The Many Worlds of Hugh Everett III: Multiple Universes, Mutual Assured Destruction, and the Meltdown of a Nuclear Family
Page 34
To simplify: Linear missile trackers continued to predict a straight line trajectory long after turns had occurred. But Cassandra created a “skyful” of all possible trajectories—branching paths—and randomly extracted a sample of the branching trajectories. The program then assigned to each sample a probability weight representing the likelihood or belief that the sample contained the true position of the missile through time. The tracker continually updated the sample with evidence from newly obtained radar data. Using Everett’s algorithmic application of Bayes theory—his Bayesian Machine—Cassandra closed in on the actual trajectory by a process of deduction.
Unfortunately, it is a law of ABM research that the enemy can cheaply thwart just about any tracking mechanism with metallic decoys. The best defense is not to get into a fight with people that can throw nuclear missiles. Defeated by the ABM paradox, Lambda successfully applied the logic of the Bayesian Machine to designing air traffic control software for the Federal Aviation Commission, and, as passenger jetliners do not usually duck and weave, it worked well for years.
Many worlds machine
Lucas said there is a strong analogy between Everett’s multiverse theory and the logic of the Bayesian tracker. The analogy is interesting because, although Everett did not use Bayes rule in the construction of his many worlds theory, some advocates of the many worlds theory at the University of Oxford approach the role of probability in that theory from a Bayesian point of view.5
Lucas explained:
The measurement problem in quantum mechanics in which a measurement of an object, say a photon or an electron, is affected by the measurement itself has a parallel in the Bayesian tracker. The likelihood that a particular virtual vehicle is the real or, more precisely, close to the real vehicle is altered by an inherently imprecise radar measurement. Bayesian statistics are used to revise the probability of the virtual vehicle being the actual one. Since the distribution function for the real reentry vehicle is reflected by the density of virtual reentry vehicles, an entirely new set of virtual reentry vehicles must be constructed with each radar measurement.
The entire process has a very quantum mechanical, multiverse-like feel to it. Very Hugh Everett like.6
31 The Death of Lambda
Life, language, human beings, society, culture—all owe their existence to the intrinsic ability of matter and energy to process information. The computational capability of the universe explains one of the great mysteries of nature: how complex systems such as living creatures can arise from fundamentally simple physical laws. These laws allow us to predict the future, but only as a matter of probability, and only on a large scale. The quantum-computational nature of the universe dictates that the details of the future are intrinsically unpredictable. They can be computed only by a computer the size of the universe itself. Otherwise, the only way to discover the future is to wait and see what happens.
Seth Lloyd, 20061
The future of software
In the late 1960s, the Secretary of Defense awarded Lambda a contract to develop the CODE 50 Model, a new war gaming program that used Everett’s multipliers to model a variety of “force postures” during a nuclear exchange. Complementing QUICK, one of CODE 50’s jobs was to calculate how much of the American nuclear force could survive a Soviet surprise attack and effectively retaliate.
As a young Army lieutenant, Jan M. Lodal worked in the Pentagon’s Office for Systems Analysis overseeing Lambda’s work on CODE 50. He says that QUICK and CODE 50 were used by the high command to double check what the Strategic Air Command’s programmers in Omaha were doing with the SIOP. “The SAC war planners had a set rule of thumb: every target needed to be hit by two warheads, unless it needed three. Prior to Hugh’s work on CODE 50, every time the air force sent up a request for 10,000 nuclear weapons they got them.
McNamara had no way to definitively answer Air Force procurement arguments by asking SAC for their calculations. They would say, ‘We can show you calculations, but that cannot prove that the Soviets would not have lots of weapons left after the initial salvo.’ But CODE 50 allowed us to determine what the real limits of a Russian attack were, so we could put in realistic numbers without making ridiculous assumptions about their capabilities. And that gave us a measure of assured destruction.2
Lodal has long been an influential voice in national security circles. He is a stickler for using the phrase “assured destruction,” not “mutual assured destruction,” because, he says, our own destruction is not part of the plan. He elaborates,
Especially after the deployment of submarines and MIRVs nobody knew how to define winning. Game theory did not make much sense given the high level of uncertainties, and the overwhelming capability of the weapons of both sides to wipe each other out. And nobody could come up with an escalation control scheme where the whole world would not be covered with fallout.
What Hugh proved to us with CODE 50 was that by 1970 there existed no totally disarming attack by the Soviets that could destroy our ability to retaliate.
Naturally, this worked both ways: a U.S. first strike was not likely to cripple the Soviet second strike force either, so, terminological caveats aside, mutual destruction was assured.
In the 1970s, Lodal was an aide to Secretary of State Kissinger. He attended summit meetings, sitting at the table with Soviet leaders, Andrei Gromyko and Leonid Brezhnev. His job was to provide Kissinger with technical expertise. He says that to the extent détente was successful, it was partially due to Kissinger’s ability to marshal the technical details of assured destruction. And he credits Everett: “If I had not had Hugh as my mentor to teach me about force levels and nuclear parametrics, and if I had not had his computer models it might not have happened.”
Lodal is still amazed by Everett’s genius. “I find it astounding that he built both sides of CODE 50. He invented the mathematics and approximation techniques and then designed the FORTRAN program to run the calculations.” He saw Everett almost as an physical extension of his computer. “The IBM 1604 had a loudspeaker that played tones as it was computing. Twenty minutes into a five hour run, just by listening to the tone, Hugh could tell where the computation was and say, ‘Shut it off, it’s making errors.’”
But what most astonished Lodal about Everett was his invention of “attribute value” programming. This path-breaking concept emerged from CODE 50 and QUICK. It was the first iteration of a “relational” database, a data sorting application of the sort later sold by Oracle and PeopleSoft. Lodal points out that the XML method now does for high speed computer systems what Everett’s attribute value method accomplished for relatively low speed, small memory computers in the 1970s.
At the time, vast amounts of the Pentagon’s war-gaming data was stored on spinning tapes or disks. Information was filed willy-nilly on the recording medium, so a search algorithm might have to parse the entire database to find a single byte. Lodal describes how Everett’s attribute values revolutionized searching, and how the method reflected his many worlds theory.
He realized that in data processing if you wanted to avoid entanglements in the code and preserve complete data sets without losing information content that you have to decide at the atomic level of the data base what items you most care about, what are the fundamental attributes of types of data, like names, ages, income in a social security database, or the various values defining destruction in a war gaming model.
Instead of mapping information by time or position on the tape, Everett assigned numerical tags to sets of data, categorizing them. Number 444, for example, would call up names of targets, 354 would call up the probability of a certain type of missile reaching a target, 666 would pop up its kill probabilities. Then, to further simplify and speed up the search process, says Lodal,
Hugh invented probably the first efficient compression algorithm. It worked exactly like a three-dimensional spreadsheet with columns and rows, so you could pinpoint a certain value. Except Hugh’s spreadsheet was N-dimensional. He
could deal with an uncountable number of dimensions, just like his uncountable number of branching universes in his quantum theory, with no limit on the number of dimensions available.
The stuff was amazing effective. Intellectual. Crisp. Precise. Like how he treated the Schrödinger equation: he said, ‘Lets not get complicated, lets start with the simplest assumption, that it evolves linearly.’
Lodal’s former boss, assistant secretary of defense, Ivan Selin, says Everett’s innovation may seem trivial today, but at the time, it was mind-blowing.3 In 1970, Lodal and Selin and another Pentagon official, Charles Rossotti, formed American Management Systems, a business consulting firm based upon Everett’s attribute value concept. AMS soon became one of the largest, most successful information management consulting groups of the late 20th century.
AMS had its first big success in the mid 1970s when it used Everett’s programming breakthrough to restructure the books of the nearly bankrupt City of New York. Lodal was featured on the front page of The New York Times as having saved the city from financial meltdown. But he gives much of the credit for that achievement to Everett’s attribute value system:
Hugh would say, ‘We have to make the values independent of each other, orthogonal.’ He conceptualized the data as consisting of glops of information in an n-dimensional Cartesian space where you could represent any collection of related data by a single point in that space. Instead of writing thousands of rules into the code, he came up with a general concept that applied to all of the data at all points.
And that was always Everett’s genius: seeing ways to solve seemingly intractable statistical problems in computer science and quantum mechanics through simplification.
Unfortunately, he did not possess a similar eye for solving business problems.
Lambda sinks
Lambda’s annual report for 1970 boasted that the company had 50 employees, most of them with PhDs, and annual sales over $1 million. In addition to designing nuclear wars, Lambda studied the best way to kill Vietnamese soldiers with conventional firepower. It had a contract to study “continuity of government,” i.e. plans to institute martial law domestically in the aftermath of natural disaster, nuclear war, or insurrection. It designed data compression systems for the World-Wide Military Command and Control System, which was the Pentagon’s global communications network. It wrote programs to train fighter pilots to maximize kill ratios in air combat. And at the same time it was tuning the war machine, it was designing an “arms control simulation study.”
Although contracts with the Central Intelligence Agency and the National Security Agency were not listed in Lambda’s official reports, they were listed in private spreadsheets kept by Everett, who personally controlled these sensitive jobs.4
Lambda also held a lucrative, politically sensitive contract with the Department of Health, Education and Welfare to assess busing as a school desegregation strategy. After Lambda produced an influential study showing that a large amount of desegregation could be obtained with a relatively small amount of busing, the company was contracted to write desegregation plans for most major metropolitan areas in the United States. But in 1972 the national press ridiculed Lambda as a bastion of clueless eggheads when it was revealed that its plan for the Washington D. C. area required students to walk across a busy interstate highway to get to the nearest bus stop.
The busing model was only a prototype and someone (perhaps an opponent of busing, Lucas suggests) leaked it to a Washington Post reporter. But the scandal came at a time when military think tanks were increasingly under attack for constructing idealized models of reality that were causing terrible bloodshed in southeast Asia and failing to work as advertised (e.g. carpet bombing Hanoi, napalming south Vietnamese families, assassinating political dissidents in Saigon). Lambda’s government contracts were cut back after Congress questioned the wisdom of paying $585 million a year to 700 think tanks that could not figure out how to defeat an army of peasants wielding sharp sticks and second-hand rifles against the best-equipped military in human history.5
Reading the writing on the walls of the Pentagon, Lambda tried to wean itself from the military teat. Everett hoped to market his word processing software, or to turn the revolutionary attribute value concept to commercial advantage. And he almost succeeded: In 1967, Lambda began a multi-year $600,000 study for Merck Corporation showing the pharmaceutical conglomerate how to transform one-drug factories into modular, just-in-time facilities producing many kinds of drugs. The computerized plan was elegant and workable—but, alas, far too expensive (a billion dollars) for Merck to adopt.
Everett’s ideas were often far ahead of his time, but he was also stubborn and arrogant, and those traits hurt his business. For example, corporations were clamoring for software written in COBOL, a business-friendly programming language. But he absolutely refused to write in any format except FORTRAN, the scientific language. To prove its superiority, he wrote a version of COBOL in FORTRAN!
Desperate for income, Lambda tried to adopt its military software to commercial chores such as assigning train track routes for millions of empty freight cars, or improving the efficiency of open pit mining. But in the end, the company was rich in scientists and poor in marketers. Dean wanted to concentrate on getting consulting contracts from industry, but Everett was dead set on making a splash with proprietary software sales. Lambda’s accountants were horrified that he spent $130,000 developing PROLOG, an anti-pirating system for software. He applied for a patent on this idea—which was to invisibly scratch a computer disk in such a way that its program would not decrypt itself and run unless the scratch was detected by an authorized user. But the Patent Office turned him down for lack of specificity in his design. He was unable to sell the product, and became convinced that IBM was trying to steal it from him.
Ironically, even as Lambda was declining, Everett’s attribute value application was doing well for Selin and Lodal and Rossotti, the firm’s former patrons inside the Pentagon. AMS knew how to sell advanced software in the corporate market, whereas Everett was adrift without a compass in the competitive world of business consulting. However, the AMS founders were so impressed by Everett’s general brilliance, that they gave him and Lambda 14 percent of their initial stock offering in return for continued access to his brain. They subcontracted jobs to Lambda; and they made Everett a vice-president of AMS, although he had no administrative duties.
By late 1972, Lambda was drowning in red ink and laying off employees. A sad internal memo, probably authored by Galiano, related:
We seem to me like a steamship, halted at sea, slowly drifting onto a reef. At first reluctantly, but eventually gaily, we begin to throw overboard the brass fittings, the cargo, and finally the crew, while hoping to float over the frightening coral. But, is the surviving derelict worth the price? Shouldn’t we be below deck, stripped to the waist, stoking up the furnaces for all we’re worth right now? Two years of drifting have left us precious little time…. We must model ourselves more on IBM and less on IDA memories…. The world, our current and prospective customers, must be made aware of our superior performance, or it may well have been wasted.6
Concerned by Lambda’s operating losses, its 50 percent owner, now called General Research Corporation, stepped in and folded the struggling company into its giant embrace, making it a division, and rewriting Everett’s employment contract. The boilerplate listed as grounds for termination conviction of “a misdemeanor involving moral turpitude.” Everett crossed out that phrase, scrawling next to it, “clarify meaning.”
The ghost of Lambda struggled inside GRC for a few more years, but Everett left in 1973 to form a new company, DBS, with a young Lambda physicist, Don Reisler. DBS (not an acronym) specialized in the analysis of discriminatory patterns in the workplace, providing expert testimony on behalf of the government in federal lawsuits alleging bias. Rossotti was on the board; Reisler was president; and Everett wrote computer programs, when he bothered to go to the office.
Selin says, “Hugh modeled the world in his own mind and then forgot about the world and concentrated on the models, sometimes with brilliant insights, sometimes with ludicrous results. Only occasionally did he intersect with what the rest of us considered to be the real world.”