DUDLEY BUCK STARTED SHUTTLING AROUND THE COUNTRY FROM one classified project to the next on behalf of the NSA. Part of his role was to straddle the various top-secret computer projects commissioned by the agency to make sure that every organization involved was aware of the latest techniques. Some of those institutions were universities or government labs like the one at Corona. Others were commercial firms like Bell Telephone Laboratories, Engineering Research Associates, and IBM.
Now that he had been brought back into the military fold, Buck’s orders came directly from Solomon Kullback, another ex-Bletchley Park codebreaker, who outranked Campaigne and Eachus as the NSA’s chief of research. Kullback was something akin to Q in the James Bond films and rose to the rank of colonel before he retired from spycraft in 1962.
Kullback had been one of the US Army’s first dedicated codebreakers—the third employee hired into the army’s Signals Intelligence Service, as it was then called. Cracking codes remained a key concern for him, as it had been since he had broken the Japanese military ciphers in the 1940s.
Although computers had shown potential for lots of other uses, it was in coding and decoding secret messages that computers had established a proven track record of performance. Through Kullback, Buck became involved in developing an encryption system that would evolve into a device called KW-26—soon the standard coding machine for the US government and later for most NATO countries.
The other long-standing perception of where computers could be of greatest use to military society was in translating languages. Once a codebreaking computer had cracked a code, the messages still had to be translated from the original language into English before its importance could be analyzed; that still required human intervention. If thousands of pages of decoded messages had to be sorted through, from listening posts all around the world, it could take ludicrous amounts of highly skilled manpower.
This was a problem that had been recognized by America, Britain, Germany, and the Soviet Union during World War II; all of the military powers attempted to build translation machines with varying degrees of success. Even twenty-first-century artificial intelligence systems struggle to master the nuances of natural language translation, yet it was viewed as a logistical challenge at the time.
Thanks to the Korean War, American intelligence services were now having to deal with more languages than they had before. This made the demands of breaking Nazi messages a few years earlier seem easy by comparison.
The North Koreans were backed by the Chinese, while the Soviet Union remained the perennial threat; this created a demand for knowledge of three languages. The sheer volume of spying going on in and around Berlin, meanwhile, ensured that there was still demand for German speakers and translators.
Postwar tensions in the Middle East were also starting to bubble up—events that would later lead to the Suez Crisis and the Six-Day War. Languages were in high demand, so the NSA was under pressure to find a way to speed up the translation process. It was not humanly possible to keep up with the flow of information being generated.
A conference was organized by Yehoshua Bar-Hillel, a prominent philosopher and linguist at MIT who was at the vanguard of a group who believed that any language could be translated by machine if given enough data. The International Conference on Machine Translation convened at the MIT campus, where the all-singing, all-dancing Whirlwind computer was already in operation. Attended by eighteen of America’s top linguists, mathematicians, and computer scientists, the conference ran from June 17 to June 20, 1952.
Bar-Hillel would later team up with a young philosopher of soaring importance, Noam Chomsky. “Professor Bill Locke suggested we use computers to do automatic translation, so we hired Noam Chomsky and Yehoshua Bar-Hillel to do it,” Jerome Wiesner, the former dean of MIT, recalled in an interview for the university about the demolition of the building where the two men worked. “It didn’t take us long to realize that we didn’t know much about language. So we went from automatic translation to fundamental studies about the nature of language.”
Buck was sent to Bar-Hillel’s conference as an observer for the NSA. MIT’s delegation was led by Wiesner, who went on to become science adviser to President John F. Kennedy and later acquired a degree of infamy during the Watergate scandal: leaked papers revealed he had been one of twenty academics denounced as hostile to the administration of President Richard Nixon.
Buck was one of only two attendees who had yet to achieve a grand title, the other being James Perry, a research associate at MIT. Everyone else in attendance was director of an Ivy League university department or ran a government laboratory. As was customary at such events, there was also one representative from IBM and one from the RAND Corporation.
The only foreigner was Andrew Donald Booth, a British computer pioneer then working at Birkbeck College in London. Booth—originally an explosives expert who had invented the magnetic drum computer memory—would have been known to many of the individuals due to a postwar spell he spent at Princeton University working with American mathematician John Von Neumann on his early computers.
Some of the sessions were open to the public, and the assembled experts took questions from the floor. Buck wrote an extensive report for the NSA about what was discussed in terms of the agency’s particular areas of interest.
In both the public and private sessions there were some attempts to tackle the eternal philosophical questions about language and the common syntax and structural elements shared among different groups of languages. They also dabbled in the definitions of machine intelligence, as mapped out by Alan Turing only a few years earlier in his work at Bletchley Park; the famous Turing Test of divining whether a machine or a human being is answering questions from the other side of a screen.
Yet most of it was a fairly pragmatic attempt to scope the problem of deconstructing languages to make words and sentences readable by machine. How would machines cope with scientific jargon? What about idioms and odd turns of phrase? Stuart Dodd from the University of Washington in Seattle suggested the solution may lie in standardizing the English language.
“In English, at least, regularizing leads only to a certain quaintness of expression somewhat similar to the sentence structure employed by the Quakers,” wrote A. C. Reynolds of IBM in his account of the session. “No attempts have been made as yet to regularize languages other than English, but at least for the Romance languages it seems on first view that such regularization can be accomplished.”
Bar-Hillel himself explained that the United Nations was processing about 175,000 pages of translated documents every year. To keep up with the flow of Chinese and Soviet messages being generated solely as a result of the Korean War, the NSA would need to process more than a million words per day. “I used the phrase ‘if a human being can do it, a suitably programmed computer can do it too’ more than a dozen times during the conference,” Bar-Hillel wrote in a 1964 paper recalling the event.
Bar-Hillel spent the next twenty years researching computers. By the time of his death in 1975, he had concluded that it was more or less impossible to build a machine that could translate accurately and consistently using proper syntax and grammar. He handed over his work to Chomsky, whose fame has subsequently outstripped that of his professor.
After Bar-Hillel’s conference, a whole new area of computing research was spawned, with programs set up across the United States to investigate this new form of artificial intelligence. As the drive for machine translation progressed, it fell to Buck to keep Washington up to speed on developments—and contribute his own ideas.
There was an assumption that building these translation machines would also mean finding a new way of building a computer. The invention of the electronic transistor in 1947 had led to a new boom in computing, replacing vacuum tubes as the driving force of computers. The next stage of development, it was assumed, would involve some yet-to-be-discovered device.
“The transistor was smaller, used less power,” explains David Brock of the Computer History Museum. “The NSA and the aerospace people ate that up. They wanted transistorized computers. Even the people who made the transistor were thinking, ‘Okay, what’s next, what’s the next device?’ Clearly the transistor was better than the vacuum tube. Obviously, there’s going to be something that’s better than the transistor. So people were looking at all kinds of things.”
Infused with the knowledge of the MIT linguistics department, the electrical engineers set to work in attempting to build translation machines. On November 28, 1952, Buck wrote to the US Department of Defense claiming that a team at MIT was making advances in its efforts to translate Russian by computer: “Work is progressing here at MIT on the short-range word-for-word translation program under the direction of Professor J. W. Perry. The chosen language is Russian. Enclosed are sample translations and a discussion of the machine simulation processes. The work is primarily aimed at automatic indexing and abstracting.”
Washington took the view that this was a task that merited a large-scale, dedicated contract. For any translation machine, the problem was how to index and file a whole language so that it could be easily retrieved in microseconds. It was a problem of computer storage, therefore, as much as a cryptographer’s or linguist’s intellectual game.
The NSA had commissioned Raytheon, the defense and electronics giant, to work on this problem at its labs in Waltham, Massachusetts. Project Nomad, as the program was named, was an attempt to create a filing system that could be used to convert a human language into a set of searchable data points. Roughly once a month, Buck was sent to Raytheon to check up on progress. Yet his dispatches back to Washington told a story of a constantly changing plan with little material progress.
As Campaigne, one of the recipients of those dispatches, explained in a 1983 oral history interview conducted for the NSA,
Nomad was a flop. We never got anything much out of it.… We let a contract to Raytheon, up in Boston, to do this and we had a bad experience with them. Their proposal sounded very imaginative. It really sounded good. The minute we signed the contract everything changed. They no longer were imaginative. They were picayune about changes.
Anything that we mentioned was a change in contract, change in scope, change in price. I suppose that’s what they did—they took the imaginative man off and put him on something else and they put some other person on it. He was a fiscal type person.
The venom in the phrase “fiscal type person” says a lot about the attitudes of the day. Budgets were there to be blown; the mission was to get ahead of the Russians.
Project Nomad ran for two years before it was ditched in June 1954. The NSA concluded that it was so far off track that the system was likely to be obsolete by the time it was finished: “When the books were closed, the contract was shown to have cost the government approximately $3,250,000,” the agency wrote years later in a review of its computer programs.
Nomad’s colossal budget was not entirely wasted, however. Buck’s notes and diaries show that he managed to get the Raytheon team to do a lot of experimentation with some other computer systems he was developing. They also dabbled with the ferroelectric memory chip that had formed the basis for his thesis.
On his trips to Raytheon, Buck found himself addressing teams of 450 or more engineers on the finer points of his newest inventions. The engineers were then sent off to do his research as part of the taxpayer-sponsored project.
As for the problem Nomad was supposed to solve, Buck later came up with a solution on his own. He created a system called content-addressable memory, a filing method that allows a big database of information, such as the entire contents of a language, to be searched in record time. It was developed as part of the work on the Whirlwind machine.
Content-addressable memory is one of the systems that allows big data computer systems to sift through vast amounts of information and pull out only the relevant parts. Rather than work its way through every piece of data sequentially, it can search them all at once.
“A normal addressed memory is like a list of telephone numbers,” explains David Brock of the Computer History Museum. “And you call those numbers to see who is home. If you are searching for someone, you would go through all those memory locations and call all those numbers until you find the right person. Content-addressable memory is a way to call all the phones at one time and ask, ‘Is Dave home’? It registers when someone says, ‘Yes, I am here.’ It’s really fast.”
Of all Buck’s inventions, content-addressable memory was the one that—thus far—has had the most direct impact on the age of the Internet and modern computing. At the time, it was just another solution to just another problem dictated by the military.
“Content-addressable memory continues to be a thing,” notes Brock. “It is one arrow in the quiver of modern data scientists. There’s a lot of artificial intelligence that does not use content-addressable memory. But it’s still in use. The first patents on that are the work of Al Slade and Dudley Buck. They built one for the NSA. I haven’t seen earlier stuff.”
Inventions were coming thick and fast across the industry. With so much money being spent, and so many brains dedicated to the cause, the American computer scientists were innovating at a ridiculous pace. What was happening under the guidance of the NSA was substantially ahead of anything available commercially. In late 1951, Computer Research Company in Hawthorne, California—one of the early computer component manufacturers—was advertising a basic computer circuit that could operate at a speed of twenty kilohertz. By the following spring, at the inaugural meeting of the NSA’s Working Group on Ultra High-Speed Computer Circuits in Corona, California, Zoltan Bay had already developed a circuit for the NSA that could do the same thing at one hundred megahertz—that is, five thousand times faster.
Buck was only part-time with the NSA. Yet, increasingly he was clearly part of the agency’s upper echelons. On occasion he would travel with Lou Tordella, who would go on to become the longest-serving deputy director of the NSA. Almost monthly, Buck would be asked to Washington to brief the Department of Defense or the NSA on the latest developments in the world of computing and how they could be applied to the military.
Bizarrely, in spite of his prolific output Buck was still just a graduate student at MIT. He would have to wait some five years before earning his doctorate. Many of the students who worked alongside Buck at MIT had no idea of his alternative existence as a government agent until they were contacted during the research for this book. To many of them he was just a cheerful young lab rat who ran a scout troop in his spare time and happened to disappear for long periods at a time.
“Dudley Buck has a remarkable ability to apply his knowledge of physics and electrical engineering to engineering problems,” wrote his lab supervisor David Brown in a December 1953 staff appraisal—which rated the young scientist as “superior” and recommended a maximum salary hike. “Of all the engineers we have known, he probably derives the most use of his academic training. His understanding of physical principles is excellent. His ideas for new techniques have been very useful for the laboratory program. Another outstanding characteristic is his enthusiasm. This enthusiasm is very stimulating for a group program such as that in the Digital Computer Laboratory. His excellent personality and breadth of interests have made him the recipient of many technical liaison assignments to other laboratories.”
With the Soviet threat escalating day by day, Buck’s work in the MIT lab and for the NSA were about to move more closely together. Buck had already started to come into contact with Soviet agents, keen to talk about exchanging equipment and information.
IN THE WEEKS after Bar-Hillel’s machine translation conference, Buck asked permission of both Dean Acheson, the US secretary of state, and General Walter Bedell Smith, director of the CIA, to visit Moscow.
The details of his request are unclear; the only reference appears in declassified interdepartmental correspondence discussing the idea. It seems that Buck was attempting to procure materials from one of his Soviet counterparts—and possibly to use that as an excuse to find out more about Soviet computing developments.
The creation of the iron curtain had interrupted trade in many basic elements and metals, including many of the key components needed to make experimental computers. It was not always easy to get hold of things that came out of the ground in the wrong countries.
“The DCI [director of central intelligence] instructed me to confirm with Doc [H. Freeman] Matthews of State that the Department had no objection to Mr. Buck visiting Moscow, upon the understanding that no barter deal was contemplated,” wrote one anonymous CIA official regarding the request. “I later confirmed this with Undersecretary Matthews, who said the department would accept our judgment as to the value of such a trip, and had no objection thereto, so long as Buck did not undertake to deliver copper or lead, or other prohibited items, to the Soviets. The Department would have no objection to having Buck purchase Manganese for cash.”
The USSR had a long-standing company, Amtorg Trading Corporation, dedicated to handling its imports and exports. Amtorg had offices across America, and in the past had signed deals to bring Ford automobiles, International Harvester agricultural machines, and DuPont chemicals to the Soviet Union. It had also brokered the sales of Soviet timber, flax, furs, and caviar.
One way or another, Buck started talking to Amtorg, whose agents seem to have turned up unannounced at MIT to attempt to broker some kind of trade.
As 1952 turned to 1953, the administration of President Harry S. Truman handed power to Dwight D. Eisenhower. The tone of follow-up letters in the new regime suggest that there was a degree of frustration in officialdom about Buck’s dealings with the Russians.
Samuel W. Anderson, the assistant secretary of commerce, wrote a memo for Allen Dulles, the new director of the CIA, informing him about Buck’s attempted Soviet trades:
Mr. Buck telephoned again to our Mr. Frederick Strauss yesterday and reported that his Amtorg contact had paid him another visit in order to talk further about trade, particularly two-way trade. Mr. Buck stated to Mr. Strauss that he had informed his Amtorg contact that the possibilities of any progress might be substantially advanced if he, Mr. Buck, knew more about what the USSR desired in the way of imports from the US. The Amtorg representative stated that he would attempt to develop this and promised to “cable.”
Mr. Strauss has made it clear to Mr. Buck that he is not to disclose his conversations in the Commerce Department. I have asked Mr. Strauss to see to it that I meet Mr. Buck, which is now planned for June 12. Mr. Strauss believes that Mr. Buck may have further information by that time. I have instructed Mr. Strauss to evince my interest in the most casual way.
As you will see from this memorandum, Mr. Strauss, before I knew about this case, has generally informed Mr. Ray Vernon of State, I have instructed him not to continue to advise Mr. Vernon of the situation, unless you wish us to do this. I thought you might have other desires as far as informing the Department of State.
A handwritten scrawl on the archived version of this memo confirms that Dulles saw the note. A second notation indicates that the CIA told Anderson’s assistant, Arthur Frazer, that they should keep informing the US Department of State of Buck’s movements as they had previously.
Buck traveled extensively during the period between these two letters. He was not only working on the machine translation project, and Blois’s experiments with sea mud, but it was also an extremely busy time in the lab.
There is no evidence to show that Buck ever went to Moscow. It is possible that one of the many excursions diarized as a visit to one military installation or another was a cover for a trip behind the iron curtain, yet that seems unlikely.
What it does prove, however, is that Buck’s name was known already to two consecutive directors of the CIA—both of whom seem to have considered him an appropriate conduit to trust with backchannel conversations about breaking Cold War trade embargoes.
More to the point, the exchange serves as proof that, seven years before his death, Soviet agents already knew about Dudley Buck—and had a habit of turning up at his office to make demands of his time.