Oral-History:Robert Kahn
About Robert Kahn
Robert Kahn was instrumental in the creation of the Internet, from its earliest inception as ARPANET, a private/public collaboration between Bolt Beranek and Newman (BBN) and the Defense Advanced Research Projects Agency (DARPA). Kahn would straddle the divide, first working at BBN and eventually becoming the director of DARPA. He was also critical in transforming Arpanet into the full-fledged Internet.
Like Leo Kleinrock, he thinks wireless will become like cell phones, and that we are currently in a transitory period. Also, like Kleinrock, the elderly Kahn is still working. He is CEO and President for the Corporation for National Research Initiatives (CNRI). Kahn views what has happened over the last thirty years or so as the “networking revolution.”
See Vinton Cerf Oral History and Robert Metcalfe Oral History for further discussion of the development of ARPANET and of the Internet.
About the Interview
ROBERT KAHN: An Interview Conducted by Michael Geselowitz, IEEE History Center, 17 February 2004
Interview #432 for the IEEE History Center, The Institute of Electrical and Electronics Engineers, Inc.
Copyright Statement
This manuscript is being made available for research purposes only. All literary rights in the manuscript, including the right to publish, are reserved to the IEEE History Center. No part of the manuscript may be quoted for publication without the written permission of the Director of IEEE History Center.
Request for permission to quote for publication should be addressed to the IEEE History Center Oral History Program, IEEE History Center, 445 Hoes Lane, Piscataway, NJ 08854 USA or ieee-history@ieee.org. It should include identification of the specific passages to be quoted, anticipated use of the passages, and identification of the user.
It is recommended that this oral history be cited as follows:
Robert Kahn, an oral history conducted in 2004 by Michael Geselowitz, IEEE History Center, Piscataway, NJ, USA.
Interview
Interview: Robert Kahn
Interviewer: Michael Geselowitz
Date: 17 February 2004
Place: CNRI, Reston, Virginia
Geselowitz:
We’re here with Dr. Robert Kahn at the Corporation for National Research Initiatives (CNRI) on Tuesday, February 17th to conduct an interview on behalf of the IEEE History Center and the Marconi Foundation.
Background
Early Years
Geselowitz:
Dr. Kahn, thank you again for agreeing to do this oral history interview. I was wondering if you could tell me a little bit about your early years and your background, how you got into engineering, before we get to your actual later work.
Kahn:
It was just a choice that I made. I think originally I had thought that industrial engineering might be interesting. I think I initially signed up for chemical engineering because there wasn’t actually an industrial program. But I really didn’t like the chemical lab work as much. It turned out that my background in mathematics got me more over into the electrical engineering side of things.
Geselowitz:
Is that undergraduate study at City College of New York that you’re talking about?
Kahn:
I don’t think I was thinking much about careers before then, but I did get my start at City College of New York.
Geselowitz:
And you were born?
Kahn:
Born in Brooklyn, moved to Flushing, Queens, when I was about thirteen. I think it was 1953.
Geselowitz:
Which high school did you attend?
Kahn:
I went to Madison for one year, and then to Flushing High School for two. I was actually only in high school for three years. I was in an accelerated program through high school. I went to Queen’s College for two years and the City College for two and a half after that and got a Bachelor of Electrical Engineering degree.
Growing up as a youth, for as long as I can remember my mother had been dealing with a heart condition that she had contracted as a child. It was kind of difficult for my dad growing up. He managed to keep myself, my sister, and the whole family well intact through a very difficult period. She had had a heart attack on April 12th of 1945, which was the date that Franklin Roosevelt died. I think hearing that somehow brought it on. Then she had a whole series of them between then and about 1953, when she went through closed-heart surgery. Those were the days before they had the blood circulation machines, so they had to get in and out pretty fast. We didn’t have a lot of money. My dad was a high school teacher and principal, so staying close to home was important. That’s basically why I ended up going to the Queens and City College, although in retrospect, it was a terrific education that we got from both places. The hard-core engineering stuff was taught by City, but the more basic liberal arts stuff—math, physics, chemistry, humanities—were taught at Queens.
Geselowitz:
Coming from that kind of background, where people might see engineering as sort of a career that was a move up for the family. Was your father an immigrant?
Kahn:
No.
Geselowitz:
Were his parents immigrants?
Kahn:
Yes. Both parents were born in the US, and all four grandparents came from abroad. He was a CPA by background, taught accounting, and eventually went into high school administration. But the motivation for me going into engineering I don’t think was that clear for me. I didn’t know quite what to expect. I did know that working with technology was really interesting; working with mathematics was really interesting. But it wasn’t until I actually got into the thick of it that I really began to see what was actually entailed there.
Bell Laboratories
Geselowitz:
What made you decide to go to graduate school instead of trying to get a job right out of college?
Kahn:
- Audio File
- MP3 Audio
(432 - kahn - clip 1.mp3)
I went to work at Bell Laboratories. In fact, I worked at their headquarters, which, back at that time was actually in New York City itself at 463 West Street in what’s now a converted set of buildings out there, turned them into condominiums or something. But back then that was where the headquarters were. They had a research lab out at Murray Hill and another one at Whippany and at a few other places around the country. This was before Holmdel had been built up for AT&T.
I was hoping at the time to go to graduate school, but it wasn’t clear how we were going to finance that exactly. While I was at Bell Labs, I won an NSF fellowship, which let me go to a school of my choice. I ended up opting to go to Princeton, got accepted that fall. I was actually at Bell Labs only from roughly January of 1950 when I graduated from undergraduate school until September, when I went off to Princeton.
Geselowitz:
In that brief time, did you do any interesting work that later you felt had been useful?
Kahn:
Well, the people who were graduating from college back then in my field typically went into things like public television, electronic devices of one sort or another that were interesting, including various things that were de rigueur at that point. I opted to go into a kind of an operation with the research side of the business, because one, I like mathematics—it’s more mathematical. Number two, the particular job I took was in a group that was worried about the overall structuring of the Bell System and so it had kind of a national reach and had to do with some of the macro-technological and economic issues that one had to deal with.
I was in a group that was run by a fellow named Roger Wilkinson, a grand old gentleman of the Bell System. I used to think of him as the Walter Cronkite of the Bell System. He’d show up in cardigan sweaters, smoked a pipe, sailed his boat every weekend. He was a self-defined gentleman, the kind you associate with that kind of an endeavor. He ran this group in a very avuncular way, and it was really delightful.
I got some fundamental jobs to deal with that involved running the program computers and doing simulations and evaluating different alternatives for operation of various parts of the Bell System. It gave me a grand overview. And it wasn’t specifically in the development of particular pieces of technology, but rather understanding how the whole system worked from a global perspective.
Geselowitz:
The things that you were doing must have been early in the simulation days of computers—computer simulation of telephone systems?
Kahn:
No. These were mostly Monte Carlo simulations. They were mainly written in FORTRAN FAP at the time: FORTRAN Assembly Program Linker. When I joined Bell Labs, they had an IBM 704 as I recall. It was still a vacuum tube machine that we were using. Very shortly after I got there, they converted over to the solid-state version and the 709 or the 7090 series. I forget the exact numbers. I had a chance to watch that transition.
We were given desks in a room that involved four or six people, generally, at that time. Some of them were pretty world-class people, like John Riordan, the mathematician, was part of the group, and a number of other people who are very well known. The woman who had the desk to my right was in charge of building some of the computer programming language infrastructure for the whole Bell System. I got to learn how to do assembly language programming as well as FORTRAN programming and to link the two together. It was probably my premiere stint as a programmer in my career.
Geselowitz:
You rested on your laurels after that?
Kahn:
Right. [Laughs]
Graduate School: Princeton
Geselowitz:
If you could go anywhere with this because it was a portable fellowship, why did you choose Princeton? Who did you work with there and what did you work on?
Kahn:
I guess the best I can do by way of answering that is to say that, first of all, it was clearly an extremely good school. Still is. In fact it’s been nationally ranked up in the top one, two, or three for as long as I can remember. I don’t think engineering was ever its claim to fame. Math and physics were probably better known, as well as some of the other departments.
When I chose to go there, I was thinking of the following: number one, it was close to home, and I knew my family needed some personal support from time to time, so I could get back home weekends if I wanted. Number two, it was a change, and in very significant ways for me. Going to a place like City College is very city oriented. It was taking the subways or busses or whatever to get to school every day and coming back at night. It was a commute from home. It was dealing with the vagaries of city commuting and really mass education.
Princeton represented a major change in that sense. You know what kind of school it was like—in a rather rural setting in a small town, much more socially oriented rather than just cramming knowledge into your head. I think I saw the possibilities of broadening out in a number of dimensions, which I think is in fact what actually happened. One of the other things, which I didn’t know at the time but turned out to be a very big plus, is I had learned to play golf while I was in high school. In fact, I was on the golf team in high school. The graduate college at Princeton was right on a premier golf course, so it was a no-brainer. You could literally walk from your dorm room onto the course in a matter of a few hundred feet. So I used to play quite a bit of golf while I was out there, mainly in the interstices of writing the thesis and the like. That turned out to be just a very different lifestyle for me.
Geselowitz:
Sounds great. On the engineering side, what were you working on for your Master’s and/or your Ph.D.?
Kahn:
The first two years at Princeton were principally either the required course work or self-instruction. I used to closet myself in Fine Library, which was the mathematics library at Princeton. I’d be reading books and just sort of self-educating myself in many of the areas I cared a lot about. The required courses were taught in different places during that time. But the end of the second year, there were just certain exams that you had to go through to go on to the rest of the doctoral program. I think the actual selection of what I was going to work on didn’t really occur until I got through that hurdle. As I recall, the hurdle was very high. Very few people that came in actually made it into the formal part of the program. Most of them either dropped out or didn’t make it across the hurdle. It was a small group of maybe 35 in the incoming class. Only a handful of that many made it into the actual doctoral program.
The advisor that I had was a professor named John Thomas going in. I’ve come to appreciate him more and more over the years. He didn’t engage a lot while I was there, which, in retrospect I realize was a way of letting me develop on my own. My guess is I was probably not all that easy to manage anyway because I was pretty independent-minded, I probably still am. But he managed to guide me to the point where I ended up being successful in my program, for which I’m eternally grateful.
A very interesting thing happened during the course of that period, which I didn’t realize until years later when he was retiring. He had a retirement ceremony in Princeton that I went to, and he invited all of his graduate students that could make it. As each of them was introduced by him, he made some comments about each one of them. One of the things that happened while I was in the graduate program is another professor had joined us part way through, named Bede Liu, and I started to work with him part way through. When I finally wrote the thesis, it had a little piece of some work that I had done when John Thomas was involved directly, and some parts that were from Bede Liu. I never really thought much about it again.
At this retirement party, when he was introducing me, he said, “Bob Kahn is the only student that we’ve had at Princeton that’s written two Ph.D. theses.” It kind of took me aback, because I hadn’t realized I’d written two of them. What he said was, “He didn’t know it at the time, but he had done a sufficient amount of work to get a Ph.D. thesis, but he hadn’t spent enough time in the program. We decided we would have him work with this newly arriving professor, Bede Liu, and do some more work. He, in effect, did the equivalent of two Ph.D. theses during that project.”
This was before I had learned about the honorary degree. When I finally heard about the honorary degree, they asked me to make some comments at a later dinner. I said, “At first I thought that was a big mistake, but on reflecting on it, I came to the conclusion that it maybe was the right thing to do, because having written two Ph.D. theses, it was just recognizing the second one.”
Geselowitz:
You deserved the other degree.
Kahn:
Right.
Geselowitz:
So your double-length thesis, what was the topic?
Kahn:
I think it was titled "Some Problems in Signal Modulation and Representation." It was a project that had to do with modulation theory. The specific problem that I was dealing with there was how do you do simultaneous amplitude and angle modulation of a signal so as to minimize the amount of bandwidth that it occupies? Typically, we know that AM systems where you wiggle the amplitude and that’s what the signal is, or FM systems where you’re modulating the phase or some function of the phase. But what happens if you do both of them simultaneously in some coordinated way?
It turned out that the answer to that was essentially directly related to the solution of the Schrödinger wave equation from quantum mechanics. This is a quantum mechanical relationship that came out of that that I guess I really wanted to spend more time understanding, but never really got back to it. I think it’s still an open issue at this point, but it must have to do with some kind of minimization problem in nature that still remains to be looked at.
Geselowitz:
Why hasn’t that been pursued? Because it didn’t have an obvious practical application? You’d think a physicist would want to know.
Kahn:
Theoretically, it’s a very interesting question. I don’t know. Maybe nobody could see any reason for trying to figure that out. I just thought it was an interesting connection. The mathematicians have this all the time, when somebody working in one space somehow finds that a theorem of what looks like a completely different space allows them to find an answer. They know their problem and they know there’s a linkage, but they don’t know quite why it is, and they do spend time thinking about it.
Geselowitz:
Right, because you would think that the linkage might have something to do with some fundamental way the universe works, and it would be very fascinating to understand.
Kahn:
I don’t know. I can’t explain it. I’d love somebody to be able to answer that. But as I said, I’m sure it has to do with a minimization principal of some sort, but I’m not sure what it is.
The second part of the thesis, though, dealt with representation of signals. We had known at that time all about the Nyquist theorem, or Shannon sampling theorem, as it’s sometimes called. And that is if you took a signal that is limited in bandwidth, that if you sample it at least twice the highest frequencies, you can perfectly reconstruct the signal.
Various people had looked at variants of that, like if you took the derivative of the signal, could you still do that and get that and so forth. What I was able to do was to show that in one of the special cases, if you took n linear operations from that signal and sampled each of them at one nth the rate, that you could still put the whole thing back together, as long as the operations were what are known as linearly independent. That is, that you couldn’t take the same signal and sample identically in time at one over nth the rate, but you could do it with time delays because that’s kind of linear in a sense.
Geselowitz:
Plus a constant or whatever.
Kahn:
I was able to show that was the case. And also to really deal with the more general question of what was happening with signals that were not band limited and how you could deal with minimizations in that space, and also with issues having to deal with optimal choices of pre-filtering to minimize error rates in non-band limited signals, and dealing with issues of time jitter in both the sampling and the reconstruction. It’s a fairly comprehensive coverage of that. The thesis actually was sort of like Problem A plus Problem B packaged together, which is why I think John Thomas probably said that it was two theses. I didn’t see it that way at the time—I was just glad to have it done.
Geselowitz:
Did you have an opportunity to do teaching as a graduate student at Princeton? Did you teach undergraduates as any kind of assistant?
Kahn:
I don’t recall that there was much of that. It was really a pretty intensive program. It was very challenging in many ways, but mostly the challenges were self-imposed by the nature of the environment there. And it was one of the best experiences of my life. As I said, I think I grew in many different dimensions, both in terms of perspective, socially, and the like. Also, I got quite involved. I was sort of the social chairman there for a while. I used to arrange all of the social events, whether they be mixers with the various girls’ schools or inviting people to come down to talk. I ran the seminar series.
Geselowitz:
For the engineering graduate program?
Kahn:
The whole graduate school. One of the interesting things happened in—I would guess—about 1961. There was a strike at the Metropolitan Opera in New York, and opera was basically shut down. I’d always been interested in opera, used to go whenever I could. In New York in those days, it was very hard to get tickets. They were sold out years in advance and subscriptions were handed down within the family so you couldn’t easily get them. I used to show up on Monday morning occasionally, whether at four or five in the morning, and wait in line to get tickets that were so far up in the balcony that you couldn’t even see or hear, but that was the best you could do. So I picked up the phone one day and I called Rudolph Bing’s office and invited him to come down, talk about issues of running the Metropolitan Opera. And he did, to my surprise! He went through all the details of how to run opera, what the issues were, the form systems. It was really a marvelous experience for me.
Massachusetts Institute of Technology
Geselowitz:
Great. And then, as I recall, you somehow ended up at my alma mater, MIT. How did that happen?
Kahn:
When I finished with Princeton, I thought what would appeal to me was being at a small school. I took a look around at what some of the options were. I very quickly realized that they didn’t have enough critical mass at these small schools to do anything substantive. I very quickly decided to seek out some of the larger engineering schools. Of course, MIT was first on my list of places to seek out. They were actually pretty far along in the process. I think this wasn’t until something like April or May that I actually applied. They pretty much set what they were going to do for the next semester. In fact I got a very nice note from Peter Elias, who was then the chairman of the department--he’s no longer alive--saying, “Thank you for your resume. We’d be delighted to have you come and visit, but keep in mind that we’ve already made our commitments, so this is just so we can get to know each other,” which I thought was a very nice thing to do.
I guess it must have clicked when I went up there, because three weeks after I got back, I got a letter from them saying, “We found a way to support you. If you could come and join us in September, we’d be happy to have you do that.” That’s what triggered it. I went up and spent a few years teaching at MIT until I ended up taking a leave of absence a few years later, basically at the recommendation of one of the faculty up there who was actually running the communications group, Jack Wozencraft, who I got to know very well. He thought that what I really needed was to cut my eyeteeth on some engineering problems of significance.
The reason that that happened was, I think I was a pretty good mathematically oriented researcher—I knew how to apply the tools of the trade. In fact, while I was at Bell Labs, I used to go back summers. One of the summers I worked for probably the best applied mathematician that I’ve ever met in my life, a fellow named Stephen Rice, who was the guy who generated most of the theories about how to analyze noise in communication systems. I just learned so much from working there at Bell Labs, both in terms of the mathematics and how you can deal with it by computers and so forth.
That’s what I was doing at MIT. It was more of the same, picking an interesting problem. Every now and then, I’d go in to visit with Jack, who, as I said before, ran the group. He would, I would say one time out of four, get interested in what I was doing. We’d go onto the blackboard and work it out. Three out of four times, he would just not really show that much interest. I knew he was listening, but he’d continue what he was doing, working at papers or whatever it was.
Finally, I asked him one time, just out of curiosity, “Why is it that sometimes you get so animated and other times not?” He said, “Since you asked, I’ll go into it. Your problems are all interesting, and they can be harder or not. But if I don’t know what to do with the answer, I’m not motivated to work on it. If I can see what to do with an answer, I can get very interested in working on it.” I said, “How do you know? How do you know that an answer to a problem will be useful or not?” I’d say, “Okay, I want to calculate the answer to this,” and the answer is 4.3. He’d say, “It doesn’t matter. You could tell me 4.3, 0.1, or ten million, and it wouldn’t matter because I don’t know what to do with any of it. So it’s hard to get motivated.” I said, “Well, how do you learn that?” He said, “It’s probably experience. It may be a gratuitous suggestion, so you can deal with it if you want, but if I were you, I’d go find somebody who knows how to build the real thing. Go apprentice with them for a year or two, learn what that’s about, and then go on back.” I thought about that, and two weeks later I made my decision. I’m going to go and do that. That’s what caused me to take a leave of absence from MIT.
Bolt Beranek and Newman (BBN)
Joining BBN
Kahn:
I went to a little outfit in the Boston area called Bolt Beranek and Newman (BBN), which at the time was an architectural acoustics firm, but in transit. They were interested in getting into information technology. I think they had a small grant that might have just come in from DARPA on dealing with natural language communication with computers. They had been an early user of some of the digital equipment technology. While they were building up a core competency in state-of-the-art computing, communications was something they would have liked to get into. I think I was the first person they really hired in that dimension.
They also hired some people from Lincoln Labs around that time that were more engineering oriented, principally to take over an area of the business that Jordan Baruch had been running before. Jordan was later an assistant to the Secretary of Commerce, but at the time he had been running their medical information systems business. He had left shortly before I got there to become head of a company that GE had helped to fund, called MediNet. They were providing, among other things, what I knew was they were providing time sharing services to Mass General Hospital, but I imagine doing other things as well. This was a group that was going to pick up that area. They got very interested in networking too, from the implementation point of view. Of course, that’s what I started to work on when I got there at BBN. Eventually we learned to work together and help to create the very first packet switching net, called the Arpanet.
Geselowitz:
Now we’re getting to the juicy stuff that everyone, I’m sure, wants to hear about. But just exactly how, if you and they weren’t in that area yet, how did you and BBN find each other?
Kahn:
I think I called Jordan Baruch on the phone one day, and Jack Wozencraft replied[?].
Geselowitz:
Jack knew him and knew the kind of things he was doing?
Kahn:
Jack gave me a list of two or three people to go talk to, and Jordan was one of the ones on that list. I called him, he invited me over, we had a great chat. He had one of those can-do, let’s-get-with-it kind of attitudes. You could just see it. He was energetic, he was alive, he just sort of vibrated all that stuff. I thought about it, and eventually, a little later that summer, wrote back saying that I’d like to accept.
Actually what happened was I think I got a letter inviting me to join them shortly thereafter, except instead of coming from Jordan, because between the time I met him and the time I joined them, he had left; he wasn’t in that plan. He had introduced me to another gentleman there when I went to visit, named Jerry Elkind, who actually sent me the letter inviting me to come and join them and I took them up on it.
I actually showed up there and became involved in networking. I reported directly to Jerry. Within a few years, Jerry Elkind ran half of the company and Leo Beranek ran the other half. He was actually a very good person to have gotten to know. We stayed in touch for many, many years, although I haven’t seen him very recently. He ended up leaving BBN a number of years after I got there, briefly went to MIT, and then eventually went out to Xerox PARC. If you followed the whole story there, Bob Taylor invited him in as a boss to form the Contiguous Systems Lab, which did all the work on personal computing and the Internet--all that stuff.
Geselowitz:
At this point, I want to hear about the work for DARPA. You still thought, at this point, you were going to go back to academia after a couple of years cutting your teeth on the real world problems?
Kahn:
My thought was unequivocal. I was there to get a little practical experience, and I thought that in a couple of years that I would be for sure back at MIT. Absolutely. That was my mindset at the time. I had grown up as a kid thinking that, even though I went into engineering school, teaching was sort of the tradition in the family. My dad, as I said, had been a high school principal. It just seemed like the right match for me. I was very interested in things conceptual and intellectual. I liked working with kids. I liked teaching. That’s what I assumed would happen, but it didn’t quite work out that way.
DARPA Proposal
Geselowitz:
Tell me now about the DARPA work at BBN and how that changed your direction.
Kahn:
- Audio File
- MP3 Audio
(432 - kahn - clip 2.mp3)
I had been working on networking. When I started, my thought was that cutting one’s eyeteeth involved thinking through how you would deal with a real engineering problem. I was thinking about, if you were going to build a computer net, what would it look like? What are the issues you’d have to deal with, whether it’s error control or buffer management or routing or whatever the issues were. I read a whole series of memoranda on those subjects.
Unbeknownst to me was what DARPA was planning. I had not even been aware of DARPA until I joined BBN. It turns out they had been funding Project MAC at MIT. I knew about Project MAC, but I wasn’t as trained on tracking where the funding was coming from as I later became. At BBN, I knew that there was some DARPA support for some of the work that they were doing. Very likely, some of that DARPA support was helping me do some of that research, although I never quite understood how they accounted for things, probably either on overhead or directly charged to that project.
At one point, Jerry Elkind came in and said, “The folks at DARPA are very interested in building a network to link their researchers. They might like to know about some of the stuff you’re doing. Why don’t you send them a letter describing it, introduce yourself, tell them what you’re doing.” Which I did. I think I wrote it to Larry Roberts, who had just gotten hired at DARPA from Lincoln Labs. Bob Taylor talks about that story all the time in what he writes. Bob, at the time, was running the office. He had just hired Larry to become the Arpanet program manager, but he was trying to get a network going. I introduced myself, told them I had been working on networking for a while, thought they might be interested in some of it. I’d be happy to come and chat with them. I had no idea that what was about to happen was that Taylor called me on the phone and said, “Why don’t you come on down and visit? I’d like to get to meet you and chat with you.” So we did.
I got to meet Larry and Bob and those guys as a result of that, and shared with them what I had been doing. Lo and behold, I would say within a year of that time, they had worked up a whole package to get people to bid on building this network for them. And when the Request for Quotations came out, they pretty much thought they had specified what they wanted. It was general, like, “Deliver packets in a fraction of a second. They’re going to have these little nodes. They’re going to be linked together with 50 kilobits”—that sort of stuff was all prepackaged. But what they didn’t describe was how to build the actual network itself. So these were kind of architectural decisions that they had made on a very structural level.
When the package showed up at BBN, I remember one of the people, Joe Martin [Joseph Markowitz], who had walked it into my office, said, “It’s just arrived. Jerry said you were going to be looking for it.” I took a look at it, and they were laying out something that looked like I almost knew what to do. I had never built anything like that before, but I looked at it and I knew how to think about it inside and out. One of the people from Lincoln Labs that they hired was a gentleman named Frank Heart, who took over Jordan Barouche’s responsibility. Frank worked for Jerry Elkind as well. Frank showed up in my office one day and said, “I understand you’ve been thinking about networking.” I said, “Who are you?” We ended up starting a dialogue on the possibility of working together on it. I told him I’d be happy to share what I knew. It didn’t occur to me at that time that I would actually get involved in building the network. We were sufficiently far down, I was thinking about winding up my tenure there and going back to MIT. He said, “Look. I’d like to put a proposal in on this thing. We’re going to build it.” I thought, well, what a great idea—actually turn those ideas into a real proposal for an actual system. I thought that might be good.
I actually wrote the technical part of the proposal. I didn’t package all of it because a lot of it had to do with striking the deal through Honeywell to buy the computers and all of the stuff that was more management oriented. But the technical part of it I wrote with some very valuable help from Severo Ornstein, who was the hardware guy. As part of the proposal I had to describe what the hardware was. I wasn’t a hardware designer, and I didn’t consider myself a real time system programmer, but I could sit down with them and say, “Here’s how the hardware has to work. There’s this thing coming down that has to do this, this, this, this, and this.” He’d be designing in real time as I was talking, just manufacturing the specs on the spot. We put the whole proposal together, and it went in, and it worked. Here I was, trying to figure out?
Geselowitz:
Had larger players bid?
Kahn:
I don’t know. I never saw the actual submission. My understanding is they got something on the order of a dozen. Could have been a little more, a little less. Some of them were fairly big players, but they pretty soon whittled it down to two. That’s what I heard. I don’t know that for a fact. One of them came from System Development Corporation, who was probably a spin-off from the Rand Corporation; and the other one from BBN. They finally awarded it to BBN.
At that point in time, it was suddenly clear to me that there was another option here, namely actually work on getting this thing built. Because having written the proposal, somebody still needed to figure out what to do. It was sufficiently complex at the time, and just keeping it all in one head was a difficult problem. It may seem less so now, but that was all new stuff back then.
Building the System
Kahn:
I had been very involved in doing some simulation work, because I knew a lot of the issues were going to involve congestion control and routing strategies and all of that. We really didn’t know how to do that very well. We ended up proposing a pretty simple strategy for doing routing. I knew that there were options to do a better job, but we needed a simulator. We had actually created a network simulation program at BBN. I presented it at the Princeton Conference on Systems in 1969. I wanted to use that program to do analyses of what was going on. This was in the ’68 time frame that we had written that. You have to remember what the technology was like at the time. Most of us were using model 33 teletypes. This was a graphics display interactive system that ran off of systems that weren’t designed to give real time display control of anything. We kind of instrumented that whole thing.
I worked with a fellow named Warren Teitelman, who was kind of a world-class programmer, to actually build that. When it came time to actually using it, Jerry came to me and said, “If you really want to actually run that, you ought to go have it charged to this new contract that Frank has just won.” Frank wasn’t interested in simulations. His view was, “We’ve got a real network. Let’s go build a real network and then you can do anything you want.”
Geselowitz:
And see what happens [laughs]?
Kahn:
To some extent, you can understand that position. Except I don’t think we had won the contract yet. If we did, it was brand new and it was going to be months before it was out there. We had to make some decisions about what to do. Frank’s view was that the only way he was willing to let me charge it to the contract was to pick myself up and move it into his group, and sort of became involved in the system design aspect within in the group. I just couldn’t see any other option. I thought that was real interesting to do, so I decided I’d take another year or two, go do that. I picked up and moved over there. I reported to Jerry at the same time I was sitting at Frank’s group, so it was a little awkward as you can well imagine. Frank’s personality was such that he really liked to control everything that was in his purview, except that I didn’t report to him. That caused him probably some unease or discomfort.
In the final analysis, we managed to get the system built and deployed. We went through a couple of iterations on some issues. One of the things that was built into the system up front was it had the property that the system could lock up. Do you know what the “Boston Box” is? This is the mechanism whereby you can have an intersection, four cars, and each one is blocking the other car from going through. So there’s nothing that any car can then do in the Boston Box situation, because Car A is blocked by Car B, and Car B is blocked by Car C, and Car C is blocked by Car D, and Car D is blocked by Car A. In the absence of a crane picking them up and moving them all out of the way, there’s no action that can be taken.
I knew that could happen in the net, but I couldn’t convince anybody. Frank and all the other people there were of the view this is like an argument that, “We better worry about an oxygen circulation system in the conference room, because of some finite probability that all the oxygen molecules will migrate up to the lower left-hand corner and we’ll suffocate.” They thought that was kind of ridiculous and therefore we shouldn’t worry about it—you know, simplicity in design. I was sort of the lone fish out of water in this case, literally, arguing that, no, it was a problem and we’d have to deal with it.
Well, guess what? Not being the coder, it didn’t get coded. When we went to do the first test of the system out in the field, the very first thing that I did was to make it lock up. We came back and reported it and they didn’t believe it. “Can’t happen,” they said. So I said, “You were saying in the conference room about all the molecules are in the corner?—That’s what happened!” “No way!”—They wouldn’t believe it. It was six months later before they took their own little simulation and found that, yeah, it really was a problem. We then went through a whole revamping of the system to make it work a little more effectively.
Testing First ARPANET Nodes
Geselowitz:
These early tests were done where, in Cambridge?
Kahn:
The actual building of the first nodes--these were Honeywell’s plasticine [DDP-516] mini-computers. The actual testing of them was done out in California, where the first node of the Arpanet was installed at UCLA in September of 1969. That’s where I went out to do the initial testing. We eventually had a four-node net. That was one node a month. In September we put in UCLA; in October it was SRI; in November it was Santa Barbara; in December it was University of Utah. We had a little triangle in California and then a little branch out to Utah.
I went out to do testing on the four-node net after doing some one-node net testing, and caused all the lockups to happen. It just really took place out there. I went out there, stayed in the hotel out near the ocean, and every day went into UCLA to do the testing. In fact, that’s where I first met Lou Ferro, who was on Leo [Leonard] Kleinrock’s team doing the network measurement stuff. He was actually the counterpart for doing all the host-related testing once we finally got the hosts plugged in. Some of the early testing was just done on the node itself before there was any connection to their Sigma 7, and that was done with the teletypes and just playing around with the internal for the M code itself. That’s where the testing took place.
Geselowitz:
So different academic teams were also funded by DARPA, essentially to build the first one?
Kahn:
We had contracts with all of them, and they were trying to hook their computer research centers together. They mainly had big time sharing systems that were state of the art for those days. The question was how do you link them to do computer resource sharing?
It’s interesting. Katie Hafner wrote a book called When Wizards Stay up Late [1996]. I don’t know where she got her information, but it was kind of interesting. Her first version of that basically reported that when I came back from doing this testing, which I had done with Dave [David] Walden, who was also part of Frank’s group and later became president of BBN Labs. We actually reversed roles and Frank ended up I think reporting to him. We came back and reported it. What she said in her book was Frank came and said to everybody, “How could we have allowed this to happen?” or something to that effect, and that I had proposed—erroneously as was the case—that, “Why don’t we build a simulation to figure it out?” Of course, we had built the simulation a whole year before, which we had built to look into that before we built the net.
She got that wrong. And then she said Frank came back and said, “No, I don’t think we should build a simulation because we have a real net out there. Why don’t we just use it?” It sounded very reasonable except the logic was all backwards. I looked at that and I said, “Katy, I don’t know how you got this story so twisted.” She didn’t quite believe it at that point. I had to produce for her the proceedings of the Princeton Symposium to show that we had actually given a paper on this subject, that this was an existing thing long before we had actually started to build it.
Geselowitz:
You had given a paper on that modeling at Princeton?
Kahn:
We actually used it. We had pictures of it in operation. It was working back in ’68. When she finally saw the paper, she said, “Hmm. I guess so.”
There were those kinds of iterations in her book that kind of get after things that were egregiously wrong and see if we can get back on track. It’s a great read. It really reads very nicely. You can sort of sort out what’s exactly right from what’s not exactly right.
Geselowitz:
There’s a more dry, scholarly history of the Internet by Janet Abbate called Inventing the InternetI don’t know if you’ve read that one. She is at the University of Maryland so she’s close by. Right when I joined the IEEE History Center, she was there, and then she left soon afterwards to come down here.
Kahn:
I don’t recall it that well in my mind, but I know that she did a pretty good job. It wasn’t an across the board treatment of everything. It was just two or three things she was interested in. My recollection was she did a pretty good job of dealing with those things. Katy’s book is a great read. Pick it up and read it from cover to cover.
International Computer Conference (1972)
Geselowitz:
So once the Arpanet was working and there were four nodes and you had an architect, why didn’t you go back to MIT? You’d done what your colleague had asked you to do.
Kahn:
We had four nodes that weren’t working, which was the problem at that time. My goal at that time was to do whatever I could to get the system design fixed, if you will. It had 80% of what I had in there originally, but there were still some problems. I spent until the next year doing that kind of thing. I recall there was a discussion that Larry Roberts—Larry had replaced Bob Taylor running the DARPA office. I had suggested that we do some kind of a public demonstration. My original thought was the Spring Joint Computer Conference. Larry actually made some arrangements to do it at the First International Computer Conference, which was to be held in October, I think, of 1972. He asked me if I would make that happen. He was actually working with Steve Crocker at the time, who was at the office. Steve and I interacted a lot on those issues.
I basically thought that would be a lot of fun, do a public demonstration. This was exciting stuff. I agreed to do a public demonstration of it, which may sound pretty dull, but it involved figuring out how to put a first node of the national net into a hotel room in the Washington, D.C. area and leave it up for the better part of a week, bring all the researchers and network people who knew anything to that area. We went around and got manufacturers to deliver terminals. We had to get all the machines up and working. We had to get the protocols on the machines. It was the driving force for making the whole network work, writing scenarios.
We had people who are very well known by now as part of that. For example, all the scenarios were developed by a gentleman named Bob Metcalfe, who you probably know for his later work on the Ethernet out at Xerox PARC. Well, Bob was there figuring out what all the scenarios are. He used to draw in this almost cartoon-like character for me, really interesting to look at, big exclamation points. He used to make up those charts. They were really wonderful. We had John Postel organizing a bunch of it. We had a lot of the key people that were instrumental in networking to help make that a reality. I orchestrated the whole thing.
One of my counterparts here, he worked up at MIT, Al Dezza, was the one that took on the responsibility of actually doing the physical implementation of the system in the hotel, which meant the day that it opened, we had a standing start at like 6 a.m. Get the whole place wired, get the lines in, get the nodes in and have it ready to be over at like 1:00 that afternoon or 6:00, whenever it started. It was a very narrow window for actually doing all that. He did a wonderful job of bringing that all up and making it work. I was enjoying it.
Geselowitz:
It sounds like that was one of those seminal events, like people working with turning on the Colossus, where then everyone who was ever anything in the British computing industry had been there at that time. It was almost something where anyone who had anything to do in networking was in on this demonstration.
Kahn:
This is where it was. Computer networking was a core activity at that time, and everything else would have been less interesting at that particular moment. We thought through and did the demonstration.
Work at DARPA
Kahn:
Larry Roberts had been trying to get me to go down to DARPA for a while. I think he first made the opportunity in something like June of 1971. Remember I told you I had moved over into Frank Heart’s group. At that time we had figured that the network was really not ready, not really done yet and I ended up moving back into the research group that I’d come from shortly thereafter. I decided I would stick it through and make sure everything worked well.
I didn’t really think of myself as a government employee or moving to Washington. But as it drew closer and closer, I finally ended up deciding maybe that was a good thing to do. I’m not sure I can articulate all the reasons for doing that, but I ended up going to DARPA right after that ICCC demo. I think October 29th was my start date at DARPA, although I don’t know that I actually showed up on that day. I think I went up to Boston, closed some things up, and ended up voting in the election of 1972 up in Cambridge and then getting on a plane. My first day was election day.
Geselowitz:
That makes it easy to remember.
Kahn:
That was the Nixon-McGovern race.
Geselowitz:
Right. Well, don’t blame you. You voted in Massachusetts, as we used to say!
What then happened, now that you were on the DARPA side, on the government side, where did the action sort of move with growing this network, of perfecting this network that had built?
Kahn:
Actually, there was some real conflict inside of BBN before I left. That’s part of the reason that I chose to go to DARPA. It had to do, I think, with the way things were developing with Frank and his group. He was very interested in controlling everything that had to do with networking and kind of left me at loose ends. In going to DARPA, I decided I was going to have nothing to do with networking at that point. I had arranged with Larry Roberts. They were just starting a new program in automated manufacturing. It sounded really interesting: how do we use computers to help in design and manufacture of virtually anything? So I agreed to go down to run that program. It was a $100 million dollar new initiative with the Congress.
Probably one month after arriving or less, it was announced to me that Congress had killed the program. It wasn’t going to happen. I had the very agonizing decision, then, of deciding do I really want to be here, since what I came to do was not going to happen? I think I probably could have gone back up and gone into teaching again at that point. But this was probably early 1973, and what Larry said to me was, “Look. You know as much about networking as anybody, if not more. Why don’t you stay on and figure out some new things to do about networking?” Well, I had been thinking about it little by little in the background, because one of the things that I was sort of doing while I was getting started here was thinking how we could get a packet radio network going. Larry had started an effort on applying packet switching to satellites. I insisted I didn’t want anything to do with the Arpanet. I just didn’t want to have those issues come up again. So I didn’t. I think the program manager for it was either some combination of Steve Crocker or later Craig Fields. I eventually said, “Fine. I’ll do that for a few years,” again thinking I’ll just do that a year or two and then I’ll head out.
I ended up setting up the packet radio program. I revamped the satellite program something fierce, ended up starting a number of other programs in what’s now voice-over IP, packet-type speech recognition, and end-to-end security—some related problems like that. We had gotten enough support to get those programs up and launched. Here I am sitting with the notion of multiple networks, and I’m trying to think, how would we actually do anything interesting with them if we don’t connect them? Because the computers were all very big. If I have a radio net, what am I going to do with the radio net? I can maybe plug a terminal into a little interfacing computer, but I’ve got to get to some big machine to do anything interesting.
So I had the problem of trying to figure in those nets, and I had a basic architectural approach to deal with it, which is what became the Internet. The one problem that I did not know how to deal with back then was how to take the functionality that I knew needed to be in the computers and actually get them into all these different machines? Because they were all different. That was exactly the problem that Vint Cerf and Steve Crocker and others had worked on in developing the original host protocol for the Arpanet. You have to remember, that protocol was one that left you with just a wire coming out. The problem was that you sent something up that wire, and that wire was connected to another net. What were you going to do to direct where a packet should go? Furthermore, the Arpanet was a perfectly reliable network, or supposedly so, so what were you going to do in a radio case where you were behind a mountain and a signal couldn’t get there or in a tunnel or somebody jammed you, or just errors in propagation had caused errors to occur. How was this all going to play out?
We knew we needed a better protocol, something that involved some kind of an error detection and retransmission or correction scheme. We knew we needed to have a more robust—or at least I did, at that point—addressing mechanism. I had a pretty good idea of what we needed to do, but I didn’t know how to take that and actually figure out how to get it in the first machine. That’s why I asked Vint to work with me on this whole mission.
Vint Cerf and the Growth of Arpanet
Geselowitz:
He was where at this time?
Kahn:
Vint had just finished his Ph.D. at UCLA the year before and had just joined Stanford as a new faculty member. He was at Stanford at the time. He and I clicked instantly on this. We had known each other before from all the Arpanet testing, and we started to flesh this out. Vint is the kind of guy who likes to roll up his sleeves and let’s get with it. I thought that was a breath of fresh air.
We had meetings at my office at DARPA. We had meetings out in California in his office and walking the streets of Roslyn, Virginia, and of course, that famous session where we actually wrote the TCP/IP paper, which he then had typed up and promptly threw away the manuscript. We don’t actually have a copy of the actual handwritten version that he and I wrote. His secretary typed it up, but I guess didn’t think to keep it. It would have been a nice keepsake.
That’s how we actually got together was that thing. Through the interaction, what I had been proposing and what he knew from before actually got much better. The synergy between the two of us caused it to be far better than either one of us could ever have done alone. We then worked together and we gave contracts to Stanford, University College London, and BBN to do some initial implementations and do some testing. Of course, it started to roll out experimentally, I would say by the late ’75–’76 time frame. We did some three-node testing in ’77 and it was reported in that book, A Century of Innovation, produced by the National Academy of Engineering. But there were probably two network tests that were done in either late ’75 or ’76, most likely at BBN where they were building some of the gateway technology. No real record of exactly who did what, when, or how much. I have in some of their technical reports, which I haven’t gone back to really look at.
That’s sort of how things got started. The Internet really remained in a very experimental form until we decided that we would cut the Arpanet over to the use of TCP/IP in the early 1980s. The actual cut-over was supposed to be on 1 January of 1983. In fact, it didn’t all happen like that. It took about six months before everybody converted over, mainly because there were some that didn’t really believe it until the day it happened; others were just slow in doing their importations. But within a few weeks we had a big chunk of the Arpanet converted over.
People were plugging in local area nets into the system experimentally before that. This was a way to make all of that local area network traffic and computers become actually part of a working composite system. Remember, the IBM PC didn’t really come into being until the early 1980s, and even then it didn’t have enough power to really run TCP/IP protocols. These were mainly experimental workstations. Sun Microsystems was one of the first out of the box to have it bundled in. By the early 1980s, there were pockets of real capability developing, and they were all on LANs [Local Area Networks]. The Internet was sort of a lot of LANs connected to the Arpanet with linkages to satellite nets.
Geselowitz:
Which had been your idea. If you had all these LANs running around, and how could they speak to what architectures each had.
Kahn:
Well, no. We started out with just the three long-haul ones: the Arpanet, the packet radio net, and the satellite net. The LANs didn’t come along till a little bit later. We weren’t thinking personal computers and workstations for LANs at the time, but we were thinking general purpose nets. Of course the architecture of the Internet was just sufficiently general purpose that it didn’t matter whether it was a LAN or anything else, you could just sort of plug it in and keep going.
Surprises
Geselowitz:
As it grew, were there any surprises or things you didn’t anticipate in your design of the way the net grew?
Kahn:
You mean like the break up of the Bell System and things like that?
Geselowitz:
All right. Yeah. How did that impact it, as you see it? That was a political, not a technological, transformational surprise.
Kahn:
I think that, to be more direct about your question, I think the very first thing that was a big surprise was all of the LANs. Instead of working with what we thought might be five or ten big nets that needed to be connected, we suddenly had the prospect of hundreds of thousands. In fact now it’s probably millions. When we did the original addressing, the Arpanet addressing is fifteen-bit addresses—perfectly good for a network that might have sixty-four or 120 nodes. We knew with the Internet we’d need more, so we said, "Let’s make the Internet addresses thirty-two bits long." We said we’ll make the first eight bits specify which network and the other twenty-four specify which machine on that net. Again, more than we thought we’d ever see in our lifetime. "We only have ten nets, we’ve got 246 more to go." We thought surely this is good enough, and very soon we realized it wasn’t, so we now had to look at ways of accommodating what now were Class B and C networks and just broadening that out. We could have networks that we have a lot of them with only a few machines on them, like 256, and some that had lots of machines but they’d be national in scope. That was the first thing, I think, that was the big issue. Other issues were technical details, which I don’t think we need to go into.
I think one of the most interesting things that developed part way through was after Vint joined us at DARPA. That was really a godsend for me because it freed me up to worry about other things. When I took over as DARPA’s director, Vint really had the responsibility for the day-to-day networking stuff. Although I had brought in a fellow named Barry Leiner to work with us also. He eventually took over some of the projects that Vinton Cerf had been working on. During that period, I got very concerned that if anything happened to Vint, it would bounce back to me to worry about all the details, and I was busy setting up the strategic computing program and managing the office and the like. I suggested to Vint he set up a kitchen cabinet, educate the community about what’s going on. I don’t know whether he didn’t like the term or whatever, but he came back and said, “Why don’t we set up a group that can help other people learn how to get onto the Internet?” We set up, really under his leadership, something called the Internet Configuration Control Board. They were going to learn about all the details and help people that wanted to get configured to it.
You had to roll your own back in those days. You were using biometrics-currency product systems. If you want to put in TCP/IP, you’ll have to figure out how to get that into your operating system and then link it to all the applications. It was a job for a wizard, as Katie Hafner has said. He set up the ICCB. It consisted of twelve people, mainly real implementers with hard-core experience in building stuff like that. He would meet with them regularly, and I’d often sit in those meetings—in fact, I think I sat in on most of them—and tell them what’s going on. There were twelve people in the community who knew what we were thinking. It was so interesting in those meetings that people would say, “Can we sit in? We won’t say anything. We’ll just be a fly on the wall.” Vint left in late ’82, September or so of ’82, to go to MCI, where he ended up being the bench engineer on the MCI net system.
I took over running the program again and trained Barry Leiner to take it over. When Barry took it over about a year later, he said, “I don’t want to have to invite 200 or 300 people to have a meeting of a core group of twelve. It’s too complicated.” So he and Dave Clark, who was then chairing the ICCB, set up the structure that’s really persisted to this day, which they created something called the Internet Activities Board (the IAB) and they set up initially ten task forces underneath it on areas like routing and end-to-end protocols and privacy or security—and whatever the topics were. One of them was called Internet Engineering. I think it was headed by Ed Cain of DCA at the time. Its job was to maintain the punch list for the Internet, all the things that needed to be done in order for the Internet to get up to speed. It was like, “We need a new board at Stanford, and this thing at BBN is broken. We need to fix it.” It was like if you were buying a house—a bunch of small things that needed to be done in the short term.
As this whole structure evolved, it had the nice attribute that people wouldn't have to sit in on a meeting at twelve; they could go to these working task forces. The number grew. In a few years, there were like 50 of them. The group of 12 didn’t want to manage all those task forces, so they had this punch list group, the Internet Engineering Task Force to manage it, and that’s what led to what’s now known as the IETF. That took on a life of its own. Eventually, after I left DARPA and Barry left DARPA, we got a cooperative agreement from NSF which had taken the mantle of leadership over to DARPA in this area. We got a cooperative agreement to help structure, run, and organize that whole process, and this was networked CI (community information).
That’s really the story of what happened at DARPA in terms of getting into the Internet.
Leaving DARPA
Geselowitz:
When and why did you leave DARPA, then?
Kahn:
People normally don’t stay at DARPA. It’s intended to be a place you come in, spend a few years, and leave. In fact, I had thought about leaving several times, but I kept getting my arm twisted. I ended up staying there about as long as anybody ever has. I’m sure that I was the longest tenured office director in IPTO. In the mid-1980s, we had grown the office from perhaps the smallest office in all of DARPA when I took it over to perhaps the largest when I left. Part of that period involved spitting out all of the technology from DARPA that was involved with the FBI program. That might have been the biggest one, had it stayed. But with that gone, I think we had pretty much become if not the largest, one of the largest. In today’s knowledge, it was probably three quarters of a billion or a billion dollar equivalent. Back then I think the actual dollars were in the $200 million range, although some part of that money had actually gotten siphoned off into different offices. We created the Engineering Applications Office and the like, but we pretty much had the dominant research support for the whole field.
What I wanted to work on was information infrastructure, equivalent to building the superhighway. That’s something that Al Gore later popularized, but I was calling it the National Information Infrastructure. The Reagan administration at the time was very supportive of this, but not with government money. They thought the private sector should do it because they thought it smacked of industrial policy. The private sector viewed it like clean water and fresh air—they were all for it too, just a question of who paid for it.
Development of the Internet
Geselowitz:
If AT&T has been broken up, then you don’t have a monopoly that has an interest in acting like a quasi-governmental agency.
Kahn:
It wasn’t clear whose job this was. I finally stepped in—having been told that this was not appropriate for government to do. Remember, we were also running into the Gramm-Rudman era, you know, budget cutbacks. It turned out in retrospect there was more and more pork showing up, especially allocated projects and more classified ones that I don’t know much about now, but I believe actually ended up put in the agency. All of that portended that the big budget build-up going on--in going from the smallest to one of the largest, if not the largest--was going to run into a lot of resistance, including from the other offices, which wanted some share of this.
In fact, the reality was the budget probably came down by a factor of two over the next several years under my successor. Had I stayed, we would have been in a very defensive mode, very hard to get new things done anymore, and I couldn’t have done what I wanted to do, which was work on national infrastructure. I set up CNRI essentially to try and provide a national leadership role for that area, and I think we’ve done a very credible job. I just was unable to raise the amount of funding that I was looking for, which I wanted something that was on the same level as I had at DARPA. Not possible. I found out I had to raise it from industry, but we did end up eventually with government help as well, helping to move the whole agenda forward.
We got initial funding from industry to get started, so there was no government money involved there. We got it from a broad range of companies that normally might be competing with each other, but were anxious to see this happen, but at a small level. And the very first government activity that we had was a grant from the National Science Foundation to help create five very high speed networking test sites in the United States. We called them Gigabit Test Sites, where the line speeds would be on the order of a billion bits per second or more. This was in an era when we hadn't even gotten the Arpanet beyond 50 kilobits. They were just now moving the Arpanet to the one and a half megabit range to an effort at NSF called the NSFnet Project.
Geselowitz:
That was around what year?
Kahn:
1988, perhaps. We proposed it starting in ’86 or ’87. Dave Farber was a major collaborator of mine at the time. He was at the University of Pennsylvania proposing that NSF do something like this. NSF finally agreed to actually make something like that happen. We got a grant that ended up being a total of $20 million dollars, which for NSF was a huge grant. It was our first one with the government, which is doubly amazing. We spent most of that money on subcontracts with universities and other participants. The project itself made the front page of the New York Times, right-hand column. It was the lead story in June of 1990 because the Times had somehow calculated that the industrial contribution was worth something like $500 million dollars, both in terms of contributed facilities, fiber optic links, equipment, people, cash, and whatever.
We set that project up and ran it. I think it was a smashing success. We not only showed that high speed networking was viable, we built five independent networks. They had a big spread on me in the Business Section of the New York Times about three weeks later with a photo taken with a fish-eye lens from a very low level, so it made me look like I was about 300 pounds bigger than I am. But it was a very important thing. It put ATM on the map. It was the first demonstration of an ATM switch in a carrier central office. We demonstrated the first all-optical link between Illinois and Wisconsin. AT&T did it—I said “we”. AT&T did it with using their erbium-doped fiber amplifier technology, I believe. We funded the University of Illinois to build what became the Mosaic browser, which opened up the World Wide Web. It was a major effort.
At the same time, we got a cooperative agreement from NSF to help manage the whole Internet sandwich process, which we still do to this very day. So those were the very first two activities.
The World Wide Web
Geselowitz:
You just mentioned the Web. In terms of before what I said and that I had mentioned to you earlier that Marconi Foundation has certain questions they wanted to make sure got answered. One of them is if your contributions were used in a way that you did not anticipate when you were working on them. What would you want to say about the relationship of the World Wide Web to the Internet and where that came from?
Kahn:
I had many meetings with Tim Berners-Lee, who is often credited with that. Tim did that work at CERN in the late ‘80s, early ‘90s. In fact, it was not really moving at a very rapid rate until the Mosaic browser came out of Illinois. That really just spearheaded the whole thing, because suddenly it became very easy for people to use and it was something you could actually get and download onto your local machine and make it work.
The World Wide Web has become an extremely important means for people to access information. What most people don’t realize is, rather than the last word on information access, it’s really one of the early ones. Just like the Arpanet was an early step into networking, many people thought this was going to propagate around the world and be Arpanet forever. Well, the Internet showed that that was probably not the case. Just like with the Web, it’s a very important first step. That’s how many people access information, but it’s basically dealing with the public kinds of information for most people—whatever they can get to people and make available to them. It’s an application, as I see it, on top of the Net. It doesn’t rebuild the Net. It’s sort of using it for like connectivity in order to allow people to access resources. As a social phenomenon, it’s almost mind-boggling what’s happened. It’s energized the whole world to put up all kinds of information. There’s hardly any topic you can’t find useful information about on the Web. And of course, all the search engines which go through and gobble up everything and index it for you, has made it pretty easy for people to gain access to the public side of it.
What I think surprised me was how rapidly that happened. It just went like gangbusters. On the other hand, by the late ‘80s, the Internet was already growing at 100 percent per year. You could use as a measure a domain name’s traffic, for example. It’s very clear the Internet would have continued that growth, I believe. But the Web just accelerated the process. When you look at it today, I think the two main things that people rely on the Internet for are e-mail and Web access. They still remain the two biggest applications, as far as I can see today.
Impact of the Internet
Geselowitz:
Would you say that’s transformed society over the past decade?
Kahn:
E-mail has been around since the early 1970s. What’s transformed society was the rapid propagation of networking, the creation of commercial Internet service providers, which really started in the late 1980s with places like UUnet and PSInet and a variety of others, both in the US and around the world. That’s what made it an international phenomenon. The e-mails were there right from the get-go, so I think what it did was enable people to essentially rely on that as an alternative to necessarily sending letters or calling on the phone. What the Web did was enable people who now were motivated to get on to the Internet through whatever mechanism, to be able to access all kinds of public information of all types. I think that has made a huge difference in how things work, because if you wanted to track what was going on in Washington, you had to either come to Washington or know somebody here. It was hard to get official documents. You didn’t even know what was out there. If you wanted to deal with your local tax regime, you had to go there, and people had to take time out of their lives to do that. Today, almost all of this information is just directly available. You simply go on your machine and download it.
In terms of public access, it has made an enormous difference. Information is now out and available. If you missed the local golf tournament and the news is finished and you don’t want to wait until eleven o’clock, you can go on the net and find out what’s happening. It’s essentially given an almost unlimited number of channels of information flow so that you don’t normally have to rely on the five or two or ten local TV channels or just the local radio. You can get what you want when you want it. It may not be in the same form and you may not get the full video of the great golf shot the hole, but it was unheard of even a decade ago.
Geselowitz:
And isn’t that what, if any, of the political impacts would be around the world, because a lot of historians of technology have written about the impact just of the fax machine and the copier on the fall of Communism and the fact of the difficulty of a centralized government controlling information when people can make duplicates of it and send it around. The Internet does that in a much more incredible scale of allowing sort of instantaneous multiple dissemination.
Kahn:
You could make whatever argument you want about what caused what. I heard at a National Academy Annual Meeting that the Internet was the most important invention because it caused the collapse of Communism. I have trouble with that argument, because it seems too simplistic.
I think it’s very hard to fight the tide of history. When the tide of history is more efficiency, better productivity—I mean the unsung story of the Internet is how much difference it’s making for businesses to improve their productivity. Yes we had an Internet bubble and a lot of the businesses that really had no good business models and were selling advertisements and eyeballs or whatever went by the board. But when we look at what difference it made for companies that knew how to leverage it, they could suddenly be selling world wide 24/7, possibly with even a smaller sales force. Companies like Cisco were using it almost exclusively to sell their merchandise. Companies like General Electric had learned to do B2B kinds of transactions more effectively. You even saw new kinds of phenomena arise, like eBay is a phenomenon that didn’t exist before in that form. You could go to local auctions, but where could you go and find anything?
Internet Economics
Geselowitz:
I hate to say this to the world and the experts, but just for myself, as a layperson, the people who have succeeded on the Internet businesses are people who knew how to run a real brick-and-mortar business and used this to enhance their reach.
Kahn:
Except eBay.
Geselowitz:
Except eBay. It’s like saying you can go to a flea market anywhere, but here’s your chance, saying, “Let’s bring all the flea markets of the world together virtually.” That was a brilliant, brilliant idea, however they stumbled on it.
Kahn:
They were not the first to think of the idea. You might actually ask, “Why did they succeed?” I’m not sure I could give you the story of why them versus somebody else. But you’re certainly right, that was not an example of a brick-and-mortar thing that was just made more efficient. We will see what the story is about the search engine business. Google claims to be about to go through an IPO. We’ll have to see how that plays out. But this was not something that existed as a brick and mortar thing. Maybe they’ll pan out. I think it’s not the search mechanism that’s really the interesting thing in Google because they don’t charge for that. It’s got to be something else. It’s got to be the way that they either organize it, or maybe it’s their dealings with advertisers that makes the difference. There’s something in their business model that’s not just what it purports to be on the surface at the site there.
I talked to somebody who works for one of the large candy companies and whose name is instantly recognized, and I was just talking to him about what that business was like. What he said to me was, in order to really succeed at this, you need to be very clever about a lot of things that people don’t normally think about. When he explained to me what they normally do, I thought, “Yeah, that’s it!” I never thought about it that way. If you buy yourself a candy bar, it must be that they figured out the economics. It turns out their margins were all in the commodities market, futures. They don’t have to actually make the candy bar. They made a lot of their money on knowing when to invest in the chocolates and the sugars and the like. The candy bar was just sort of the icing on the cake. It’s like American Airlines, another company. I was talking to some of the senior people there. They fly airplanes and most people think that’s how they make their money, but I think they make most of it on information services that they provide, whether airline reservations on their own Saber System or whatever. I don’t know the economics of it, but here’s a case where they fly the airplanes because they like to fly airplanes in part, because they’ve done it and it’s a way of keeping their credibility up.
Geselowitz:
Their brand name or whatever, right?
Kahn:
Yes. I’m not the expert on this and you shouldn’t think this is a definitive statement. But sometimes when you look at it on the surface, what the real business is and what makes it work is not what it appears to be. That’s the comment I was thinking about a place like Google. EBay, I think, is what it is. But Google’s got some other way of dealing with things that probably makes them effective. You’d have to go speak to them to find out what it is.
The Future
Geselowitz:
Finally, on behalf of the Marconi Foundation I wanted to ask you, how do you see what you’ve done as a basis for future science and engineering? I guess they’re asking you just to look a little bit ahead now. We’ve got the Internet. We’ve got the World Wide Web. What do you think the next generation is? You’re not ready to retire. You’re here working at CNRI. But what are you working on and what do you see other people working on that’s going to build on the PC?
Kahn:
Let me say three things. One is the whole notion of wireless I think is going to become even more ubiquitous than we know. Some people have pretty much made the transition to the cell phone world, but I think that’s going to continue. It’s going to be a means of information access just on an ongoing, anytime, anywhere basis provided you’re willing to enable that. You’ll have administrative control, but they’ll be just wherever. That has a lot of interesting science to it, because if you can do that, if you know where you are, there are all kinds of questions you can ask about that are location-dependent, like who’s near me and what’s near me and how do I get from here and there? Things like that that normally might be problems.
Second thing is that I think the whole broadband world is going to continue and that you’ll have that as an alternative, whether it’s all fiber, or whether it’s on copper wire for a while more, or whether we find some new way of doing it. It’ll mean that more and more information access is possible wherever you are, in different forms. On a lower bandwidth line, you may get smaller images or less data, or where it is critical you may get more enhanced ones in different places. But I think that will happen.
The third thing—I think the thing that I’ve been working on that is most interesting—is what I call the digital object architecture. It was my attempt to try and reconceptualize the Internet around the management of content. The whole idea that today you have say a wire in the original context of the Arpanet, or a machine in the context of the DNS, or a particular file. What do you do when you want to be in a world where you just track your material through cyber space? Maybe put it here, maybe there tomorrow. It could be anywhere. The whole idea was that you no longer are dealing with machines and files, but you’re actually identifying data structures themselves. If every data structure, which we call a digital object, has its own built-in, unique identifier, and you have a system that lets it persist in cyber space indefinitely, then you have the basis for managing information over the long haul. You can deal with your own personal information as well as with public and private, so you can help identify or store your financial information, your medical information. The military can use it for classified information or sensitive information. It allows you to organize things in a way that doesn’t require somebody’s search mechanism to go out and find your stuff, and hopefully find the right things. You can actually structure it yourself using these identifiers and create your content and manage it over the long haul and never have to do it again.
Geselowitz:
It’s looking at cyberspace as if it was sort of a reality independent of the hardware? It doesn’t matter where I’m logging into cyber space? It’s not like, “Oh, this is my C drive and I better copy this up to the server.” It’s just all there?
Kahn:
It’s out there. You need to know how to identify it. That is, you can’t throw it out and expect that you’ll find it without something. Part of the notion here is that there are metadata registries that you will have created yourself that let you navigate through it, or places where you actually build the records of what you have. Your financial records—here they are, but they’re in the form of catalogs of identifier stuff. It’s in effect your own personalized metadata registry.
Esther Dyson, the “futurist,” just wrote about this her monthly report called Release 1.0. I’d be happy to give you a copy. In the September 2003 issue she talks about lessons learned from DNS, where she used to be chairman of the board. Then she goes into the transition from names to numbers, and then she talks about the whole digital object architecture. In fact, I just was notified that as of last year, it won the Digital ID World Award for Balancing Innovation with Reality. It’s in the field of identity management. It’s a very powerful system for letting you do that.
Copyright Issues
Kahn:
I got interested in this back in the ‘80s, when we were starting to work on mobile programs and the Internet. In the 1990s, we ended up working with the US Copyright Office on applying that particular approach to the management of copyright materials at the Copyright Office, where if you want to get a copyright registration, you have to apply for registration of your claim in copyright, and they’ll register your claim if you’ve got it. You’ve got to supply them one of these objects. What they do is, they will basically hang onto it there over the long haul. My view is if they ever wanted to, the digital world can create an enormous amount of stuff real fast. Maybe they shouldn’t have to keep everything in the Copyright Office. They may choose to, but can we give them an option where they don’t have to? If you’re creating software, do they really need to keep all your software there? You can’t look at it. All you can do is maybe execute it. But if it’s intended just for a record of what was done, maybe it would be better if you let the people that created it hold onto it. Maybe they’ve got to submit it once so you can see what it is, but if you know what the identifier is for it, and they care to have their registration held up and maintained, then if they can click on the identifier and go find this version, wherever it is in cyber space.
Geselowitz:
20 years late.
Kahn:
But that’s when people first saw it. It was visible in the marketplace that year, for the first time.
Deep Gratitude
Geselowitz:
I thank you very much for an incredibly fascinating session. Is there anything else? I think we’ve covered everything I wanted and everything that the Marconi Foundation wanted. Is there anything you’d like to add?
Kahn:
Well, I’ve been privileged to be part of this networking revolution almost literally from day one. In fact, from day one of actual implementation. Nothing like this could have happened on such a scale without a lot of people being actively involved. So from the point of view of creating nets where I think I played a very key role as principal designer of the Arpanet, not to take anything away from Larry. Because it wouldn't have happened without Larry Roberts, who really architected the whole thing and produced it and came up with the concept. I was actually the detail designer of how to make it work. To all the people that helped in the creation of the Internet, of which Vint Cerf is my sort of alter ego on this because he and I really did the basic work on that together. There are thousands of people around the world who have been involved in helping to make that a reality, many today who would probably not even recognize our names as original contributors because they probably weren't even born when we did that, that the whole evolution of this thing has really been amazing to me.
If I had to predict it, I never could have predicted it as it turned out, because we started it out as a little research experiment, three nets that DARPA had been building and watched all of these things that pulled around us. No one could have predicted the chain of events: The creation of the personal computer market; the willingness of the Defense Department to open up their networks to broader scientific use; the willingness of the National Science Foundation to open to all of science and education; the congressional willingness to open up the NSF Tech to commercial use; the evolution of the Dell system, which brought more competition in the market. If you had known that up front, it never would have happened. Probably the reason I think we were successful was that so few people saw it as that meaningful when we started it; whereas today in the world of the digital object world, everybody’s dealing with managing information, whether it’s a library community, the publishers, the banks--so it’s a much more difficult road to hoe. But what happened with the Internet probably could not have happened if people understood what a big impact it was likely to have. So for that, I think I’ve been very fortunate to have a very special place in history, and to have watched it really from square zero.
Geselowitz:
Thank you and I think we’re very fortunate to now have your words on that on tape. So thank you very much.
Kahn:
Thank you, Michael.
