Tech Policy Seminar

A few more thoughts on Carr

My apologies to the class for the late posing.

I enjoyed reading the first portion of Carr’s book on the modern technological shift towards centralized computing.  I have a few comments to make on it regarding its scope, feeling that Carr should (and maybe will later) expand his discussion toward some more of the possible spill-over affects.  Apart from the obvious legal questions of possible antitrust regulation of Google, should their market share and conduct warrant it some time in the future, or possible treatment (price regulation) of the supposed utility as a public utility, I would liked to have seen some more discussion of the social impacts (beyond the historical) and privacy issues.

It seems like there’s a lot more to focus on, other than the history of production and economic savings by corporations.  The information revolution is so different than the energy one because not only will it enhance production, but it changes the way we communicate ideas and makes the world increasingly smaller in unprecedented ways.  Like his discussion of the world post electrical revolution, I think he should have (assuming he doesn’t pick up on this a bit later) spend some time talking about the spillover social affects that high-speed ubiquity could have. 

Evolving Job Market

            For example, he characterized the electrical revolution as leading to a “virtuous cycle” (pg. 95) on the production end and the growth of the middle class (pg. 94) on the worker/consumer end.  I’d like to see a bit more exploration of the affects on employment in this country the shrinking of IT departments might have.  For example, not only will massive data centers like Google’s Dallas facility cut the raw number of laborers needed to achieve the same ends, but it might kill demand for domestic workers to fill those shrinking positions. 

If indeed the virtue of consolidated IT its ability to function effectively across great distances, a vice might also exist in that the political location of the technician is irrelevant as well.  In other words, the streamlining of IT, might encourage a further exportation of education-intensive jobs to countries like India      I think other practical concerns would severely cut back on that specific example, but there is ample room to see how something similar at least could be a real consequence of Carr’s vision of the future.  What does this mean for trade policy and the globalization debate?  How should the law respond?  How do you explain this to the American computer science graduate who has been told for years that globalization will bring prosperity and her place will be assured provided she get an education?  Perhaps, from an American Policy standpoint, it deserves more of a suspicious eye than is being offered in his account thus far.

, where skilled workers can be found at cheaper prices.  Technically lower costs are usually a good thing, but without a further look into the potential affects on the American economy, I think the discussion is incomplete.  It goes beyond IT as well.  Imagine: Kirkland & Ellis, Chicago branch.  Reduced to a hundred attorneys managing the leg work locally, supported by 500 research staff in Bangalore. 

Social Change

            The last chapter, “The White City”, made some interesting implicit comparisons between the current pending paradigm shift and the shift to cheap, ubiquitous electricity.  I found this particularly interesting in its description of its affects of business and home life.  This passage seemed particularly appropriate, “as women took over all the work required to keep house, they often found themselves spending more of their time alone, isolated in their suburban residences.” 

            I can’t think of a better way to describe the potential social affects of Carr’s vision of increased internet-based computing- maybe we’re seeing it already.   I liked the inclusion of some of the fantastic oddities contemporaries of Edison predicted for electricity, but I am unconvinced that the rise of the centralized computing network will have comparable affects on out daily lives without some further exploration of the extant technology and projects under development which could effectuate such changes.  For example, the growth of bandwidth is poised to threaten many forms of physical interaction, and physical media formats- PeaPod obviates the need to drive to the grocery store. iTunes is slowly killing the CD industry.  Netflix, AppleTV and peer to peer file sharing programs are beginning to threaten the survival of DVD’s.  There is ample room for further exploration, and I think Carr’s comparisons to past paradigm shifts would be enriched for it.

Illicit Uses and Privacy

            In Chapter 4 Carr discusses the demise of the traditional PC, suggesting that the future will be dominated by stream-lined machines that function as little more than terminals to the grander network (and indeed some exist already), on which the actual work and data storage occurs.  With respect to illicit uses of the digital format, there are a few ways to look at this. 

Firstly, traditional and wide spread illegal activity (such as copyright-infringing file sharing) would seem to put downward pressure on the shift to off-sight data storage.  Users accustomed to stealing music, films or software would be unwilling to risk storing their contraband on an off-sight server, where it might be subject to a greater rate of detection by authorities, deletion, or termination of service and fines.  Apart from that is the arguably paranoid reluctance to store personal material where other people have potential to access it, regardless of how safe it is.  Couple that technophobia with the fact that a personal computer can be purchased for a price lower than a Playstation, Carr’s internet terminals might be as likely as take over home computing as “electrified water” was to revolutionize home sanitation.

On the other hand, if such a shift becomes reality, it could have a potential spillover affect in helping to enforce the current copyright law.  It would be an interesting policy question to discuss whether or not such a computing service should have an affirmative duty to prevent unlicensed material from being stored on their hardware.  If Carr is indeed right about how the computer (or internet terminal) of tomorrow would function, it could potentially eliminate copyright infringement as a serious problem in this country in a way that traditional enforcement has thus far been unable to achieve. 

With respect to corporations, there would certainly be resistance to such a shift owing to obvious privacy concerns that don’t need too much explicit discussion beyond what Carr suggests (pg. 70).  It is at least worth mentioning that a corporation might consider retaining some portion of its IT needs on a local server- such as email- in order to control more confidential communications.  That being said, his point would still be valid regarding the large cost-savings as other services are out-sourced.

In Sum

            I like where Carr is taking the book so far, but I would enjoy it more if he spent a little less time discussing history (although quite interesting) and a bit more discussing the practical policy consequences- more substance, less filler.

Switching Paradigms

Nicholas Carr’s The Big Switch is an engaging and entertaining read, but ultimately vulnerable to many of the delusions and biases which plague all efforts to forecast beyond the “next big thing.” Carr is certainly correct in noting some real, confirmable trends in the integration of the online and offline worlds, but he suffers from some fundamental misunderstandings of the information technology (IT) industry. Most importantly, he misunderstands many of the technologies which are available now or soon to come.

IT Will Always Matter

My commentary that follows hinges on a critical misunderstanding that Carr demonstrates repeatedly, namely that IT services and structures really can be commoditized and made fungible in the same way as 60-cycle, 120-volt, alternating current. Despite many minor successes in this arena, the fundamental needs of businesses and consumers will not support such commoditization.

Carr’s article IT Doesn’t Matter in the Harvard Business Review launched a minor firestorm inside and outside the IT industry. A quick Google search reveals nearly ten thousand web pages and articles discussing the article (and the later book version). The problem, as countless commentators have noted, is that IT really does matter. Carr's fundamental point in that article was that IT and IT departments no longer provide competitive edges in the marketplace. If anything, however, an IT department now can provide a greater competitive edge than in years past. Google, for example, is well-known for hiring disproportionately large numbers of newly-minted Ph.D.s and masters degree recipients. While some may question the value of doctoral education in the IT industry - Bill Gates, after all, is a college dropout, and the Google founders never received doctorates - the Google strategy has resulted in an unquestioned success (at least so far), because it has so many truly talented people creating, updating, and maintaining its fantastically large and complex systems. Google, Amazon, Facebook, and countless other companies demonstrate the advantages of innovative IT.

The counterargument, of course, is that these are all IT companies. The point, though, is not that these are IT companies, but that they are big companies. The biggest and most complex businesses will always have needs for more creative and efficient ways of doing things; there is a reason that most large enterprises each obtain dozens if not hundreds of patents each year. Carr's examples of on-demand grid computing, such as 3Tera's AppLogic, only serve to prove my point. AppLogic does let an individual or enterprise rapidly create large and scalable networks, but it does not - cannot - ensure that those networks are being used in any efficient manner, that they are secure, that they are meeting end-user needs, or that countless other concerns are being met. IT departments are likely to evolve further and further from "the people who fix computers" to "the people who fix computer systems (i.e., grids, applications, etc.)," but they are not going away.

Limitations on Economies of Scale

Economies of scale are fundamental to Carr's conception of the future. Carr's premise in The Big Switch is that computing will become more and more like a utility, because computing utilities will be able to provide computing and storage at ever-lower costs until failure to use those utilities cripples one's ability to compete (or "be cool" at the individual level, etc.). There are a number of major problems, however, with the analogy to a traditional utility.

Traditional utilities offer services and facilities - e.g., power, communications, water, gas - that are prohibitively expensive if provided by individual families or businesses for their own uses. The startup costs are typically very large and unlikely to decline over time. That is, the tenth nuclear power plant may have been marginally cheaper to build than the first one, but building the same plant today would still be a major undertaking, comparable in inflation-adjusted price to the one that has already been built.

The same cannot be said of a computing or storage utility. When I was young, my father sold computer systems that cost roughly $27,000 for a tiny monitor, a few kilobytes of memory, and the ability to calculate and print (at additional cost) very basic spreadsheets. Our computer when I was in middle school had a 100 megabyte hard drive and initially cost a couple thousand dollars. My iPod, at a cost of $300, is faster, prettier, tinier, and loaded with far more memory and storage than either of those ancient machines. It even has more storage than the laptop I owned before my current one.

The point is this: costs in the computing world drop constantly, even as Moore's Law continues to push performance up and storage density climbs apace. At some point, it ceases to make economic sense either to build a datacenter or to create a virtual one through a service like 3Tera - at some point, a few machines will be able to handle most businesses' needs. We will, ironically, have moved back from the desktop to the mainframe, because a single system will have so much processing power that thin clients are all most people need. Ironically, as processors and hard drives get ever smaller and cheaper, there will be even less need for huge datacenters and outsourced processing - the desktop computer will meet more and more of a company's needs, without the need to rely on third-party help.

The Costs of "Plugging In"

Carr's grid sounds like a wonderful thing to an accountant. A critical feature of commoditized computing, however, is that applications and data must be structured to run in the right environment. Virtualization through products like VMWare promises the ability to consolidate a room of ancient machines into a handful of off-the-shelf PCs. Yet this promise is premised on portability. Many mission-critical computer systems accumulate a slew of hacks and hard-coded (programmed-in) tweaks which are not terribly portable or easy to detect and fix; just ask any programmer who worked on a Cobol system prior to Y2K how easy cleaning up any given program was.

Replacing an existing computer system - much less an entire datacenter - with utility computing frequently requires huge expenditures for recoding critical applications and scripts, not to mention tweaking countless settings and data entries. When a computer system moves, every system which interacts with it has to be informed of the change, and most large systems are not set up so that massively relocating an entire datacenter is a simple matter of flipping a switch.

At the same time, the kind of commoditization Carr envisions requires that applications be made increasingly system-independent, so that the code created can run on whatever system the utility provides. This is the motivation behind languages such as Java, which promises "compile once, run anywhere" computing - a single Java program can run on a PC, a Mac, a Linux system, or a Java-enabled handheld device. The problem is that such workarounds entail massive sacrifices in performance compared to system-tailored code (for the uninitiated, these hits can range from twofold to factors in the hundreds). Thus, even very cheap utility computing may require more machines (real or virtual) than a custom-built system, making the "utility" less economical. This is especially true as a company with huge needs, like Google, becomes large enough to develop its own economies of scale.

Risk Costs

Carr also overlooks one of the great dangers of utility computing: utilities succeed because of their stability. Companies and individuals did not connect to the power grid because of any particular utility's capacity to meet their needs, but because the grid itself guaranteed a supply of power from many utilities, even if one fails. This is not true of offerings like Salesforce.com. If ComEd goes under, Chicago's lights will stay on. If Salesforce.com goes under, thousands of businesses may find that much of their customer data is either lost or returned to them in a form they are unable to use. By comparison, using a desktop-based CRM (Customer Relationship Management) solution at least guarantees that the application will still work, even if the company which created it folds. The additional cost of the standalone solution might be seen as an insurance policy of sorts.

Give Them an Inch...

A further complication of Carr's world comes from his assumption that demand for computing power, storage, and bandwidth will trail supply. Historically, this has not been true. The invention of compression algorithms for MP3 files, for example, made it possible to transmit and store music far more efficiently than previously - and users promptly downloaded and stored correspondingly greater amounts. In fact, users often tried to get their hands on more music than before, overwhelming even high-speed networks, especially at universities. As bandwidth and storage become more affordable, rather than finding that users can download an entire musical track in seconds, for example, we will find that users demand more and higher-quality tracks; rather than an MP3, which is a "lossy" version of a recording, why not stream digital masters of a track, at many times the data volume?

The point of this is that users and businesses will find needs to match the supply of processing power and storage capacity; a utility will find it hard to scale accordingly and economically to continue attracting users.

Ten years ago, a ten-gigabyte hard drive was considered extravagant. Today, a terabyte drive costs less than its ten-year old counterpart, and people are buying them and filling them with music and video.

Everything Must Go

The movie Live Free or Die Hard was premised on a "fire sale," the act of taking down an entire country's infrastructure on the premise that "everything must go." **** Spoiler Alert **** The motivation for the fire sale committed by the "bad guys" in the movie turned out to be a massive data warehouse, containing effectively all of the financial information in the United States. It didn't much matter whether the villains accessed the data or merely held it hostage; they could demand massive payment for letting it go.

Every few days, a major company or government entity announces that its data has been compromised, sometimes to catastrophic effect. Yet it is in this environment that Amazon opened up its database system, SimpleDB, Google is opening up BigTable, and other companies are rushing to provide greater and greater data warehousing and processing capacity at a utility level. All computer systems, however, have vulnerabilities, a fact which both consumers and businesses fear. As computing becomes more like a utility, being off the grid may become a viable model for attracting customers.

Schrodinger's CPU

In a few years, the promise of quantum computing may well come to fruition. A quantum computer operates on the principles of quantum mechanics, which allows it to compute nearly infinitely many things at once, subject primarily to the limitations of hardware components. The promise of such computers is that they will render traditional computing obsolete - your computer will seem even slower next to a quantum computer than an Apple II seems next to your current computer. Computations which require one of the world's fastest supercomputers will require a single desktop.

But there is a threat in this promise, as well. Obviously, it will redefine the industry and probably destroy some computing utilities, but there is a legitimately scary side to quantum computing, as well. Quantum computing will render current encryption methods obsolete; rather than hundreds of years, it will be possible to crack encrypted data considered secure today in mere seconds. Today, if somebody breaches your bank's database security, they may get gibberish, encoded data they will never be able to use. If the same thing happens, but the hacker is hacking into a major computing utility and has access to a quantum computer, that data may as well not be encrypted at all.

In other words, in a few years, the confluence of increased utility computing and the dawn of quantum computers may make today's data breaches look trivial. Live Free or Die Hard may not be so far off the mark.

Conclusion

So far, I am genuinely enjoying Carr's book. I even think he is right, in that utility computing is becoming more and more the norm, especially for businesses a step or two ahead of the curve (or their hardware budget). That said, at least so far, he has overlooked a number of concerns businesses and consumers are likely to share, which will undermine widespread adoption of utility computing. Additionally, the combination of rapidly increasing power and dropping prices makes the world of IT less suited to the formation of utilities than traditional utility industries. Finally, rapid technological advances may render off-the-shelf computing so cheap and so powerful that outsourcing either storage or processing seems unnecessarily complicated and risky, especially after the first major data breaches at computer utilities. Carr is likely to be right in the short term, but wrong in the long term, like those who predicted Utopian results from electrification. Utility computing is just not something we really need.

A Few "Big" Questions

In The Big Switch, Nicholas Carr predicts a forthcoming mass transition from localized computing based on individual processing power to a centralized utility model of cloud computing. Throughout the first half of the book, Carr goes to great pains to relate the history of parallel transitions, focusing principally on the creation of the modern electricity utility. In so doing, he clearly seeks to establish precedent and explain the economic moves typical of such a “big switch”.

Building upon this basis of sound economic principles, Carr proceeds to assert that the current state of personal computing mirrors that of the electricity industry under Edison’s DC paradigm: one of wasteful duplication, rife with inefficiency. Carr argues that the emergence of a high-bandwidth infrastructure has now made it possible for the elimination of these inefficiencies in the computing market, a move that would resulting in the death of the personal computer and software industries as we know them, followed by the emergence of a new market of competing computer services utilities. He points out that the move has already begun in some sectors, citing the rapid growth of businesses like Salesforce.com and the resulting damage done to traditional business software vendors like SAP. Small moves like this will in theory create a virtuous circle, creating economies of scale for software services firms, and driving down costs—thus encouraging more and more firms to switch over until the transformation is complete.

While Carr’s approach is mostly positivist, a subtle undercurrent of approval seeps into his writing on the topic, as he extols the virtues of centralized utility models. This gloss includes few references to historic and predictable drawbacks such as economic (natural monopoly), regulatory, and other potential concerns.
Given the recurrent pattern of similar industries over the last century, Carr’s idea has intuitive appeal and may yet prove correct. Some careful scrutiny of a few of his claims is in order, however.

Infrastructure

First, Carr claims as a matter of fact that the requisite physical infrastructure for this transition is already in place.  This is empirically less than obvious, as the adequacy of current global network infrastructure is a point of disagreement among analysts. Some experts mimic Carr’s claims while others warn of an impending bandwidth squeeze. Given the relatively recent emergence of large, newly-industrialized economies such as India and China, along with other countries emerging from technological infancy, the number of new Internet users each year is immense. And, given the proliferation of peer-to-peer technologies and the expectation of ubiquitous video-on-demand, individual bandwidth consumption is likewise growing at a rapid pace. Both of these forces make the notion of a network with sufficient bandwidth to handle the content of virtually all the world's computer processes an unsure one for the next several years at least.

Finally, this claim has been made before. Thin clients have been trotted out by numerous players in the industry over the years—perhaps most notably in the form of "Net PC" and Internet appliance projects that were all the rage during the dot-com era.  Carr doesn’t meaningfully and convincingly establish why things are different now, other than to say the pipes are bigger. Thus as a matter of pure timing, Carr must provide more proof to support his contention that the Internet as now constituted (or as it will be in the coming decade) will be sufficiently robust to handle the complete transfer of nearly all personal computing through its pipes.

Consumer Wariness

Carr also largely ignores consumer opposition. He says little about the likelihood that many—at least at first—will be wary of a wholesale move of all data and processing power to another entity. This concern stems from both privacy and reliability fears.

The reliability concern is likely to be the less problematic of the two. Although naysayers will always raise the specter of power outages and other worst-case scenarios,  firms and individuals have shown a continued willingness to cede control when offered lower costs and other efficiencies. In the electricity market, several businesses were initially reluctant to relinquish control over their own energy production systems, but Carr aptly describes how these concerns gave way when utilities were able to provide significantly lower prices, freedom from upgrades, and freed-up factory space. A similar move is likely in computing, along with the adoption of similar emergency backup measures in the event of a localized or mass outage.

However, because the privacy issue does differentiate modern computing from the electrical grid and other utilities, it poses a more substantial problem. While transfer of the relevant resource is unidirectional in electrical systems--requiring nothing of the consumer but the flick of a switch or press of a power button—centralized computing requires forfeiture by the user of all personal and confidential data. When Ingersoll rolled out his centralized AC model, users faced no quid pro quo and had no real decision to make; adoption was a clear win-win proposition. With centralized computing, however, users and businesses must decide that they are comfortable with the whole of their relevant data being stored elsewhere. Recent reports of successful credit card and personal info hack attacks only serve to underscore the fear of many that their most precious possession—their personal or business data—will be stolen and used against them maliciously by competitors, data pirates, or criminals.

While this somewhat rational fear is a powerful obstacle to the ultimate success of the “big switch”, it will eventually give way to the indefatigable power of the market. After all, consumers and businesses managed decades (or centuries) ago to relinquish immediate control of another precious resource—their money—to banks, recognizing the economic advantages of increased physical security and investment potential. Carr’s failure to treat this issue is a glaring omission—provided, of course, he fails to do so in the second half of the book.

Natural Monopoly & Regulatory Concerns

As mentioned above, Carr also fails to discuss many policy concerns raised by the “big switch”.  First, if PC speed as currently distributed in duplicative local IT centers isn’t clearly much more expensive and wasteful than cloud computing—at least at present—then perhaps natural monopoly and privacy concerns may justify the lost efficiency. I’ve discussed privacy above, and now treat natural monopoly briefly. In his rather helpful and detailed recounting of the history of both electricity and computing in America, Carr concludes his narrative at the point of mass adoption of the technology in utility form. What he doesn’t mention (again, at least not yet) is the persistent and often intractable regulatory problems created in many similar industries.

Utility paradigms almost always involve immense upfront fixed costs, and as such are clearly most economically efficient when controlled by one single player. This avoids wasteful duplication of very costly infrastructure and the lost resources and time resulting from ruinous competition scenarios. The obvious problem with the presence of a single dominant firm in any market, is, of course, the predictable abuse of market power. These abuses take the form of decreased output, muted innovation incentives, and other economic distortions. Furthering the problem is that regulatory approaches throughout the last few decades have been unable to solve this market power conundrum without introducing other issues of a similar magnitude.

While it is difficult to predict the eventuality of things, Carr's utility computing industry is unlikely to devolve into a natural monopoly. First, unlike with the telephone or electricity, the necessary hardware and know-how for creating such a network is already largely distributed and in the hands of many. This distribution of resources precludes the formation of one single front-runner similar to a Bell Telephone Company or Edison Electric. Second, organization of the necessary hardware for providing such services is more a matter of combining individual component pieces than of outlaying huge amounts of capital for a few gargantuan pieces of machinery. This makes entry on a small scale much less difficult for existing IT firms, which own large amounts of server hardware and software already. It also makes entry tenable for those with the capital to incrementally build a system, rack server by rack server. These lower fixed costs should allow for entry and profit margins sufficient to support competition among multiple players.

My analysis may prove inaccurate, however. If so, utility computing would create the same natural monopoly problems we've faced for decades. In light of this fact, we might prefer to not make the switch at all; we might think that the current level of wasteful duplication is worth swallowing as an alternative to creation of yet another natural monopoly mess. This is preference is an option at all because computing hardware has been successfully miniaturized to accommodate individual use and has proven itself a workable model for the last few decades. (Unlike in classic natural monopoly industries, home users needn't have the equivalent of a gigantic electric switch situated down the street to meet their computing needs, and most businesses don't house the computer network equivalent of an early-Twentieth-Century telephone switch in their office basement.)

Such normative arguments are of little moment, however, if utility computing is indeed a natural monopoly industry. If it is, we can expect that individual efficiency incentives will inexorably lead us to it eventually, creating yet another headache for regulators, firms, and the computing public.