Starclouds

As I promised last week, I wanted to talk a bit about my chapter in the new book Blindside, edited by Francis Fukuyama. Because the book (like the conference it was based on) focuses on prediction and forecasting, I framed the chapter as a discussion of the near-impossibility of trying to forecast technological outcomes-even in areas like information technology, where trend lines like Moore’s Law would seem to make it easy. But it’s really a meditation on the nature of innovation, using examples drawn from computer history (many of which I talked about in my own book, The Dream Machine.)

I can’t just post the chapter whole, because it’s copyrighted to the Brookings Institution Press. But basically I make two points:

Innovation doesn’t just happen.

As often as not, they originate from very specific efforts to solve immediate, practical problems. It’s only later , when the innovators are really immersed in the thing, that they begin to appreciate the visionary implications. To quote from the chapter:

Back in the 1920s and 1930s, for example, academics tended to be rather contemptuous of raw number crunching, on the theory that a real mathematician or scientist gains insight by abstract reasoning, not reckoning. Slide rules were acceptable, for engineers. But brute number-crunching was just arithmetic, a task for desktop adding machines-women’s work. (The word “computer” was still a job description in the 1920s, and had much the same pink-collar connotation as “typist.”) The building of computing machinery was, by extension, a job for mere tinkerers.

As a result, the road to modern electronic computing began with a handful of very practical pioneers. They were motivated in large part by desperation: modern technology was already beginning to demand calculations on a scale that humans could not manage, even with adding machines.

A prime example is Vannevar Bush,

who orchestrated the Manhattan Project and all the rest of nation’s war-related scientific research during World War II. Bush is probably best known today for his 1945 article about the “Memex,” a hypothetical knowledge-access tool that could link one concept to the next in a manner that anticipated the World Wide Web by nearly half a century. [The Memex story is told in a wonderful book, From Memex To Hypertext.]

But the path that led Bush to the Memex began more than 20 years earlier, when he was an MIT electrical engineering professor trying to analyze the instability of electric power networks. The computers he created were mechanical, analog machines (as the Memex would have been). But for the class of problems they were meant to solve, they were the most powerful calculating machines on the planet in the 1930s.

Other examples:

• The digital, all-electronic ENIAC, widely regarded as the first modern computer, was constructed by engineers at the University of Pennsylvania to calculate artillery trajectories.
• The Hungarian-born mathematician John von Neumann, who pioneered numerical processing and scientific supercomputing in the late 1940s, first became interested in the subject because he was a participant in the super-secret Manhattan Project, and was looking for computing machines that could help out with the horrendous calculations needed in that effort.
• Claude Shannon, the creator of modern information theory, was an AT&T Bell Labs employee who was originally trying to quantify the communications capacity of telephone networks.
• The MIT mathematician Norbert Wiener, who combined information theory, computing, and feedback control in his highly influential 1948 book Cybernetics -a word he coined, and the origin of today’s ubiquitous “cyber” prefix-was originally inspired by his World War II work on the control of anti-aircraft artillery.

The list could go on and on. Tim Berners-Lee originally created the World Wide Web as an easier way for his colleagues at the CERN particle accelerator facility to access online scientific documents. The Google search algorithm started as a grad student project funded under a federal research program on search and retrieval for digital libraries. Facebook was originally just a way for undergraduates at one school-Harvard-to keep track of one another.

I certainly don’t want to claim that innovation always grows out of efforts to solve practical problems; I’m still a huge fan of basic, blue-sky research. (My Ph.D. was in elementary particle physics, which is about as blue-sky as you can get.) But from a policy perspective, this is yet another vote for robust funding of research in Pasteurs Quadrant.

Innovations don’t happen in isolation.

To quote from the chapter again:

They almost never involve just a single idea, but the convergence of many ideas. And they are not inevitable: they result from social needs and interactions. The modern digital computer, in particular, required the convergence of at least half a dozen innovations-most involving not just another gadget, but a shift in the way people thought about computing. Well into the 1940s, moreover, people were struggling to fit the pieces together in the right way; it took a decade of trial and error (and a war) to get a combination that was workable. Among the most important of these pieces:

Digital computing: solving problems by numerical calculation as opposed to building a physical model of the problem. It was far from obvious in the beginning that digital was the right way to go, especially given the success of analog machines such as Bush’s Differential Analyzer.

Binary mathematics, as opposed to the base-10 arithmetic that humans had been using since they first began counting on their fingers.

Logic: the recognition that a simple on-off switch could embody the notions of true and false, and that a network of such switches could embody all the standard operations of Boolean logic-the operations of binary arithmetic among them. In particular, the network could make comparisons, and thus take alternative courses of action according to circumstances-as in, “If the number X equals the number Y, then do operations P, Q, and R.” That ability, in turn, was what made the digital computer so much more than an ultra-fast adding machine. A switching circuit could add and subtract-but it could also decide. It could work its way through a sequence of such decisions automatically. In a word, it could be programmed.

All-electronic switching, using vacuum tubes for speed as opposed to mechanical switches. Again, the choice of vacuum tubes was far from obvious in the early days, given that a computer would need tens of thousands of them to do anything useful, and that even a single burnt-out tube could bring the system to a halt. How would you ever finish a calculation?

Program control-giving computers the power to carry out long sequences of operations on their own, as opposed to relying on a human operator to press the buttons, watch the meters, load and unload the punch cards, and generally intervene at every step.

Stored program control-that is, storing the program as binary code in the computer’s memory, as opposed to reading it in each time from punch cards or paper tape. Once again, the usefulness of this approach was not entirely obvious at the beginning; many of the early computers, including the ENIAC, had at least some of their programming wired into their physical structure. Implementing the stored-program approach was also a good deal harder than it sounds today, given the primitive state of memory technology at the time; no one was able to field a working stored-program computer until the late 1940s. But the stored-program approach had the obvious advantage of convenience: once all the instructions were stored electronically, so that the problem-solving sequence was entirely separate from the hardware, you could change the function of the computer without having to touch the wiring. Or to put it another way, the act of computation had become an abstraction embodied in what we now know as software.

The history of information technology offers many other examples of invention-by-convergence. Among them:

• The modern concept of information and information processing was a synthesis of insights developed in the 1930s and 1940s by Alan Turing, Claude Shannon, Norbert Wiener, Warren McCulloch, Walter Pitts and John von Neumann.
• The hobbyists who sparked the personal computer revolution in the late 1970s were operating (consciously or not) in the context of ideas that had been around for a decade or more. There was the notion of interactive computing, for example, in which a computer would respond to the user’s input immediately (as opposed to generating a stack of fan-fold printout hours later); this idea dated back to the Whirlwind project, an experiment in real-time computing that began at MIT in the 1940s. There were the twin notions of individually controlled computing (having a computer apparently under the control of a single user) and home computing (having a computer in your own house); both emerged in the 1960s from MIT’s Project MAC, an early experiment in time-sharing. And then there was the notion of a computer as an open system, meaning that a user could modify it, add to it and upgrade it however he or she wanted; that practice was already standard in the minicomputer market, which was pioneered by the Digital Equipment Corporation in the 1960s.
• The Internet as we know it today represents the convergence of (among other ideas) the notion of packet-switched networking from the 1960s; the notion of internetworking (as embodied in the TCP/IP protocol), which was developed in the 1970s to allow packets to pass between different network; and the notion of hypertext-which, of course, goes back to Vannevar Bush’s article on the Memex in 1945.

I think this point is quite generally true about innovation: it’s not just about individual gadgets, but how multiple gadgets-and ideas-are integrated into coherent systems. This is hardly an original insight-but one that a lot of people seem to miss, so it’s worth making again.

From a policy perspective, it argues for doing everything possible to promote the open exchange of knowledge, so that innovators can combine and reuse that knowledge in ways that are impossible to predict. But of course, it also leads to a long and complex debate over the limits to that openness: what’s the right balance with intellectual property protection, national security and the like, all of which act to restrict the free flow of information?

Comments?