Signs are everywhere that the AI frenzy may be cresting. As Ed Zitron notes in an acerbic essay on the shaky financial underpinnings of ChatGPT pioneer OpenAI and the associated inability of Microsoft to make money out of this technology, “the literal leader in productivity and business software cannot seem to find a product that people will pay for, in part because the results are so mediocre, and in part because the costs are so burdensome that it’s hard to justify them.”
Little Profit Down the Road, But Hallucinations Aplenty
Microsoft customers, for example, are skirting the firm’s ballyhooed and costly AI “Copilot” feature in droves because they “don’t find it that useful.” This raises awkward questions for the data-center industry currently paving over large portions of Northern Virginia and elsewhere, powered by $150 billion and more from Microsoft, Amazon, Google and Oracle - money that they might never get back if or when the expected bonanza from products such as Copilot fails to materialize. At a basic level, billions pumped into AI development have failed to crack the problem of “hallucinations,” the tendency of AI to make things up, or come up with the wrong answer.
As example, consider this recent exchange I had with Google Gemini:
Spoils of War is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
How Many People Will Want a “Ferrari” Chip?
Much of the AI investment from the major tech companies has been in hugely expensive GPU chips endowed with massive computing power from the Nvidia corporation . Nate Koppikar, co-founder of the San Francisco investment firm Orso Partners, who takes a dim view of the whole AI phenomenon, compares those chips with enormously powerful Ferraris - as much as 1000 horsepower - which are not only very expensive but essentially useless for the purposes of everyday living, so the vast majority of car-buyers opt for comparatively modestly-powered machines, yet the big tech companies, he says, are proceeding on the assumption that there is a mass consumer market that will justify these vastly costly supercharged chips. “There’s not an obvious consumer use for AI yet, and we’re most of the way through 2024,” two years into the bubble launched by OpenAI’s ChatGPT in November 2022,” Koppikar told me. “No one at the big tech companies knows what they’re doing with these chips, but it looks like they’ll turn to Uncle Sam” as the one customer who will loyally pay up, however disappointing the returns.
Share
Uncle Sam to the Rescue!
Uncle Sam, in the form of the Pentagon, is ready and waiting, with legs akimbo. These days no utterance from the Defense Department is complete without deferential reference to the essential role that artificial intelligence will play in any and all aspects of our defense apparatus. “New operational concepts must also incorporate emerging technology, such as artificial intelligence (AI) and autonomous systems, as these technologies are fundamentally changing modern warfare” exhorts The Commission on the National Defense Strategy, a bipartisan body set up by Congress in its recently released final report. The Senate version of the 2025 defense authorization bill directs the Pentagon to set up an “Artificial Intelligence-Enabled Weapons Systems Center of Excellence.” Ask Sage, one of the innumerable tech startups getting in on the action, touts all the tasks its AI system could automate, ranging from software development to delegating to machines the burdensome business of buying weapons, including tasks such as “scope of work, defining requirements, down-selecting bidders and much more.”
Plan? What Plan?
Thirsting for a dose of reality, as opposed to hype and fantasy, I consulted a Pentagon insider currently and intimately involved in some of our more costly weapons development programs, for his observations on the excitement over AI. “There’s no coherent plan,” he assured me, “People are just saying ‘we have to have some. They don’t have a problem in mind that they’re going to solve with artificial intelligence.” As example, he cited the Collaborative Combat Aircraft program, a project currently generating enthusiastic excitement among the Air Force high command, its industrial partners, and Congress. The scheme envisions sending human-piloted combat planes into action accompanied by “loyal wingmen,” drones under the overall direction of the human but enabled by artificial intelligence to maneuver and fight independently.
Curb Spending? No Way!
To a generation nurtured on video games, the scheme may seem an entirely feasible proposition. The drones would supposedly act as “force multipliers” in battles with enemy fighters, or deliver missiles or bombs on targets in areas too risky for a human pilot. So confident is the Air Force leadership in the concept that money has already begun to flow in quantity. In April, Air Force Secretary Frank Kendall unveiled contract awards to Anduril Industries and General Atomics Aeronautics for “detailed designs, manufacture, and testing of production representative test articles under the Collaborative Combat Aircraft program,” leading ultimately to a production run of as many as 2,000 at a cost of $30 million each. This first phase is slated to cost just under $9 billion over the next five years. As a sure sign of things to come, the Air Force is already demanding an extra $150 million - a 40 percent boost over previous estimates - to its CCA budget for this year alone. Some lawmakers had the temerity to suggest a cap on program spending, but Kendall has said he is “strongly opposed” to any such curb.
Make it Like the Movies!
Despite all the high-level boosterism, my insider source reports underlying chaos: “There’s no coherent plan for this. There’s no vision for how this resolves problems with the manned aircraft, it’s just ‘go to work on this project and bring me back something that smells like what I saw in the movies.’” The Air Force team tasked with the effort are, he observes, “very young and very junior officers” who display a touching faith in the assurances and blandishments of the Anduril and General Atomics staff.
Who Will Talk to the Drones?
Most pertinently, there has been no constructive thought devoted to the tricky problem of how the “quarterback,” i.e. the manned aircraft, will communicate with and direct the drones, which is hardly surprising because there has been no decision about which aircraft will actually do the controlling, whether an existing plane such as the F-35 or the F-22, or a whole new yet-to-be developed system. (Bear in mind that the F-35 has suffered from endemic software problems even without the hugely complex task of outfitting the system for CCA-related software.) Until recently the Air Force was reportedly embarked on a “Next Generation Air Dominance” fighter, endowed with wondrously sophisticated futuristic technology. Kendall called it his “highest priority” and let it be known that development plans had reached the engineering and manufacturing phase. The Defense Department Inspector General was suspicious enough of the boasts to announce an investigation, following which the project appears to have been shelved, at least for the time being.
“I have no idea what [the CCA team] are doing,” my friend continued. “I can’t see work being done. I’ve seen a lot of stuff on PowerPoint and a lot of made-up stuff on the schedule. As far as I can see, the only reason we’re doing it is to spend the money.”
Political Engineers At Work
To the discerning eye, there is much to be gleaned from official pronouncements regarding the program. Speaking at an Air Force “Life Cycle Industry Day” in July, air force General Jason Voorheis revealed that both Anduril and General Atomics might end up sharing the ultimate spoils, each garnering a production contract. So much for competition. Voorheis also indicated that the political engineering in which contracts are spread around early, thus establishing a lobby as a guard against any effort to kill the program, might extend to foreign countries. All this before the machines in question are designed, let alone tested.
Gliding Toward $3 Trillion
The National Defense Strategy Commission referenced above is calling for putting defense spending “on a glide path to support efforts commensurate with the U.S. national effort seen during the Cold War.” The authors discreetly refrained for putting a dollar figure on this, but my friends at the Stimson Center’s National Security Reform Program calculate that we will be gliding toward a $3 trillion defense budget, at least, by 2034.
That should take care of big tech’s $150 billion investment, and then some.