Saturday, April 6, 2024

"The Pentagon’s Silicon Valley Problem"

From Harper's Magazine, March 2024 issue:

How Big Tech is losing the wars of the future  

Three months before Hamas attacked Israel, Ronen Bar, the director of Shin Bet, Israel’s internal security service, announced that his agency had developed its own generative artificial intelligence platform—similar to ChatGPT—and that the technology had been incorporated quite naturally into the agency’s “interdiction machine,” assisting in decision-making “like a partner at the table, a co-pilot.” As the Israeli news site Tech12 explained in a preview of his speech:

The system knows everything about [the terrorist]: where he went, who his friends are, who his family is, what keeps him busy, what he said and what he published. Using artificial intelligence, the system analyzes behavior, predicts risks, raises alerts.

Nevertheless, Hamas’s devastating attack on October 7 caught Shin Bet and the rest of Israel’s multibillion-dollar defense system entirely by surprise. The intelligence disaster was even more striking considering Hamas carried out much of its preparations in plain sight, including practice assaults on mock-ups of the border fence and Israeli settlements—activities that were openly reported. Hamas-led militant groups even posted videos of their training online. Israelis living close to the border observed and publicized these exercises with mounting alarm, but were ignored in favor of intelligence bureaucracies’ analyses and, by extension, the software that had informed them. Israeli conscripts, mostly young women, monitoring developments through the ubiquitous surveillance cameras along the Gaza border, composed and presented a detailed report on Hamas’s preparations to breach the fence and take hostages, only to have their findings dismissed as “an imaginary scenario.” The Israeli intelligence apparatus had for more than a year been in possession of a Hamas document that detailed the group’s plan for an attack.

Well aware of Israel’s intelligence methods, Hamas members fed their enemy the data that they wanted to hear, using informants they knew would report to the Israelis. They signaled that the ruling group inside Gaza was concentrating on improving the local economy by gaining access to the Israeli job market, and that Hamas had been deterred from action by Israel’s overwhelming military might. Such reports confirmed that Israel’s intelligence system had rigid assumptions of Hamas behavior, overlaid with a racial arrogance that considered Palestinians incapable of such a large-scale operation. AI, it turned out, knew everything about the terrorist except what he was thinking.

Such misplaced confidence was evidently not confined to Israeli intelligence. The November/December issue of Foreign Affairs not only carried a risibly ill-timed boast by national security adviser Jake Sullivan that “we have de-escalated crises in Gaza,” but also a paean to AI by Michèle Flournoy. Flournoy is a seasoned denizen of the military-industrial complex. The undersecretary of defense for policy under Barack Obama, she transitioned to, among other engagements, a lucrative founding leadership position with the defense consultancy WestExec Advisors. “Building bridges between Silicon Valley and the U.S. government is really, really important,” she told The American Prospect in 2020. Headlined ai is already at war, Flournoy’s Foreign Affairs article invoked the intelligence analysts who made “better judgments” thanks to AI’s help in analyzing information. “In the future, Americans can expect AI to change how the United States and its adversaries fight on the battlefield,” she wrote. “In short, AI has sparked a security revolution—one that is just starting to unfold.” This wondrous new technology, she asserted, would enable America not only to detect enemy threats, but also to maintain complex weapons systems and help estimate the cost of strategic decisions. Only a tortuous and hidebound Pentagon bureaucracy was holding it back.

Lamenting obstructive Pentagon bureaucrats is a trope of tech pitches, one that plays well in the media. tech start-ups try to sell a cautious pentagon on a.i. ran a headline in the New York Times last November over a glowing report on Shield AI, a money-losing drone company for which Flournoy has been an adviser. Along with a review of the company’s drone, the story cites:

the many hurdles that the new generation of military contractors face as they compete for Pentagon funding against the far bigger and more entrenched weapons makers that have been supplying the military for decades.

But the notion that the Pentagon is resistant to new technologies is hardly fair—it reportedly funds at least 686 AI projects, including up to $9 billion for a Joint Warfighting Cloud Capability contract awarded in 2022 to a slew of major tech companies destined to, per one Pentagon official, “turbocharge” AI solutions. Another AI project, Gamechanger, is designed to enable Pentagon employees to discover what their giant department actually does, including where its money goes. A press release from the Joint Artificial Intelligence Center from early 2022 that celebrates Gamechanger’s inauguration noted “28 Authoriative [sic] Sources’’ and quoted a senior Pentagon accountant’s excitement about “applying Gamechanger to gain better visibility and understanding across our various budget exhibits.” Even so, the Pentagon failed to pass a financial audit in 2023, for the sixth year in a row.

Artificial intelligence may indeed affect the way our military operates. But the notion that bright-eyed visionaries from the tech industry are revolutionizing our military machine promotes a myth that this relationship is not only new, but will fundamentally improve our defense system—one notorious for its insatiable appetite for money, poorly performing weapons, and lost wars. In reality, the change flows in the other direction, as new recruits enter the warm embrace of the imperishable military-industrial complex, eager to learn its ways.

The belief that software can solve problems of human conflict has a long history in U.S. war-making. Beginning in the late Sixties, the Air Force deployed a vast array of sensors across the jungles of Southeast Asia, masking the Ho Chi Minh trail along which North Vietnam supplied its forces in the south. Devised by scientists advising the Pentagon, the operation, code-named Igloo White, and designed to detect human activity by the sounds of marching feet, the smell of ammonia from urine, or the electronic sparks of engine ignitions, relayed information to giant IBM computers housed in a secret base in Thailand. The machines were the most powerful then in existence; they processed the signals to pinpoint enemy supply columns otherwise invisible under the jungle canopy. The scheme, in operation from 1967 to 1972 at a cost of at least hundreds of millions a year, was a total failure. The Vietnamese swiftly devised means to counter it; just as Hamas would short-circuit Shin Bet algorithms by feeding the system false information, the Vietnamese also faked data, with buckets of urine hung in trees off the trail, or herds of livestock steered down unused byways, which were then dutifully processed by the humming computers as enemy movements. Meanwhile, North Vietnamese forces in the south were well supplied. In 1972, they launched a powerful offensive using hundreds of tanks that went entirely undetected by Igloo White. The operation was abandoned shortly thereafter.

The IBM System 360 computers at the center of Igloo White were a prominent icon of the industry we now call Silicon Valley. Born of the electronics industry that helped secure victory in World War II, this sector flourished under Pentagon patronage during the Cold War. The development of integrated circuits, key to modern computers and first produced by Texas Instruments in 1958, was powered by an avalanche of defense dollars, and initially deployed in the guidance system for the Minuteman II intercontinental nuclear missile. Beginning with personal calculators, the microchip revolution eventually found a commercial market severed from the umbilical cord of government contracts, generating an industrial culture, eventually both physically and spiritually centered south of San Francisco—far removed from the Pentagon parent that had spawned it.

As America’s manufacturing economy gradually declined from the Eighties on, its Rust Belt heartland increasingly studded with decaying industrial cities, the digital economy grew at an exponential rate. Less than a month after its stock market debut in December 1980, Apple Computer was worth more than Ford. It was an industry happy to be independent of the irksome constraints of government contracting. Apple’s Macintosh personal computer, the young company imagined, would free citizens from the Orwellian world of government control “by giving individuals the kind of computer power once reserved for corporations.” But even though the original freewheeling iconoclasts of the tech industry saw themselves as “cowboys,” “rebels,” and “revolutionaries,” in the words of historian Margaret O’Mara, the divorce from defense was never absolute. The internet, hailed as a liberating technology, grew out of ARPANET, which had been developed by the Pentagon’s Advanced Research Projects Agency. According to Yasha Levine in Surveillance Valley, the proto-internet was deployed almost immediately to collect information on the antiwar movement. Google founders Larry Page and Sergey Brin’s early work at Stanford was funded in part by the same Pentagon agency, which by now had added “Defense” to its name to become DARPA. And Google Earth started as Keyhole EarthViewer, a mapping system partly funded by the CIA’s venture capital offshoot In-Q-Tel, which was eventually acquired by Google.

Meanwhile, the Pentagon never lost sight of the unfulfilled dream of Igloo White: that computing power could make it possible to control the battlefield. Following the American retreat from Vietnam, vast sums were poured into Assault Breaker, in which powerful airborne radars would peer deep behind Soviet lines in Eastern Europe. It, too, abjectly failed its tests—the system could not distinguish tanks from cars, or from trees blowing in the wind—and the project was canceled the following decade. Yet military optimism was undimmed; senior commanders broadcast the notion of “netcentric warfare,” and their aspirations found fruit in such projects as the Future Combat Systems program, which linked sensors and weapons via high-powered processors to strike targets so effortlessly that, or so its proponents claimed, it would no longer be necessary to install defensive armor on tanks. But after consuming almost $20 billion of taxpayer dollars, it came to nothing. Same with the Department of Homeland Security’s Secure Border Initiative Network, marketed as a “virtual fence” equipped with computer-linked radar, cameras, and other surveillance sensors to detect intruders, which was canceled in 2011. Boeing had been a prime contractor on both of these baroque endeavors, a sclerotic contrast to the fast-paced dynamism of Silicon Valley. Surely the flourishing entrepreneurial culture of the West Coast could succeed where the old guard had failed.

Peter Thiel certainly thought so. A former securities lawyer, he had garnered his initial fortune as a co-founder of PayPal and grew vastly richer due to an early investment in Facebook....