The Internet’s grandfather, an inventive fraudster with many descendants
Lester D. Earnest, Senior Research Computer Scientist Emeritus, Stanford University
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License.
Please note that this is a work in progress and some parts will likely need to be expanded or otherwise modified. If you have any questions or see something that doesn’t look right please send a note to les at cs.stanford.edu.
Introduction. If you have read some of the hundreds of articles or books about the history of the Internet you likely have the impression that ARPAnet was the first interactive computer network, that it began operating in 1969 and that it was based on packet switching technology invented by the late Dr. Paul Baran, or by a British physicist, the late Dr. Donald Davies. You might also have the impression that the point-and-click graphical user interface was invented by Apple or that it was really invented by Xerox PARC or by Doug Engelbart at SRI. You may also think that computer networking depended for its success on the introduction of personal computers and that ARPAnet was developed by the U.S. Defense Department with the goal of it surviving a nuclear attack. However none of that is true.
You might even have the crazy impression that Al Gore helped create the Internet. In fact he actually did.
Widespread misinformation about networks is evidently attributable to three main things: changes in terminology over time, the tendency of authors to think that a given field started when they first engaged it, and the tendency of innovators to expand the scope of their contributions in retrospect. I share the latter two tendencies but, having entered the computer field in 1952, I have a longer perspective than most and have kept a lot of documentation on early activities, which tends to keep my feet on the ground.
One of my goals in writing this account is to fill in some of the missing history of computer networking. Another is to refute some of the bigger distortions and fabrications and to point out that even though capitalistic free enterprise has played a role in the development of the Internet, its basic technologies all came from inventors motivated by altruism, who patented nothing, gave away their ideas and helped others develop them. A third goal is to expose some of the more crooked financial transactions involved in this evolutionary process and to point out that similar practices are continuing today.
There have been four big steps so far in the development of computer networking, whose developmental and operational time spans were as follows;:
First computer network (SAGE air defense system, 1953-83)
First general purpose computer network (ARPAnet with email, 1967-90)
Standardized interconnection of networks (Internet, 1973 on)
Standardized user interfaces (World Wide Web, 1990 on).
By chance I contributed to each of those developments and, just to brag a bit, am evidently the only person in the world who did that. Having entered the computer field fairly early I was privileged to have created the first of a number of new technologies, sometimes by accident but generally with a lot of help from my colleagues, including the first spelling checker (1961), search engine (1961), cursive handwriting recognizer (1962), autonomous road vehicle (1967), online restaurant reviews (1973), network news service (1974), and social networking and blogging services (1975), though the terms “social network” and “blog” did not come into use until about 25 years later. For more on these and other innovations as well as some embarrassments see my list of Brags and Belches.
The first computer network was an air defense system called SAGE (Semi-Automatic Ground Environment) that was initiated by the Massachusetts Institute of Technology (MIT) with financial support from the U.S. Air Force. I began helping design it in 1956 and it became a technological marvel that included six levels of packetized data communications networks that were more tightly connected than the later ARPAnet and Internet. I was assigned to design guidance and control functions of missiles and manned interceptors based on my earlier experience, having graduated from the California Institute of Technology (Caltech) in 1953 and spent 3.5 years as an aviation electronics officer in the U.S. Navy, mostly doing simulations of flight control systems for aircraft, using rudimentary digital computers.
By the early 1960s, SAGE networks spanned the U.S. and Canada, connecting twenty-three main computer sites. While it was a technological marvel it was also a gigantic fraud as an air defense system in that it had fatal flaws at multiple levels. However its flaws were kept secret from the public, enabling the contractors to make enormous profits over a twenty-five year period, during which time SAGE evidently became the biggest military-industrial-political fraud of the 20th Century, though President Ronald Reagan's Strategic Defense Initiative (also known as “Star Wars”) was a close competitor for that honor. Unfortunately their modern descendants are still going strong today.
Two good things did come out of SAGE based on further development in the MIT community in the early 1960s, which made ARPAnet possible:
A practical way to do interactive computing, which came to be called timesharing
Packetized data communications using six levels of networking.
I observed those developments directly, having been spun off from the MIT Lincoln Laboratory into the nonprofit MITRE Corp. In 1965 I moved to Stanford University, which I represented on the start-up committee for ARPAnet, the first general purpose computer network. Our committee helped formulate its performance specifications during 1967-68, reviewed alternative proposals and made recommendations. Still later, in 1972, I helped Vint Cerf set up his Stanford project that developed the Internet Protocols.
Given that ARPAnet grew up in a friendly academic environment and gave birth to the Internet, I view it as Mother of the Internet. Since ARPAnet was spawned by SAGE, a military system, I view SAGE as the grandfather and the Cape Cod System, its prototype, as the great grandmother, but that is enough genealogy for now. I leave it as an exercise for the reader to figure out the gender of the Internet. It looks rather bi to me.
Who's the daddy? Speaking of sex, over the years there have been many articles identifying various people as a “Father of the Internet.” Some offer plausible arguments while others seriously distort the facts and quite a few are pure fiction. In any case, claiming that this technology has a human male parent is an example of sexist egotism. Why no mothers? And why must the parent be human?
Claiming that various people were the fathers of advanced technologies implies that they had sex with earlier feminine technologies. While robot prostitutes are now becoming available they will not be able to produce viable offspring in the foreseeable future, so I suggest that claims of technological fatherhood by humans be dropped from public discussion.
SAIL. Beginning in 1965 I worked with John McCarthy and others to create a graduate-level computer research facility that I ended up calling the Stanford Artificial Intelligence Laboratory (SAIL). Its participants subsequently started dozens of successful companies including Sun Microsystems, Cisco Systems (actually more of a ripoff than a spin-off), RSA, and D.E. Shaw & Co. Others made substantial contributions, directly or indirectly, to the creation of Microsoft, Xerox PARC, Atari, Apple, and Amazon.com and also made major contributions to the success of Yamaha and (for awhile) of Digital Equipment Corporation.
Another measure of SAIL's success is that sixteen of its graduate students and other affiliates subsequently received ACM Turing Awards, which are widely viewed as Nobel Prizes for computer science. For a more detailed listing of my brags as well as some embarrassments see Brags and belches.
Overview of Networking. Looking at the modern Internet, with its rich collection of services that evolve daily and are accessible almost worldwide, you may be tempted to suppose it is the result of a careful long range plan. In fact the planning was incremental and this fortuitous outcome was the happy result of a number of separate inventions and a fair amount of luck.
Some mistakes were made along the way but fortunately this remarkable series of developments somehow kept going. I don’t think that anyone saw much beyond the next step as it developed, though some claim otherwise now. In any case, we now have a global public utility that keeps diversifying. There certainly will be further important developments from networking provided that governments and corporate interests that are striving to get control of it do not succeed.
Given that the network has enabled many new industries to become very successful, it is tempting to suppose that it was created by greed-driven entrepreneurs. However, as mentioned above, all of the key inventions came from altruistic inventors and developers who sought no patents on their creations and helped others adopt them without seeking compensation. In fact, had they sought to exploit or control the use of their work it is likely that computer networking would not have come nearly as far and as fast as it has in the last sixty-some years.
Changing names. As mentioned above, one source of confusion about technological histories is changes in terminology. This article generally uses modern terms, which are not consistent with some older documents. Even if you have access to older documents you may find it difficult to learn about earlier versions of a given technology unless either you or your search engine knows about such changes in that field. For example, modems were used for about twenty years before they began being called that.
In the 1950s SAGE pioneered the use of modems for computer networking and used specialized forms of what we now call “packet switching,” a packet being a linear array of binary digits. The term “packet” was introduced in 1968 by Dr. Donald Davies and was later picked up by others, who started talking about “packet switching” in the early 1970s. Subsequently Dr. Davies and his supporters claimed that he invented packet switching, which was a fabrication. Dr. Davies proposed to build such a system but got only enough funding to build a simple Star network (several computer links connected through a single hub), which was far less sophisticated than what SAGE used earlier. Dr. Paul Baran also sought funding to test his scheme, aimed at a military communications system that could survive an attack on many of its links, but was unable to obtain funding so we will never know how it would have turned out. For some reason a number of people (not including Baran, who had become a friend of mine) then claimed that he had designed ARPAnet.
Another set of terminology shifts began with what was called “real time computing” in SAGE, which needed to process a lot of radar data in real time, though it was actually not able to do that for aircraft using radar jamming such as any bomber built after World War II. The computer programs that did automatic tracking of cooperative aircraft also handled human inputs and showed the results within a couple of seconds, which was very fast compared with the batch processing computer systems then in general use.
Based on a 1999 proposal by MIT Prof. John McCarthy, several groups in the MIT community and others then developed what came to be called “general purpose timesharing” which allowed a single computer to provide interactive services to a number of people concurrently. That development was an essential predecessor to computer networking and came into fairly widespread use in the mid-1960s. When ARPAnet was created later it was composed exclusively of timesharing systems of various types. Later still, as the World Wide Web came into use in the 1990s, all web services were provided by timesharing systems, which came to be called “servers”. More recently some marketeers introduced the term “cloud computing” and have pretended that it is a new invention even though it continues to be provided by plain old servers.
Chronology. Below are some key inventions that made computer networks work.
Electronic digital computers
These were mostly byproducts of World War II technological developments. You might think that modems were first created for computer networking but that was not their initial application.
Point-and-click graphical user interface using light guns
Magnetic core memory
Packet computer networks
Packet radio networks
All of these but the last were initially developed as parts of SAGE (Semi-Automatic Ground Environment), an air defense system that was initiated by the Massachusetts Institute of Technology (MIT) with support from the U.S. Air Force.
General purpose computer timesharing
Mouse supersedes lite pen
Sparsely connected general purpose network designed (ARPAnet)
Development of all but the mouse was also initiated at MIT, drawing on the SAGE networking technologies. ARPAnet became operational around 1971 though some like to think it was 1969. International communications organizations were laughing at us inasmuch as they thought we were pursuing a bad idea.
Warning unheeded. U.S. President Dwight Eisenhower gave a speech as he left office in 1961 warning about the growing military-industrial establishment. That was the period in which SAGE was being deployed and he almost certainly had SAGE in mind when he gave that speech. However he was ignored and SAGE went on to become the largest military-industrial-political fraud of the 20th Century as discussed below. Its “success” then spawned a horde of “Command and Control” systems that have been bilking American taxpayers ever since and are still going strong today.
U.S. President Dwight Eisenhower, Warning of the military industrial complex, 1961 Jan. 17, 2 ½ minute video.
Local computer networks
Internet Protocols (TCP/IP)
Creation of the Internet through widespread adoption of the Internet Protocols
Personal computers became popular
Laptops began appearing
The latter two were inevitable results of the many advances in manufacturing technologies predicted by Moore's Law.
World Wide Web
Commercial services allowed on the Internet
Web-based search engines
At this point networking was spreading into many aspects of business and government around the world while political censors began controlling the net in certain parts of the world.
Search engines and social networks take off
Spreading high performance networks will enable Internet TV to replace broadcast, cable and satellite services.
Net-coordinated autonomous road vehicles will begin providing robot taxi and delivery services.
Myths. All cultures seem to develop myths about their origins and the Internet is no exception. Each day new stories appear, some about recent developments and others about the more distant past. Upon closer examination some of these stories turn out to be pretty accurate while others are pure fiction, but most are a mixture of the two. I used to be surprised at how many inaccurate creation myths were appearing and attempted to refute some of them until I figured out that would be a full time job and would be rather unproductive.
Let me first review a commonly repeated Internet myth that came out of the U.S. Presidential campaign in 2000 between Al Gore and George W. Bush. As you may recall, the outcome of that election was resolved by the U.S. Supreme Court in a rather mythical way.
Gory details. Many people like to think that Al Gore claimed to have invented the Internet. Statements to that effect are typically delivered with a smirk and a giggle, implying that the speaker thinks he has said something clever. In fact, Gore claimed only to have taken the initiative in “creating the Internet,” which he actually did by rounding up financial support for it in the U.S. Senate. Had he not done that in a timely manner it might not exist today. The “invention” myth was created by his Republican opponents for political purposes, just as seems to happen in all elections. See the Internet of Lies at snopes.com
Fantasy worlds. It is clear that telling tall tales is a natural human tendency and that some people don’t even know when they are lying. The reason is that even though we all must deal with reality to some extent in order to survive, we also spend much of our lives in fantasy worlds shaped by our heritage including genetics and culture, family values, community viewpoints, and our choices of religion or political party, if any. These things are augmented by a natural tendency by many people to brag.
This article attempts to present a fairly complete and accurate history of computer networking based on my somewhat unusual perspective but necessarily subject to my own fantasies and biases, though I like to think that I’m well grounded. I readily admit that my memory is imperfect, but I also happen to be a pack-rat, having kept filing cabinets and boxes full of documents from earlier times. Only occasionally am I surprised when I reread that material. However my observations below are admittedly a bit uneven because some inventions come with more interesting stories than others.
Keith Windschuttle says that each historian actually makes his own history. See The Killing of History, Simon & Schuster, 1996.
Now that you have been warned, I offer below my opinionated take on what happened. Please note that I do not claim infallibility; some of my past recollections have turned out to be false based on evidence discovered later. Catch me if you can!
Bigotry. As computers and robots continue to play increasingly important roles in our lives, some view them as inferiors that must be kept in their place and such people make statements about robots resembling those made by racial bigots. Computers have shown superiority over humans in performing complicated arithmetic calculations since the 1940s, but that is seldom mentioned in polite company. The Internet now provides many services around the clock in such areas as communicating with others, purchasing, banking, and requests for services that used to be accessible only by making a trip to a specific site during regular business hours. Robots are increasingly playing parts in manufacturing, farming and animal husbandry and comments on their alleged inferiority are on the increase.
Robots have demonstrated for years their clear superiority over humans in doing planetary exploration but there is a strong movement to bypass them by sending people to Mars, which has great political appeal but makes no practical sense. My expectation is that robots will continue to dominate the exploration and development of the Solar System and the Universe. For interstellar travel, robots are the only plausible choice. Instead of viewing them as inferior beings that need to be put down I see them as our natural descendants who will continue to develop a wider range of capabilities than we have.
Our society is about to take another big step forward by introducing robotic delivery vehicles, taxis and personal chauffeurs. I expect that they will be safer drivers than the average human, in part because even though they may consume alcohol or other substances as energy sources they will not use them to enlarge their fantasies, as we do. While banning humans from texting while driving is justified we should allow robots to do it because they can manage it by doing parallel processing without being distracted. In any case, we likely will need to cope with increasing bigotry toward robots, which is a result of traits we inherited from our hunter-gatherer ancestors.
As intelligent machines diversify their abilities they likely will come to regard humans as their assistants, though that will not be for quite awhile. Let us hope that they do not indulge in bigotry to the extent that we do.
SAGE. The attractiveness of the SAGE air defense system was noticed by other military commanders, who decided that they too needed something like it. One of the earliest was General Curtis LeMay, who ordered one for his Strategic Air Command (SAC). It too turned out to be useless but the name used for that system profoundly affected further developments. It was originally called the “SAC Control System” but when that was written out fully as “Strategic Air Command Control System” the chance juxtaposition of “Command Control” somehow caught military imaginations and set off a major growth industry, as discussed in my ACM Communications millennium article. See L. Earnest, E2A is worse than Y2K, CACM, July 2000.
Divorce. In the late 1950s, MIT administrators found that negotiations with the Air Force were becoming increasingly difficult and they eventually decided to step away from SAGE in 1958 by spinning off a major part of the Lincoln Lab staff into a new nonprofit corporation called MITRE. I would like to think that this was the result of MIT’s realization that they had been participating in a gigantic fraud but I'm not certain of that. In any case, I and many others went along with that move.
Meanwhile, things changed in the new world of MITRE. Overall goals eventually were no longer set by senior staff members but by Air Force officers who had almost no understanding of the technologies they were supposedly overseeing. Of necessity we adopted the Air Force way of doing things, in which the main objective was to spend all the funds for each project by the end of the fiscal year so as to ensure that a larger amount would then become available regardless of what, if anything, had been accomplished. Thus the development costs of SAGE and other projects soared further. Nevertheless SAGE somehow came to be regarded as the model of what the next generation of military control systems should be. Never mind that it was useless--it looked good!
The main problem with all of those systems was that the objective was to computerize things regardless of whether that improved overall system performance. There were almost never any quantitative performance measures specified nor analysis made of price/performance measures of alternatives. They just hired a bunch of nerds and turned them loose!
Getting funding for these projects was enabled by public fears derived from the ongoing Cold War. In the 1950s there was a lot of public concern about the possibility of a nuclear attack by the Soviet Union and this was encouraged by politicians and government officials in order to justify expansion of government powers and increases in Defense Department budgets. In urban areas, sub-basements of some buildings were designated as bomb shelters and signage was erected to show people how to get there. Civil Defense groups were organized in residential areas and some people built underground bomb shelters at their homes and stocked them with food and water so that they and their family might survive a nearby blast.
Those feelings of insecurity were further expanded when the Soviet Union launched the Sputnik satellite in 1957, which suggested that they were ahead of us in rocketry. That led directly to the creation of the Advance Research Projects Agency (ARPA, which later became DARPA, then ARPA again then DARPA again) and further increases in military budgets.
The Air Force promptly endorsed this growth industry by starting its Air Force Command and Control Development Division (AFCCDD) which was generally called “C-squared D-squared” to emphasize its technological roots. Their development programs were organized into a series of “L-systems” as follows:
412L Air Weapons Control System,
416L Continental air defense (SAGE),
431L Traffic control and landing,
433L Weather observation,
438L Intelligence handling,
465L Strategic Air Command and Control System (SACCS),
474L Ballistic missile early warning (BMEWS),
480L Air communications,
496L Satellite surveillance.
The 474L BMEWS and 496L satellite surveillance evidently worked but many of the others did not. In practice the tasks undertaken usually revolved around computerizing existing processes without any stated performance objectives, which often resulted in a net decline in performance or a total malfunction as happened in the case of SAGE.
That clearly was a growth industry and the parent organization, Air Force Command and Control Development Division (AFCCDD, generally called “C-squared D-squared”), went on to change its name every few years because bureaucrats like to leave their mark by reorganizing whenever they can. Other Defense Department boondoggles also blossomed in this period and even some municipal police and fire departments decided to build their own “Command and Control” systems that generally turned out to be expensive and useless. President Ronald Reagan’s absurd Strategic Defense Initiative (aka “Star Wars”) fit nicely into this scheme and wasted more billions of dollars.
Since the 1960s all U.S. Federal administrations have invested in the attempted development of anti-ballistic missile systems, which was a reasonable thing to do, but President George W. Bush started deploying them even though, like SAGE, they didn’t work. Like SAGE, anti-ballistic missile systems work only against simple missiles that do not deploy decoy warheads unlike real intercontinental ballistic missiles.
The Obama administration has continued this absurdity, which has given the Russians an excuse to complain even though they too know those systems don’t work. Politics is more costly than national health care.
Diving underground. The last SAGE-related project I did was to help plan the North American Air Defense Command (NORAD) Combat Operations Center that was to be constructed deep inside Cheyenne Mountain near Colorado Springs, Colorado. NORAD was responsible for overseeing all continental defense systems for the U.S. and Canada. I visited their headquarters at Ent Air Force Base in Colorado Springs a few times to learn more about their operations, including their satellite tracking system called Spacetrack.
In the 1960s there was still a lot of public anxiety about the likelihood of nuclear war and some people talked bemusedly about the need to have both men and women inside Cheyenne Mountain in case a nuclear exchange wiped out everyone else in the world, in which case they could later emerge and repopulate the World.
In the early 1980s I began making frequent trips to that former Air Force Base site after it became the Headquarters of the U.S. Olympic Committee, a government regulated nonprofit agency that has come under the control of corrupt commercial interests that run it to suit themselves, but that is another story.
In 1962 MITRE finally began work on a “Back Up Interceptor Control” system (BUIC) that would provide a more dispersed and redundant control system with computer facilities located mainly at radar sites rather than at SAC bases. BUIC systems began to be deployed in 1965 even though the threat of a first-strike manned bomber attack had disappeared because of ICBM developments, for which no defense was available.
Suppressing an unacceptable answer. Sometime around 1960 I was assigned the task of analyzing and reporting on the utility of a new kind of manned interceptor that was tentatively identified by a number somewhere in the range of F-107 to F-110, but I no longer remember which one. This aircraft was to have a fairly long range and carried both a long range radar and multiple air-to-air missiles that would enable it to operate semi-autonomously in finding and attacking manned bombers. In other words it wouldn’t need SAGE. After examining plausible deployment schemes in which such interceptors would be placed at many airfields across the country, so that an attack on any given site would not take out many of them, and knowing that the SAGE system really didn’t work, I wrote a classified report advocating the use of this new style of air defense.
As soon as I my report was released, an Air Force Colonel wrote a refutation and demanded that anyone who received my report be given a copy of his. MITRE promptly complied. It turned out that the reason we had been asked to do this study in the first place was that since MITRE had been directly involved with the creation of SAGE it was assumed that we would oppose any system of a different kind. When we failed to live up to their expectations their counterattack was distributed without our getting a chance to refute it.
Oddly, a couple of years later some people in the Pentagon, who somehow learned that I had studied this aircraft but didn’t understand that I supported it, asked me to tell them how to show that it would be ineffective. I bemusedly pointed out that if this new kind of interceptor was deployed far out so that by the time they could get up to altitude they would be in a tail chase situation relative to the bombers then they would be much less effective than in a situation where they could move toward the bombers head-on. They apparently used that misleading argument and managed to shut down the program.
Who invented timesharing? The short answer to that question is “A lot of people”. The feasibility of interactive computing first came to light as an accidental spinoff of the SAGE air defense system. See L. Earnest, Who invented timesharing?
Who invented the modem?
Who invented packet switching?
It was Jack Harrington’s group around 1953, after they moved from AFCRC to MIT Lincoln Lab. As part of the experimental Cape Cod air defense project (the SAGE prototype) they had Burroughs Corp build the FST-2, a special purpose computer used to process radar data from radars for transmission to the Whirlwind computer at MIT. The same scheme was used in its offspring, SAGE. Overall the SAGE network interconnected hundreds of sites across North America beginning in 1959.
However the packet switching done by SAGE was special purpose and did not, for example, provide a way for someone to send a message to anyone else on the system. Telephones were used for that. Also you won’t find packets mentioned in the literature describing SAGE because that term was not introduced until over a decade later.
Note that SAGE used a combination of “star” networks each of which connected directly to a central computer. That was quite different from the sparsely connected ARPAnet and Internet.
Several graduate students were sharing the TX-2 computer with me when I was working on Curse, so we each got to see each other’s work and sometimes offered suggestions. Among my comrades were Ivan Sutherland, who was developing his Sketchpad drawing system, Larry Roberts working on perception of physical objects from photos, and Len Kleinrock doing simulations of sparsely connected networks, which proved that they could work without getting clogged. That later was used as a foundation of ARPAnet.
L. Kleinrock, Message delay in communication nets with storage, PhD dissertation, MIT, 1962 December.
I. Sutherland, Sketchpad, a man-machine graphical communication system, PhD dissertation, MIT, 1963.
L Roberts, Machine Perception of Three-Dimensional Solids, MIT Lincoln Laboratory Technical Report #315, 22 May 1963. http://packet.cc/files/mach-per-3D-solids.html
L. Kleinrock, Communication Nets: Stochastic Message Flow and Delay, McGraw-Hill (New York), 1964
As it turned out, all of us were later involved in the creation of ARPAnet.
Television for the President and Moon Men
In the Spring of 1962 I heard that the Defense Department was planning to build a secure television system connecting the White House and the Pentagon, presumably going both ways. I assumed that this system would be digital and that they would employ the common cryptographic scheme of using pseudo-random bit stream generators at each end to encrypt and decrypt the data.
It occurred to me that this would mesh nicely with a video encoding scheme given in Larry Robert’s 1961 Masters thesis, which substantially improved video quality by synchronously dithering the low order bit at both ends of a digital video transmission using synchronized pseudo-random number generators.
Lawrence G. Roberts, Picture Coding Using Pseudo-Random Noise, MIT, S.M. thesis, 1961
I wrote up this idea in a classified report that got the Pentagon’s attention and Larry and I were invited to visit them. We did so and were asked a lot of questions but when we asked whether they were likely to use this idea we got only evasive answers. My guess is that they did use it but we will likely never know for sure.
On our flight home from Washington Larry mentioned a problem that Ivan Sutherland was having in Sketchpad when dealing with line drawings that were over-constrained, such as a line that was supposed to be parallel to another one but also perpendicular to a third line that didn’t quite line up. I suggested using one of my favorite tools, namely least-squared error fits to the constraints. Happily that apparently worked.
As we learned later, Larry’s TV dithering scheme was used by NASA when they went into show business by sending television pictures of astronauts landing on the moon back to Earth.
Three kinds of intelligence
Some people like to observe that there are three kinds of intelligence—human, artificial, and military—and none of them has much to do with the other two. My cursive handwriting recognition project was in the field that was beginning to be called “artificial intelligence” and I subsequently spent most of my career in that field. John McCarthy had introduced that term to replace the less sexy name he used earlier, namely “automata studies.”
C. Shannon & J. McCarthy (ed.), Automata Studies, Princeton University Press, 1956.
I and some others prefer to call this field “machine intelligence” on the grounds that there is nothing artificial about it. I view the “artificial” term as a form of bigotry based on the premise that human intelligence is superior to all others. While it is true that machines are not yet very good at what we call common sense reasoning, machine intelligence has been superior to that of humans in some areas for a long time and will continue to become more diverse over time, though I expect that the so-called “singularity” when they will become dominant is still a couple of hundred years off, contrary to the expectations of some enthusiasts.
Note that human intelligence is but one branch of animal intelligence, yet we arrogantly assume that all forms of intelligence are inferior to ours. As you know, dogs are very much better than we at detecting the scents of many substances. Is that just because they have better sensors in their noses or is it because they have better scent analytical abilities than we? I bet it is principally the latter.
We now know that elephants use very low frequencies, lower than we can hear, to communicate with each other over long distances and it appears that dolphins and whales use complex acoustic communications. We do not yet understand their languages but assume that they are not saying much of importance.
Some people view military intelligence as being a mixture of spying and lying, but the hard part is resolving inconsistent and incomplete information so as to meet the demands of superiors. While working on SAGE I sometimes saw intelligence reports claiming that the Soviet Union was developing a computerized air defense system similar to SAGE. Those evidently were fabrications used to help sell SAGE to Congress. I knew that those reports were bogus because the Soviets didn’t have computer technology comparable to ours and ours clearly was inadequate.
Going to L
Late in 1960, given that the design of SAGE was more or less complete, I and my group were reassigned to provide systems engineering support for Air Force System 438L, the “L-system” dealing with military intelligence. It was headed by a Colonel M who I soon learned knew almost nothing about computers. His prime contractor was an IBM group that believed all information processing tasks should use batch processing computers with punched card inputs.
Col. M promptly took me on a tour of Air Force intelligence sites in Europe, during which our plane, borrowed from a General, nearly crashed twice. He then asked me to review an RFP (Request For Proposals) written by IBM for a prospective small computer system. Looking it over, it was clearly a description of the IBM 1410 computer.
I and my colleagues then compared those performance specifications with various machines then on the market and concluded that the Bendix G20 computer could do that job better at a lower cost, but when I gave that report to Col. M he was horrified. “How can I ask the IBM people to program another company’s computer?” he asked. I responded that any competent programmers could do it. That response apparently put the project in limbo.
Two years later, as I was about to move on to another project, I mentioned in passing that an IBM 1410 actually could do that job, which he evidently took as acquiescence. I learned later that he issued a purchase order for a 1410 an hour after I said that. So much for providing systems engineering support!
Keeping up to date
In the meantime I learned more about the main project of System 438L, which was to store and retrieve intelligence reports at SAC Headquarters near Omaha, Nebraska, using an IBM 7090 computer.
As time passed with no results from that project I asked the IBM manager what the problem was. He said they were having trouble developing the data retrieval software that would extract information from magnetic tapes. I said that didn’t sound too hard to me and asked about the data formats. When he gave them to me I took them to one of our able IBM 7090 programmers and asked if he could figure out how best to do it. He gave be a set of retrieval macros the next morning, which I passed along to the IBM manager.
Two weeks later I learned that my colleague was moving to Omaha, having been recruited by IBM. I thus learned to be more circumspect in my bragging. Unfortunately the next time I saw my friend he had become rather depressed.
Awhile later IBM declared that the new system for SAC was working, so I took a closer look and figured out that the whole thing was senseless. They were taking in intelligence reports that had been transmitted over cryptographic links and printed on paper, then put them on punched cards, which they then verified and error-checked to some extent. As a result, the online files were always about three days old, making them useless for following fast changing situations.
I also learned that in the earlier system the intelligence reports were read as they arrived and filed in appropriate folders while grease pencil summaries were maintained on Plexiglas displays for use by the Duty Officer, who could request the original reports as needed. Not surprisingly, the old system remained in use because it worked whereas the new computer system didn’t.
After learning that embarrassing fact, the General in charge decreed that each Duty Officer would make at least two inquiries of the computer system on each shift. We later checked the logs and found that the number of queries made over a period of a couple of months was precisely two times the number of shifts that had occurred. So much for advanced technology!
In the early 1960s computers were being used increasingly to compile databases and to search them but nobody was doing that with ordinary text messages other than in the ill-considered 438L system. In the account below I will use modern terminology such as “search engine” even though that term didn’t come into general use until the new millennium, as I recall. In view of the useless and expensive text processing system built by IBM, in March 1961 I initiated the development of what was evidently the first search engine.
L. Earnest, ROUT, the first search engine. http://www.stanford.edu/~learnest/rout.htm
1962, a year of transitions
In 1962 Lick was recruited by ARPA to start an Information Processing Techniques Office (IPTO) and he set up a conference in Hot Springs, Virginia, in November 1962 that was essentially a pep rally for timesharing. I and others enjoyed it in spite of the fact that it was conducted at a racist resort.
Lick had inherited several “legacy” command and control system development projects that he and his successors shut down over time while he focused on supporting things he thought would have long term payoffs such as interactive computing, enabled by timesharing, and artificial intelligence.
Early in 1963 Lick revealed his vision of networked interactive computer systems, which eventually turned into ARPAnet.
Also late in 1962 John McCarthy moved from MIT to Stanford and started a new AI Project there, which Lick funded beginning in 1963.
Out of the frying pan
In late 1962 I too made a switch when I was pulled off the useless System 438L and was assigned to work for a year at the Central Intelligence Agency (CIA) Headquarters on a “database integration project.” Rather than transplant my family I chose to commute there from Boston on a weekly basis.
I was aware that CIA liked to engage in what some call “security theater.” So does the more recent Department of Homeland Security, but I learned more than I wanted to know about this phenomenon while working there for awhile.
L. Earnest, CIA’s phantasies and lies, December 2012. http://www.stanford.edu/~learnest/cia.htm
As soon as I got into the database integration project I realized that it didn’t make sense. There were two very large collections of data that were quite different in both structure and content and there was no practical way to shove them together, though it appeared to me that it might be possible to create a query system that could translate into one or two searches of the existing files. After a month I wrote a memo pointing this out and suggesting that they consider instead starting a new database that incorporated both kinds of information but only after we had investigated and documented how the incoming information was selected or rejected, how it was encoded, etc.
However the head of our project said that was not an acceptable plan in that it did would not “integrate” the existing databases, a requirement that he viewed as coming to us on stone tablets. In the following months we continued to run around in circles, visiting various intelligence services in the Washington DC area including the National Security Agency (NSA), the old Army Map Service, the Defense Intelligence Agency, etc. Along the way we learned how they used stereoscopic aerial photos to construct 3D information of what was on the ground.
Some months later I was relieved to learn that I was being reassigned to a different project. On my last day at CIA Headquarters the head of our project shook my hand and said he had concluded I was right about what they should have done for the integration project, but I don’t know if they actually went that way. I did learn later that he had risen to a lofty position in the Agency.
In May 1963 I was assigned to work on the Air Force’s Project Forecast, which was supposed to assess current technology trends and point the way to future developments. This work was to go on for several months at the Aerospace Corporation in Los Angeles, so inasmuch as my wife and I both had families in San Diego we pulled our three kids out of school and drove there, then I began commuting on a weekly basis to Los Angeles.
There were a number of projects going on there and I recall that the computer memory group predicted that magnetic core memories would soon be replaced by magnetic thin film memories, which was not their only misassessment.
I decided to focus on what I believed was an ongoing planning error, namely the top-down design of military command and control systems that typically began with designing what they would like the General to see, then defining progressively lower levels of support and designing the whole system as a unit. When approached in that way these projects always took years to complete, cost enormous amounts, and didn’t work when they were “finished.”
I advocated a “bottom up” approach starting at the lowest level of the organization and getting prototype systems going there as quickly as possible, then moving progressively up one level at a time and redesigning the whole system each time, which I claimed would lead as quickly as possible to a system that actually worked.
However I didn’t have access to a typewriter so I wrote this proposal up on thirty-some pages of block printing and gave it to the Colonel heading our group. After reading it he remarked that “There is some interesting stuff here but you can’t tell them they have been screwing up.” He then produced a Bowdlerized version of my report that carefully avoided inferences of erroneous design in the past. Thus some more months of my time were wasted.
When I returned to the Boston area I was invited to join a new office in Virginia across the river from Washington DC that was charged with designing a National Military Command System, based in the Pentagon, that was to support the Join Chiefs of Staff. Even though they were again planning to do this in what I viewed as the upside-down approach I agreed to go. My wife and I allocated three days in September 1963 for buying a house there and managed to do that, then discovered that since there were few new residential tracts in that area a fellow who worked for me was directly across the street, a nemesis of mine was two houses away and my boss lived three houses away in the opposite direction. While shooting off rockets the next 4th of July I managed to put three of them on the roofs of fellow employees.
I started off spending some time in a two story briefing facility in a subbasement of the Pentagon that was about the size of a small basketball stadium. It had a large darkened-glass enclosed balcony where the Joint Chiefs, their senior staff, and upper level Defense Department officials received daily briefings. It was acoustically isolated so that no one below could hear their discussions but they had button-operated microphones so that their questions would boom out on the public address system below.
The briefings were given by a duty officer standing at a podium on the floor below, looking up at the gods in Valhalla, and answering their questions. There were three large screens on the wall opposite their balcony where typically the left screen would show some kind of organization chart, the right screen would show slides taken by spies in various places and the center screen would show videos sometimes taken by spies but more often derived from commercial news broadcasts. Peons like me were allowed only on the lower level, of course.
I learned shortly that I needed to reapply for a special security clearance that I had earlier while working at CIA, so I filled out the application form and, having been sensitized by the blatantly racist policies of the State of Virginia, gave a slightly provocative answer to one question. That led to my accidental discovery that doing so greatly accelerated the processing of my application so that I got the clearance much faster than my colleagues.
L. Earnest, Can computers cope with human races? CACM, Feb. 1989.
One of the first things I decided to investigate in the Pentagon culture was military geography. Whereas most literature shows geopolitical boundaries on maps, military commands have a different view of the world in which a given command often spanned multiple countries and great expanses of ocean. One day I sent a staff member to a Pentagon office to collect information on this but that afternoon I received an angry call from a Colonel who told me never to send him there again. His offense? He had a beard, which they did not allow their personnel to wear. I wrote up a humorous report of this incident but my boss decided to avoid further confrontations over this matter.
Things continued in that vein and I could see that the new top level command and control system would likely be as expensive and useless as its lower level predecessors, so I became increasingly frustrated.
Meanwhile Lick had been continuing his good work in IPTO, located in the Pentagon, and after Ivan Sutherland completed his PhD, Lick recruited him to come there and take over in September 1964 as Lick returned to MIT to take over Project MAC. Sutherland managed to concurrently fulfill his military obligation by serving as an Army officer while running IPTO.
Following up on Lick’s networking idea, in 1965 Sutherland funded a contract with Larry Roberts at Lincoln Lab to construct an experimental link between two recently completed timesharing systems, namely the TX-2 at Lincoln Lab and the Q-32 at SDC in Santa Monica, California. While that was the first link between timesharing systems, in my opinion it did not demonstrate a significant advance in networking. Sutherland also offered to support a network connecting three computers at UCLA, where Leonard Kleinrock had gone, but that fell apart as a result of academic politics.
Along the way Ivan and I and our wives socialized a bit. Awhile later he remarked that he had to review many funding requests but had no help and asked if I would be willing to give my assessments of a few proposals. I agreed and after looking over three or four papers that he gave me I visited his office in the Pentagon and recommended that he fund all but one, which was a request for continuation of an existing project. I suggested that he terminate that one.
Ivan promptly said he agreed with me and then suggested that I join him in the IPTO. I inferred that I had actually just been given a final exam. That offer had a profound effect on my life but only indirectly in that my twelve years of experience in the Defense establishment caused my ethics to boil over and I told him that my goal was to get as far from the Pentagon as possible.
After a thoughtful pause, Ivan said he might have something for me inasmuch as he had recently awarded several million dollars to Prof. John McCarthy at Stanford University and then learned that John had little interest in administration of any kind. He suggested that I go talk to the Stanford folks and he apparently put in a good word for me there. He later recruited Bob Taylor to join him at IPTO.
On July 11, 1965, I loaded my family in the car, did some sightseeing on our way to Stanford and was interviewed by several people. I then was offered the position of Executive Officer of the small project led by McCarthy that we eventually developed into the Stanford Artificial Intelligence Laboratory (SAIL), a research organization primarily for PhD candidates.
A new beginning
My escape to Stanford turned out to be the best career move of my life in that I landed in a place where there were a bunch of very bright and highly motivated people who created lots of interesting new technologies while earning advanced degrees, then generally either became entrepreneurs or went on to academic careers or both. Several of their new businesses became well known.
Based on earlier interviews I expected an opportunity to work on developing an automated biological laboratory to be sent to Mars but when I got there I learned that it had not been funded. In fact for some reason NASA didn’t get around to doing that until just now, some 47 years later, which in my opinion was a major blunder on their part. Meanwhile, under the influence of politicians, NASA spent billions of dollars on astronaut programs that produced little of scientific value compared with the accomplishments of robotic space explorers that cost far less.
I found that our project had some funding from the National Science Foundation (NSF) but was mainly funded by ARPA. Thus I was still feeding from the same military-industrial trough but in a more honest and productive way.
Upon arrival at Stanford I allocated three days to locate and purchase a new home for my family and managed to do that. However finding a suitable home for the project I was working on proved more difficult. For one thing, a large computer system and a number of peripherals were on order but there was no place to put them. I proposed bringing in a portable building to be placed near our offices at Polya Hall but the Stanford Administration said they would not allow the use of portable structures on aesthetic grounds. I then sent them a photo of the local aesthetic competition, taken from the back door of Polya Hall, which showed the Applied Physics Department’s junk yard full of rusting steel and coils of wire right next to us. However the Administration didn’t bend.
They proposed that we move into a partly constructed building they had recently purchased in the foothills about six miles off campus. I later figured out what was actually going on—Stanford did not have the occupancy rate in this building high enough for it to qualify for overhead charges on government contracts but we would put them over that threshold. I also observed that a short time later the administration purchased hundreds of portable buildings which they used for student housing, making it clear that this alleged prohibition was actually flexible when it suited them.
Nevertheless I accepted their offer because it provided an accessible site with plenty of parking adjacent to the building in a lovely setting with direct views of San Francisco up the peninsula, the Bay, and surrounding mountains including Mt. Tamalpais, Mt. Diablo and Mt. Hamilton. Just over the hill, also on Stanford land, was Felt Lake, which turned out to be a good place to go fishing or skinny dipping.
D.C. Power Laboratory Building, the semi-circular structure in the center right of this 1968 photo, housed SAIL from 1966 to 1979. An arm of Felt Lake can be seen on the lower right.
Given that the building was a hollow shell I asked who would design the interior. “You” was the answer. Though I had no experience in that area I did it, putting the computer room with false flooring in what was originally to have been a large cafeteria. I also put a coffee and food room in the center with a lounge next door to encourage discussions resulting from chance encounters.
Bucking tradition, I provided little secretarial space with the expectation that everyone would do their own typing. However under hindsight I followed a foolishly sexist tradition by making the secretarial spaces accessible to the public without lockable doors, which meant that the only way they could secure personal items was to lock them in desk drawers. We ended up with one secretary for John McCarthy, who needed one, and two more for the hundred and some other people working there, including me.
Given that organizations often change names, especially when a new bureaucrat gets control and wishes to confirm his or her power, there is often confusion about which name to use. In general I will use the latest names but note that the U.S. Defense Department organization that provided the principal financial support for SAIL was originally called the Advanced Research Projects Agency (ARPA) up through the time when ARPAnet was created, then “Defense” was added in March 1972, making it "DARPA", then renamed "ARPA" again in February 1993, and then renamed "DARPA" again in March 1996.
I note that Brits insist on replacing “Defense” with the homonym “Defence”, which has always puzzled me. Why do they think that security can be improved by removing fences?
After John McCarthy introduced the term “Artificial Intelligence” in a 1956 conference at Dartmouth College, he and Marvin Minsky subsequently put together a project at MIT they called the “AI Project.” When McCarthy moved to Stanford University in 1962 he brought that name with him. Having become executive officer (i.e. principal bureaucrat) of this group in late 1965, I eventually concluded that the group of projects that had joined us in sharing the computer facility should be called a laboratory and, with McCarthy’s blessing, in 1971 began calling it the Stanford Artificial Intelligence Laboratory (SAIL). The name “SAIL” is now commonly applied to both the original AI Project at Stanford and subsequent developments through 1980 when it was dismantled. A later group of Stanford projects assembled by Prof. Sebastian Thrun began using the name SAIL again in 2003.
Rodney Brooks, who did part of his graduate studies at SAIL, apparently liked that name and, after taking a faculty position at MIT, he merged the AI Project there with an existing computer facility and called the new organization the Computer Science and Artificial Intelligence Lab (CSAIL).
Computer system had a shaky start
Upon arrival I was pleased to learn that a DEC PDP-6 timesharing computer had been ordered with a line printer having the same expanded character set as my old friend, the TX-2 computer at MIT Lincoln Lab, including a number of mathematical symbols and Greek letters. However a number of administrative problems showed up, including poor choices in a number of peripheral devices that had been ordered before I got there.
Getting the computer system to pass its acceptance tests turned out to be a challenge and things got worse when the SAGE System apparently tried to get even with me for badmouthing it.
L. Earnest, A hummingbird with range, Jan. 1, 2009. www.stanford.edu/~learnest/umunhum.pdf
Also DEC informed us that the price of the computer was actually much higher than the total amount shown on their offer and the resulting contract. Specifically a number of devices, such as magnetic tape drives, had been listed at one-tenth of their book value and the sum of the prices listed was substantially higher than the amount shown on their bid. In other words, none of the three parties to this transaction, namely DEC, the AI Project and the Stanford Purchasing Department, had checked the arithmetic. The Purchasing Department appropriately agreed to correct the arithmetic but refused to pay higher prices for individual items such as the magnetic tape drives.
In the review process I noticed that thousands of dollars had been spent to add a special instruction in the PDP-6 that made LISP programs run slightly faster but also made software written for that computer incompatible with others of the same type, which looked to me like a bad idea.
Image dissector showed a conflict-of-interest
I saw that an image dissector camera for visual image processing had been ordered from Information International Incorporation (III) at a cost of around $30,000 and came to be called the III Eye. That struck me as odd since such cameras, which were the earliest electronic television cameras, had low performance compared to more modern cameras such as the vidicon. I was subsequently surprised to learn that John McCarthy had arranged to purchase this device while ignoring his conflict-of-interest, namely the fact that he was a big shareholder in III and a member of its board of directors. I subsequently learned that he was not the only faculty member who ignored such matters.
Still later I learned that McCarthy had given III an additional contract for thousands of dollars to examine the feasibility of building an image dissector system with an image intensifier front end, which their resulting report recommended. I was suspicious of that conclusion, so in my spare time I did an analysis of alternative camera systems to determine their relative speeds and signal/noise ratios and published a report showing that ordinary television cameras such as the vidicon, which was much cheaper, substantially outperformed the III Eye and that the recommended addition of an image intensifier would not improve its performance. Thereafter we used ordinary television cameras for image input and eventually gave away the III Eye.
L. Earnest, Choosing an eye for a computer, Stanford AI Memo AIM-51, April 1967
The PDP-6 computer system was delivered with a bunch of Model 33 Teletypes to provide timesharing services and a new high performance display system with six consoles had been ordered from a different manufacturer. However they reneged on that contract.
I then wrote a new Request for Proposals (RFP), sought bids and got just two that were plausible. Curiously one was the company that had just reneged and the other was III. For obvious reasons I chose III even though their price was somewhat higher. I was then threatened with a law suit by the losers but when I told them to get lost they disappeared.
However when the III displays were delivered they did not meet the specification for the amount of text that could be displayed. When I asked them to meet the specification they cranked up the speed of the character generator but that made the text unreadable. I continued to press for better performance with the expectation that if they failed to meet the specifications there should at least be a price adjustment. However at that point McCarthy again ignored his conflict of interest and removed me from overseeing this project, assigned it to Arthur Samuel and told him to accept the defective system, which he did. That seriously annoyed me but I chose to do nothing about it.
On order was an unusual Librascope disk system using disks four feet in diameter that had a capacity of just 22 megabytes and cost about $330,000 plus special installation costs because of its reported susceptibility to vibration. We dealt with that by not placing it on the false floor but excavating into the building structure to support it directly on the building’s steel I-beams. Nevertheless it out to be extremely temperature sensitive in that a one degree change in room temperature would make it start forgetting, so it could not be used for long term data storage. It also experienced massive magnetic head crashes that scarred the disk surfaces.
We were able to use it for awhile as a swapping store by making two copies of all stored data, check summing them periodically and quickly making another copy whenever one of the copies went bad.
[Much more to come]