future even when faced with the needs of quirements implied by that purpose drove the present. every design decision of the telephone sys- There is another aspect of generality: the tem; thus, the engineers from the world of applications that run over the basic trans- telephone systems were confounded by the port service of the Internet are not designed task of designing a system without know- or distributed by the same entity that pro- ing what its requirements were. The early vides the basic data-transport service. This history of the Internet was therefore writ- characteristic has been called the “open” ten by people who came from a computing Internet, and again, this separation made background, not a classical network (tele- sense to a computer engineer but did not ½t phone) background. Most computers are conceptually with the telecommunication built without a singular purpose, and this engineer. The telephone company installed mind-set drove the Internet’s design. that wire to your house to sell you telephone But this generality has a price. The ser - service, not to enable some other company vice the Internet delivers is almost certainly to sell you theirs. From the telephone com-

10 Dædalus, the Journal ofthe American Academy of Arts & Sciences pany’s perspective, it is expensive to install phone system, which at all levels reflects David D. all those wires, and how could they get a the centrality of the telephone call). If the Clark reasonable return on investment if they design of the Internet required that the net- were not the exclusive service provider? work understand what the application were In the early days of the Internet, the only doing, deploying a new application would way to access the Internet from home was to require its designer to somehow modify the use a modem to make a dial-up connection core of the network to include this knowl- to an Internet service provider (isp). A res - edge. To the early designers, this was a fork idential user paid the telephone company in the road down which they did not want to for the telephone service, and then paid the go. If an application designer had to alter the isp for providing access. This seemed then network before deploying a new applica- like a minor shift in the business model of tion, this would both complicate the process the telephone companies. But as the possi- of innovation and create potential for the Downloaded from http://direct.mit.edu/daed/article-pdf/145/1/9/1830691/daed_a_00361.pdf by guest on 01 October 2021 bility of expanding broadband services to network to block one or another applica- the home emerged in the 1990s, the corpo- tion. rate resistance to an open platform became The Internet has been called the stupid net- quite clear. One telephone executive ex- work, the telephone system being the intel- plained to me at the time: “If we don’t come ligent network; the open-design approach of to your party, you don’t have a party. And the Internet makes perfect sense–that is, we don’t like your party very much. The on- un til things go wrong. If the network itself ly way you will get broadband to the home is impairing the operation of an application, is if the fcc forces us to provide it.” the network cannot always detect or correct That was a fork in the road, and the In- this. The network may be able to detect that ternet certainly might have taken another one of its components has failed, but more path. In fact, the force that led the Internet complex failures may go undetected, leav- toward residential broadband was, to a con - ing frustrated users who can see that their siderable extent, the emergence of the cable application is not working, but who have television industry as a credible and com- no remedy available to them. Had we taken petitive provider of high-speed residential the fork in the road that enabled the network Internet. to know more about what each ap plica tion We continue to see echoes of this tension was trying to do, the network might have between the Internet as an open platform been less supportive of easy innovation, for third-party applications and broadband but might also have been less frus trat ing to access as an expensive investment that use when unexpected problems inevitably should work to the advantage of its owner. arose. The current debates around the concept of Finally, the division of responsibility be- “network neutrality” are at their heart about tween the provider of the data-transport whether broadband providers should be ser vice and the provider of the application regulated to provide a neutral, open plat- means that responsibility for core require- form for third-party services, or if they have ments like security is divided among several the right to de½ne the services they offer actors. This both makes the objective harder (and perhaps favor) over the infrastructure to achieve and adds incentive to delegate they invested in building. the task to another party. In this way, the de- Another consequence of generality is that sign decisions that shaped the Internet as we the data-transport layer of the Internet has know it likely did not optimize secure and no concept of what the application is trying trustworthy operation. to do (as opposed to the design of the tele-

145 (1) Winter 2016 11 The These design choices led to differences in signals ambushes in a power struggle between Contingent the technical character of the Internet, but carriers and computer industry. Everyone Internet many choices also led to particular out- knows that in the end, it means ibm vs. Tele - comes in the industrial structure of the In- communications, through mercenaries. It ternet ecosystem. When we made design may be tempting for some governments to let decisions about system modularity in the their carrier monopolize the data processing early Internet, it was not entirely clear to us market, as a way to control ibm. What may that our design was both a technical struc- happen, is that they fail in checking ibm but ture and an industrial structure. Some of the succeed in destroying smaller industries. An - early network pioneers, though, certainly other possible outcome is underdevelopment, did understand this. In the 1970s, there was as for the telephone. It looks as if we may need a substantial debate between advocates of some sort of peacemaker to draw up bound- two sorts of networks: datagram and vir tual ary lines before we all get in trouble.2 Downloaded from http://direct.mit.edu/daed/article-pdf/145/1/9/1830691/daed_a_00361.pdf by guest on 01 October 2021 circuit. Datagram networks have a simpler Pouzin saw the battle over control of the core, with more functions shifted to hosts global network as a battle between the com- at the edge. Virtual-circuit networks have puter industry and the telecommunications more function in the core of the net, and industry. At the time, the computer indus- thus more power and control shifted to the try was dominated by huge players like ibm, network operator. The Internet is a data- giving shape to Pouzin’s “Battle of the Ti- gram network; the arpanet (Advanced tans.” ibm was a vertically integrated cor- Re search Projects Agency Network) that poration, just as the telephone companies preceded it was a virtual-circuit network. were: if a ½rm got its hardware from ibm, One of the most vocal advocates of the it likely got its software from ibm as well. datagram approach was the French comput- Pouzin may not have foreseen the coming er systems designer Louis Pouzin, who was shift in the computer industry to support building a datagram network called Cyc - more open-hardware platforms, but he lades at the same time that the Internet was clear ly saw different technical decisions as taking shape. In 1976, he published a paper shifting the balance of power from one in- that reached the following conclusion: dustry sector to another. The controversy dg vs. vc in public packet In contrast to the Internet, Pouzin’s Cyc - net works should be placed in its proper con- lades network was ultimately unsuccessful. text. Its failure is (speculatively) attributed to the hostility and resistance of the French ptt First, it is a technical issue, where each side has (postal, telegraph, and telephone govern- arguments. It is hard to tell objectively what ment unit). a balanced opinion should be, since there is no unbiased expert. This paper argues in favor of dgs, but the author does not pretend being One of the lessons of the past is that the unbiased. Even if no compromise could be users of the Internet are an active force in de- found, the implications would be limited to ½ning what the network is, both by choice some additional cost in hardware and soft- of application and by the creation of unex- ware at the network interface. So much re- pected applications never anticipated by sourc es are already wasted in computing and network engineers. This trend continues communications that the end result may not today with the success of user-created pro- be affected dramatically. grams for peer-to-peer music sharing, for example. Sometimes users take the network Second, the political signi½cance of the con- down a fork in the road that the designers troversy is much more fundamental, as it

12 Dædalus, the Journal ofthe American Academy of Arts & Sciences did not contemplate, or perhaps had even neutral platform that only moved data from David D. dismissed. In the very early days of comput- sender to receiver, such peeking would seem Clark er networking, the designers were focusing unnecessary; but that interpretation of to- on remote access to the expensive, high- day’s Internet is an oversimpli½cation. Op - power computers of the time. One of the erators claim that they need to see what early network engineers asserted that mes- users are doing in order to optimize their sage services were “not an important moti- experience–which may at times be true– vation for a network of scienti½c comput- and to selectively influence what they are ers.”3 Of course, users proved him wrong, doing. Many of these interventions by In- flocking to email en masse. ternet service providers–such as the mod- The second, related lesson is that the open i½cation of data in transit to insert advertise - character of the Internet is what allows this ments–have been met with protest, but sort of user-driven evolution to take place. other interventions–like reformatting data Downloaded from http://direct.mit.edu/daed/article-pdf/145/1/9/1830691/daed_a_00361.pdf by guest on 01 October 2021 In a more structured and vertically integrat- to ½t content onto the small screen of a mo- ed vision of computer networking, the net- biledevice–are more easily justi½ed by us - work provider might not even choose to of- ers. The tussle over the use of encryption is fer an email application. The Internet, by its only the latest chapter in the struggle be- structure, is amazingly open to exploration tween the providers of the data-transport by users and third-party innovators. This service of the Internet, the providers of bene½t, though, was clearly contingent on higher-level services and applications, the earlier design choices. users, and the state system for control of the user experience. Is the core technology and structure of The experience of using the Internet is be - the Internet set for the inde½nite future? coming more diverse. While the data-trans- Or are there further forks in the road that port service is more or less uniform across might change the basic character of the In- the globe, that service does not de½ne the ternet? One important and ongoing debate user experience. The experience of the user concerns the extent to which encryption is de½ned by the applications that run “on should be used by default to protect com- top of” that service. If different applications munication between users from observa- are available or are preferred by users in a tion (and modi½cation). If the goal is priva- given region, the resulting Internet experi- cy of communication among communities ence will, in turn, be different. Perhaps the of users, encryption is a powerful tool. But most obvious example of this today is the In- encryption thwarts the goals of many other ternet experience in China, where the state actors: the intelligence community bene½ts government has blocked access to many of greatly from being able to spy on content, the applications that de½ne the Western us- and in some nations, this right of spying is er experience, such as Facebook and Twitter. not subject to debate. If the Internet were to But there are domestic equivalents to these move to a posture of “encryption by de- services within China that make the Chinese fault,” would certain nations opt out of the Internet experience a vibrant space of in- public Internet as we know it and essentially teraction, even if it is heavily policed. Still, create a separate, state-controlled network there is no easy way for a Chinese and Amer- for their citizens? Further, Internet service ican to “friend” each other using Facebook. providers ½nd encryption between commu- The global community will need to de- nicants problematic, since it prevents them cide the extent to which we ½ght against also from seeing what their users are doing. this diversi½cation of the user experience. If the Internet were still the totally open, Some of the early Internet visionaries con-

145 (1) Winter 2016 13 The ceived the Internet as a platform for global if imperfectly, that different applications Contingent discourse and a vector for a global civil so- might call for differing degrees of account- Internet ciety. Diversi½cation of the Internet experi - ability; for example, while a transaction be- ence would seem to erode that vision. On tween a customer and his or her bank calls the other hand, differences of language, cul- for strong mutual veri½cation of identity, an - ture, and norms are real, and would suggest other user may not be comfortable search- that as the Internet matures, the experience ing for information on aids, for example, in any region would evolve to conform to if the query is not anonymous. De½ning those factors. The generality of the platform which actors can issue globally trustworthy does not mean that everyone has to use it identity credentials is, of course, another the same way. In fact, the generality, com- challenge to uniform identity management. bined with the ability of users to vote (via These concerns remain valid; but at the their usage) for the applications of their same time, the pressure to improve our abil - Downloaded from http://direct.mit.edu/daed/article-pdf/145/1/9/1830691/daed_a_00361.pdf by guest on 01 October 2021 choice, almost makes regional diversity in- ity to hold users accountable and deter ma- evitable. If the nations of the world were to licious behavior may push toward stronger push for some sort of global alignment of identity tracking. In many nations today, regulation and incentive, it would likely lead one must provide a national identity num- to a more homogenous but less satisfactory ber to use the Internet; respect for anony- Internet experience. Perhaps the ideal is an mous action may only be a local preference, Internet that accepts the diversity of expe- which could erode under global pressure for rience for most users, but permits interac- accountability. tion on a global level among users who seek The explosion of mobile devices signals it. We should urge China not to block Face- another inflection point for the future of the book (which they view as a threat to regime Internet. The traditional industry narrative stability) but we should at the same time ac- pitted application designers against the In- cept the outcome that most Chinese users ternet service providers: application design - prefertheir domestic alternatives. ers wanted an open, neutral platform on Another critical issue that will shape the which to innovate and the Internet service future Internet is the poor state of Internet providers wanted control over the services security. We hear almost daily about theft of offered in order to effectively monetize the data, computers corrupted with malicious user experience. Computer manufacturers software (malware), cyber-crime, and many were seen as neutral in this dynamic, lacking other breaches of the security of both users a business model that bundled proprietary and service providers. This state of affairs applications on the device. But the trajectory could play out in a number of ways. One is of the mobile device is very different: the that the current state of insecurity persists, makers of smart phones show a much great- which might eventually prevent sensitive or er interest in shaping (and monetizing) the important transactions from taking place user experience. Apple charges a fee (cur- online. In this way, poor security could be a rently 30 percent) for the sale of a paid appli- barrier both to the uptake and utility of the cation for the iPhone or iPad through their Internet. An alternate future is that in the app store. The regulators in the United States attempt to improve Internet security, the In - would be quick to intervene if isps tried to ternet mutates to hold users more account - charge a fee to customers for using a speci½c able for their actions. For a number of rea- app, but so far there has been little criticism sons, the original design of the Internet did of device makers doing exactly that. not include any mechanisms to deal with The interplay of device maker, isp, and identity management. It was understood, application creator and provider is partic-

14 Dædalus, the Journal ofthe American Academy of Arts & Sciences ularly interesting in developing nations. our hope in competition, as if with enough David D. Getting developing states online is socially competition the market will converge on Clark desirable; the power of connectivity to im- what users prefer. Sadly, however much we prove the conditions of citizens is evident. wish, there is simply not much competition But what strategies are acceptable in pursuit to build residential broadband access net- of this goal? Here, Facebook has launched works; the investment and risks required a clever scheme: they developed a stripped- to become competitive are too great. And down version of the Facebook application at the layer of applications, there seems to (called 0.facebook.com; the zero implies be a recurring tension between building an zero cost) and have negotiated an arrange- application that attracts users and develop- ment with mobile service providers in many ing an application that makes money, per- developing countries ensuring that use of haps by capturing information about the the application is free to users, not counting user to be used in more selective (and thus Downloaded from http://direct.mit.edu/daed/article-pdf/145/1/9/1830691/daed_a_00361.pdf by guest on 01 October 2021 against any data quotas. By making use free more pricey) advertisements. Competition –and in some cases even arranging for a can not discipline this behavior if the behav - dis count on the device–the uptake of Inter - ior alone allows commercial providers of net in the developing world may increase. applications to make money. But as a consequence, a generation of users Thus, another fork in the road is how the will equate the Internet not with sending Internet will be paid for. Today, aside from email, not with searching the Web, but with some public-sector money that supports using Facebook. This is not a hypothetical spe ci½c challenges like rural deployment, outcome; surveys suggest it is already hap- there are only two important sources of pening.4 Is this degree of corporate capture mon ey to pay for the Internet: the fees we acceptable? And what growth potential pay as service subscribers, and advertising. does it limit? Of course, this is an issue that Internet users today pay for broadband ac- each country will decide for itself through cess, and they pay for an assortment of ap- domestic regulation (or deregulation, for plications and services, including stream- that matter). But again, such decisions will ing video, online games, and music services. likely diverge the character of the Internet But advertising pays for the “free” Internet experience as it evolves across the globe. experience, those websites that cost users nothing to visit. And advertising dollars will The design alternatives I have described not grow without bound. Spending on ad- may seem to concern principally the user vertising can only be a fraction of commerce, ex perience, but at a deeper level, they are or ecommerce, as is most common on the In- strug gles over control. The previous exam- ternet. In 2013, $42.8 billion was spent on In- ples illustrate tussles between isps, applica - ternet advertising in the United States.5 The tion designers, device makers, and govern- total monthly advertising expenditure per ments (among others) over control of the household with broadband access (eighty- Internet. Depending on how the balance of eight million households in 2013) is about power among these actors evolves, we may $40.6 In other words, all of the advertising- see different outcomes with respect to de- supported Internet content is ½ghting over ployment, openness, innovation, and user an amount that is smaller than the average experience. The immense power of private- monthly cost of broadband access. Could sector actors is notable; society has largely the Internet experience stall because we run left the future of the Internet in the hands out of advertising dollars?7 of far-reaching, pro½t-seeking entities. In Online advertising will indeed grow as it the United States, there is a tendency to put cannibalizes traditional tv advertising. But

145 (1) Winter 2016 15 The what if a new “Internet experience econ- Internet is now an engine of economic in- Contingent omy” arises, in which users pay a small novation. But from another point of view, Internet amount for access to a broad spectrum of there might be a richer, more diverse set of Internet applications that do not track their uses for the Internet if nonpro½t actors were usage and do not run advertisements? Users motivated and supported to develop “non- are largely accustomed to paying for both commercial” applications. As the prolifer- premium apps on mobile devices and mu- ation of apps for mobile devices suggests, sic and video. Perhaps there will be a shift it is not hard to launch a new application in how they pay for access to Web content. today. Perhaps one way to pick among fu- And were this to happen, who would con- ture alternatives for the Internet is for in- trol that payment ecosystem? terested parties to vote with their dollars, The next major tussle over control in- funding the development of applications volves the governments of the world. The that are not motivated by the pursuit of prof- Downloaded from http://direct.mit.edu/daed/article-pdf/145/1/9/1830691/daed_a_00361.pdf by guest on 01 October 2021 private sector has a common set of motiva- its, but by interest in civic, cultural, or po- tions: be pro½table, grow, survive. Govern - litical participation. ments have a range of concerns: national se- It is possible that if we leave the future of curity (which can include regime stability), the Internet in the hands of powerful pri- law enforcement, taxation, control of “un- vate-sector players, we will get the outcome acceptable” content, and protection of the we want. It is possible that if we allow the rights of powerful private-sector actors governments of the world to make decisions (such as protection of copyright), among that shape the future of the Internet, we will many others. Different countries have dif- get the future we want. But there is great risk ferent priorities, different laws, and differ- in being passive about the Internet’s future; ent approaches to governance. In some cas- it may simply be too important to leave ei- es, these priorities put them at odds with the ther to the forces of commerce or the mech- private sector that, in most countries, has a anisms of global politics. Perhaps the most dominant influence over the character of important question is how the voice of the the Internet. In late 2012, at the World Con- users of the world can be injected into the ference on International Telecommunica- decisions that will shape the future of the tions in Dubai, the International Telecom- Internet. munications Union (a division of the United Nations) proposed an international treaty that would give it the right to regulate inter - national interconnection in the Internet. This idea was supported by a number of powerful nations, but it failed to gain trac- tion. However, this preference for state con - trol over important aspects of the Internet will probably continue to grow in certain quarters. But in all of this contention over the future of the Internet, there is one set of actors that has faded from view: the federally funded research community that designed and built the Internet. From one point of view, this trajectory is proper: they did their job, the commercial world has taken over, and the

16 Dædalus, the Journal ofthe American Academy of Arts & Sciences endnotes David D. Clark 1 J. C. R. Licklider and Robert Taylor, “The Computer as a Communication Device,” Science and Technology (April 1968). 2 Louis Pouzin, “Virtual Circuits vs. Datagrams: Technical and Political Problems,” Proceedings of the June 7–10, 1976, National Computer Conference and Exposition (New York: Association for Computing Machinery, 1976), 483–494. 3 Lawrence G. Roberts, “Multiple Computer Networks and Intercomputer Communication,” SOSP ’67 Proceedings of the First ACM Symposium on Operating Systems Principles (New York: Association for Computing Machinery, 1967). 4 Leo Mirani, “Millions of Facebook Users Have No Idea They’re Using the Internet,” Quartz, Feb - ruary 9, 2015, http://qz.com/333313/milliions-of-facebook-users-have-no-idea-theyre-using-the -internet/.

5 Interactive Advertising Bureau and PricewaterhouseCoopers, IAB Internet Advertising Revenue Downloaded from http://direct.mit.edu/daed/article-pdf/145/1/9/1830691/daed_a_00361.pdf by guest on 01 October 2021 Report: 2013 Full Year Results (New York: Interactive Advertising Bureau and Pricewaterhouse- Coopers, 2014), http://www.iab.net/media/½le/IAB_Internet_Advertising_Revenue_Report_ FY_2013.pdf. 6 National Telecommunications and Information Administration, United States Department of Commerce, “Household Broadband Adoption Climbs to 74.2 Percent,” June 6, 2013, http://www .ntia.doc.gov/blog/2013/household-broadband-adoption-climbs-724-percent. 7 Moreover, how will the growth of programs like Adblock, which allows its two hundred million users to surf the Web without seeing most conventional ads, affect advertising revenue?

145 (1) Winter 2016 17