The Manhattan Project and AI: a Great Nation Coming Undone for Private Profit?

Our nation’s Manhattan Project developed nuclear weapons from nothing more than bare theoretical physics, plus a single experiment that proved nuclear fission possible. The Project did so in an extraordinarily short time: 957 days, or about 2.6 years. Our government kept it all so secret that virtually no one outside it knew of it. There was almost zero public notice, even during the short time between the visible first “Trinity” explosion at Alamogordo, NM, and the first use of a weapon on Hiroshima, Japan.

The Project was entirely run and financed by our government. The top dog, absolutely in charge (below President FDR), was an Army general with an engineering background named Leslie Groves.

Was he as smart as the brilliant nuclear physicists he managed? Probably nearly so. They respected him and followed his leadership because he was knowledgeable, a good “people person,” and a tough, practical manager. He was able to understand enough of the abstruse physics to do his job and to distinguish fundamental roadblocks from problems that just needed more people and more money to solve.

The scientific leader, Dr. J. Robert Oppenheimer, had similar skills. His mild personality and high intelligence gave him the respect of his fellow physicists, and he could understand the physics challenges in detail. So high a priority did our government give the Project that, at one time, ten percent of the nation’s entire electrical output was devoted to spinning the centrifuges that separated fissionable from non-fissionable Uranium for the Bomb.

Simply put, the Project was the greatest, most successful, and most successfully secret crash science/technology/military project in human history. It still is. And most historians and political thinkers still thank whatever God they believe in that the world’s foremost democracy, then imbued with and practicing Enlightenment values, got the Bomb first, instead of Adolf Hitler’s Third Reich.

Now let’s compare today’s development of AI. First of all, how important is AI?

At very least, it’s a world-beating advance on Egypt’s ancient Library of Alexandria—the first centralized collection of written work since writing had evolved from scratches on clay tablets to record deals in the marketplace. In theory, and mostly in practice, today’s AI is an index to all of humanity’s written knowledge (at least all in English and a few Roman-alphabetic languages). And it’s an index accessible to any literate person with online access, anywhere on Earth. Even non-English speakers can use it with the aid of Google Translate.

This is the closest our species has ever come to having a “hive mind.” Think of it. Anyone with a computer terminal or a cell phone—or even access to one—can reach the sum total of human writing, including access to great thinkers long dead!

There are downsides, of course. Not every recorded human thought is brilliant, let alone enlightened. The sum total of human writing contains a lot of false, rotten and even dangerous stuff. And today the total written output of all nine billion of us is exploding exponentially as many of us pound away at our keyboards (or now dictate into machines) and throw the results online.

AI mavens of course limit what goes into “training” their AI’s, both for economic reasons and to separate the wheat from the chaff. But that gives the people who select the training materials a lot of implicit power, doesn’t it? It raises again, in a different form, the old Roman question “Quis custodiet ipsos custodes?” or “Who will guard the guardians?” It’s eminently possible to push the hive mind gently, or even firmly, in one direction or another by carefully selecting what you feed it as training materials.

Is that going on today? I think so, if only inadvertently. Somehow I can’t see the “tech bros” as focusing as heavily on Greek, Enlightenment and Renaissance philosophers as they do on current events and stuff they personally value in their daily lives, if only because the latter categories have so much greater volume. And the volume is increasing exponentially as vastly increasing private crews expound on daily news and bloviate on social media. (One practical advantage of older materials for “training” purposes is that they have fallen out of copyright and thus are free of the influence of the big lawsuit in which copyright owners are demanding control over their content and/or royalties.)

Is there any chance that ancient wisdom will prevail over presently proliferating nonsense as time goes on? Or does it all depend, in the end, on what the tech bros and their assistants (both human and robotic) choose as “training” materials?

This is only one of the unintended consequences of AI. Another has been widely discussed: how do people—especially young people—learn anything personally when all of human “wisdom” is available to them online and nearly instantaneously? What’s the motivation to learn facts and history when all of that can be had for the asking? What’s the motivation to learn math, for example, when machines can do it for you faster and more accurately?

One answer, I think, is the philosophy I adopted as a professor of law. My job, I thought, was to teach students how to think, not what to think. That’s why I bent over backwards never to reveal my politics or even political philosophy to my students until after I retired. (To that end I kept my personal blog anonymous until then.)

But that was relatively easy to do in teaching law. In the US, law schools use the so-called “Socratic method,” in which professors pose the questions and students answer in class as called upon (with occasional volunteers). And the students give their answers, each day, based on a common set of assigned readings and their general knowledge accumulated in other courses and their earlier education.

Can you teach history, physics, biology, ethics, mathematics, etc. that way? I suspect so, but I can’t be sure because I’ve never tried.

The point here is simple but profound. How do you teach students to reason when all of recorded knowledge is available at their fingertips? When masses of facts and thoughts are available online that far exceed the capacity of the human brain, how do you teach students to select and absorb what matters? What matters enough to commit to memory? What skills require practice, and how do you inculcate their importance? How can you train students to think clearly when so many facts and events vie for attention, let alone with exotic half-clothed dancers distracting their attention in the margins of their screens? How do you inculcate the value of learning when almost anything worth learning can be had by “asking an AI”?

Aren’t these epistemological questions existential? As our daily news reflects, AI puts the entire subject of education, let alone evaluating students, up for grabs. At the end of the day, it puts the entire structure of society up for fundamental re-evaluation.

It also raises the risk that the bros who control the AIs and feed them with training materials can become a new isolated priesthood—a new ruling class—much like clerics in the Dark Ages. After all, they can influence, if not determine, not just how, but what, AI users think. AI can become a choke point for the general knowledge of a population far more powerful than the scriptures of old.

So now we come to the sharp point of this essay. The Manhattan Project was the greatest, most secret technological crash program in human history. Yet it was set up, run, and funded entirely by our government, ostensibly for the benefit of the entire population, and (given the dread possibility global Nazi supremacy) the world. In contrast, the development of AI has been left, by design or default, to a handful of tech bros and private companies, for their own enrichment and aggrandizement. That fact, along with other economic inattentions, is visibly leading to a Second Gilded Age more widespread and even more unequal than the first.

Is this appropriate for an ostensibly democratic, egalitarian society? Or is it the harbinger of a new age of aristocracy, oligarchy and sophisticated social control more dreadful and persistent than anything Orwell could imagine?

Finally, there’s the question of using AI as a weapon. Silicon chips “think” several orders of magnitude faster than human brains. So if you want to create malware to attack, subvert, or just shut off a rival nation’s computer systems and everything that depends on them, AI is by far the best way to see how. There’s probably no software program ever written (at least by human programmers) that is not vulnerable to hacking by AI.

Furthermore, hacking a rival’s or enemy’s computer systems is consistent with the mostly-unnoticed, recent fundamental changes in the nature of warfare. The current wars in Ukraine and in the Middle East are prime examples.

The days of vast armies marching across the empty countryside and into cataclysmic battles, or into big cities to occupy them, are waning. The last century’s two world wars were of that kind. So were the Korean War and our dreadful misadventure in Vietnam.

But the trend is clear, for two reasons. First, no one wants such wars anymore, and people who have a choice are more and more reluctant to fight in them. Second, as a practical matter, armies of humans are far too vulnerable. Things like tank busters and napalm from the air started the trend, and autonomous AI-controlled drones (as in Ukraine) have brought it to its logical conclusion. Even in tanks or humvees, humans are too slow, vulnerable and weak to fight the machines we can build. All human armies can do now is slaughter each other mechanically and autonomously. (This is one of many reasons why Secdef Hegseth’s physical preening and emphasis on physical fitness, let alone his tattoos, are so deeply misguided.)

The second trend is mostly unnoticed, at least in the military field: globalization. Global humanity is too interconnected economically for any assault on one part not to have repercussions on another part. The war in Ukraine demonstrated that fact, as Putin’s military assaults on its “breadbasket” imposed famine as far away as Africa and East Asia, leading to near-global economic pressure. Trump’s and Israel’s misguided war on Iran is an even better example: by cutting off the global flow of fossil fuels and maiming oil and gas infrastructure, the war imposed hardship on people around the globe as far away as Japan and China, including US drivers.

Even Trump, with an IQ that motivated legal action to keep his college grades and test scores secret, finally seems to have gotten the message. Sheer military force, let alone if applied indiscriminately, doesn’t work so well anymore. The larger its scope and the more indiscriminate its use, the broader its global (and unintended) economic impact. And the larger the political pushback from all corners of the Earth.

So our species is now entering a new era of military strategy that turns Von Clausewitz’ famous dictum on its head: economic pressure and politics are war by other means. Isn’t that precisely what’s happening now in the Middle East, if not also in Ukraine? Our slow-to-learn species is finally beginning to understand that we all live on the same planet.

If that is so, then hacking vulnerable computer systems fits right into the big picture. If you can bring a rival’s industry, transportation, communication and military systems to a halt, or to a state or relative impotence, without (directly) killing anyone, then doesn’t hacking become war by other means? And isn’t it actually preferable to wholesale slaughter?

If hacking, like autonomous drones, is the future of warfare, and if AI is by far the most effective and expeditious way to design and thwart hacks (and maybe to control drones), isn’t AI potentially the best focus for both defensive and offensive military effort? And, if that is so, what the hell is AI doing in the exclusive hands of fabulously wealthy and anomalously youthful tech bros, for their own financial aggrandizement, without carefully structured government supervision in the public interest?

Imagine an alternative history in which the Manhattan Project had been run by General Motors and the Rockefeller Trust. We’d probably all be speaking German. Guten tag! And New York, Chicago and maybe Dallas might have been like Hiroshima and Nagasaki.

With all his undeniable Irish charm, Ronald Reagan may have been our second worst president, despite encouraging Mikhail Gorbachev to “Tear down that wall!” He jet-propelled the trend among Americans to distrust their own government in favor of private greed. What a smart and stealthy way to build a durable oligarchy!

Now we are experiencing the consequences of that transformation. One of the most consequential technologies in human history is under the complete control of self-important and unregulated young men to use willy-nilly for their own profit and personal aggrandizement, without a thought to social, political or military consequences. (And all this is without regard to the ultimate risks of an AGI emerging or AIs starting an otherwise unprovoked nuclear war.)

Who elected them? And what could possibly go wrong?

In light of these facts, Trump’s feeble and belated attempts to keep at least the hacking-trained parts of that technology out of rivals’ hands appear justified, if not particularly effective. No doubt Leslie Groves and J. Robert Oppenheimer are rolling over in their graves, fast enough (if harnessed to generators) to help solve the coming AI-provoked electricity crisis.