Log in
When I was at West Point in the 1960’s, we took a class called “Ordnance Engineering.” They waited until you were a senior to teach it, so you had the advantage of three years of studying calculus, differential equations, and multiple disciplines of applied mathematics such as mechanics of solids, thermodynamics, and mechanics of fluids. In other words, they waited until you had the tools to know how to do it before teaching you what it was they wanted you to do.
The problem at the end of the semester, as I recall, was something like, “Design the tube of a howitzer cannon capable of firing a 100 mm 12-pound warhead 10 miles.” They left out some of the difficult parts of the design, such as the spacing of the rifling in the barrel and the design of the mechanism of the barrel’s breach, where the barrel opened to receive the gunpowder and the projectile. But there were plenty of other variables: How long should the gun tube be? How thick was the tube, from its base to the muzzle where the projectile left the tube? How thick was the steel around the bore of the base of the gun tube holding the propellant and the projectile so when the thing fired, it wouldn’t explode the tube into a thousand pieces? How much standard black powder propellant, by weight, would be necessary to send the 12-pound warhead 10 miles?
There was an early model of one of the first computers, the IBM 360, occupying an entire ice-cold air conditioned and heavily cushioned room in Thayer Hall at that time, but we weren’t using it. We used a slide rule to work the problem, as had been done for decades before then.
Cannons are ancient weapons that date to the 13th century in China. Cannons in 1969 were the basic weapons of the U.S. Army in assaulting an enemy. Fire the cannons, “soften-up” the target, then move infantry firing small arms forward to take the objective. Warfare was thus across the world and had been for centuries. The problem we were given tested our ability, after three years of study, to move into the future of designing new weapons. If you could design a cannon, one day after more study, you might be designing missiles which would be used in wars of the future.
I can’t recall what my grade was for the problem in Ordnance Engineering that year – probably not sterling – but I completed the work and designed my own gun tube to fire the warhead the required distance.
My point in recounting this story is to describe what it took to design weapons – years of study of very difficult mathematics and science – and then what you got was the most basic of a part of a weapon, the tube of a cannon. You didn’t get the rest of the thing, its breech, firing mechanism, aiming devices, all the stuff that would make it a usable weapon on a battlefield, which would take more years of study of even more difficult math and science. And still, in the end, what resulted was a modern version of an ancient weapon, a cannon.
The U.S. had and still has very, very good cannons – the modern 155 mm Howitzer M114A being the latest variant is capable these days of firing a GPS guided warhead making the weapon extremely precise. At one point, there was even a nuclear warhead for the 155 Howitzer. They showed us a highly classified dummy example in class one day.
At the time I took that class in Ordnance Engineering, we had shipped all kinds of cannons to Vietnam and were firing them at the North Vietnamese army and Viet Cong. Today, we’ve got them all over the world, in Asia, in Africa, in Europe. We used them in Iraq in two wars, in Afghanistan, and we still have them in Iraq and on bases in Syria and Jordan. We also have our modern ground-to-ground missile system, the HIMARS, all over the world, which fires missiles further than artillery warheads with great accuracy.
But that was the past.
Today, it doesn’t take all that education in math and engineering and a modern economy to come up with deadly weapons. All it takes is Anthropic’s Claude and a laptop computer, maybe even a smart phone, to design and use a weapon, as Anthropic reported last week, a missile capable of traveling 2,000 kilometers, carrying a hypersonic “glide vehicle,” meaning warhead. Anthropic said “a cell of threat actors based in northern Yemen” using Claude “in place of human software engineers” managed to create control and guidance and navigation devices, and fired a ballistic missile. The missile failed, Anthropic reported, and “within hours, the actors returned to Claude to work out why it failed.”
The “cell” was the Houthi army in Yemen, a country with a population of 32 million and a nominal GDP of $17 billion, with a per capita GDP of $416. And those are the numbers for the country of Yemen, the one with a government. The Houthi rebels are a breakaway force that controls only part of the country, but the part they control includes the coast of the Bab al-Mandab Strait at the mouth of the Red Sea, and as of this week, Mayun Island, located right there in the middle of the strategic strait through which billions and billions in trade flows on its way to and from the Suez Canal.
The Houthis have missiles and drones that have been supplied to them by Iran and other “threat actors,” to use the phrase used by Anthropic, and they have been firing those missiles and drones and disrupting shipping in the Red Sea already.
But the news this week is that using the AI technology of Anthropic’s Claude, the Houthis have been able to do for themselves something that until now only countries with great wealth and education systems have been able to do by educating and training people, as I was educated and trained at West Point, to design weapons systems that will one day very, very soon disrupt the entire world.
What should we call this? The democratization of warfare? When people in a country as poor as Yemen are able to develop and use the technology to fire a weapon such as a missile with a warhead capable of traveling 2,000 kilometers – that is not only news, it’s big news. And all they had to do was use an AI system called Claude.
This is why the CEOs of three of the most advanced AI platforms on earth – Anthropic, OpenAI, and SpaceXAI – agreed with each other in calling for a “slowdown” in development of AI “because of its dangers,” as the Washington Post put it yesterday.
Gee, ya think?
Anthropic reported that Russian “threat actors” designed a kamikaze drone swarm that included targeting technology for class of “person” targets “without a person in the loop.” Anthropic reported there were other uses of Claude by China and Mali. One of them simulated an electronic warfare attack on 12 targets in Taiwan, and another targeted “specific US anti-torpedo and anti-submarine programs,” according to Anthropic. The examples went on and on and included Claude ghost-writing official testimony delivered at the U.N. Human Rights Council. Another instance reported Chinese AI systems such as DeepSeek infiltrating Claude users to get American and European “names, email addresses, company data, and other sensitive data of hundreds of end users.”
And this is just one admission by one AI company, Anthropic, about the way its systems have been misused. Or should it be “used?” It’s becoming apparent that the AI companies have developed this technology without putting controls on who uses them and for what purposes. So what is using and what is misusing? When you turn loose this kind of power on the world, and it becomes available to anyone who can access it…
The three dots at the close of the above sentence are there on purpose. There are no answers to these questions. There is not even an answer to the most important question of them all: what is right and what is wrong when it comes to AI? It’s not built into the systems themselves, because that would mean giving them a sense of morality, which machines do not have. I guess you could build morals into one of the AI systems, but whose morals? We know that the people who have developed and built and own AI companies think that making money and getting rich is something they can all agree on, because that’s what they’re doing. They live in nice houses, they drive great cars, they fly around on Gulfstreams, they give themselves zillion dollar weddings and own massive yachts. But could they agree on such a basic moral as “killing bad, helping others good?”
The beast is in the wild now. The AI systems are out there for anyone to use. Even the Houthi rebels are tapping in and using AI to program and fire a rocket. How long until someone uses AI to develop a nuclear or biological weapon?
Ooops. A young AI engineer from Anthropic already warned us about that last week, before the AI CEOs galloped like crazy trying to get ahead of each other to look like they suddenly give a shit about what they’re doing. But all they did was copy what the young engineer said – it’s happening too fast, and it’s got too much power which is making it too dangerous.
The world has been going in this direction for decades. Humans have been working to make things easier and easier for as long as there have been humans, I guess. Stone age humans developed stone tools and moved on to metal hatchets and knives that were better at cutting and chopping vegetables or meat or each other. Now, in the morning, right beneath this desk at which I sit, a robot vacuum whirrs and makes its rounds picking up cat hair and dust so Tracy and I don’t have to, and I don’t have to tell it a thing. The robot does it on its own, and then it returns to the laundry room and its base vacuums up all the hair and dust into a paper bag. The only human involvement by me or Tracy is to remove the bag and insert a new one.
Apparently, we’ve reached the point where we have made killing as easy as running a vacuum. Robots will do the killing for you. AI will even design the killing machines and pick targets and guide them for you.
Soon – I guess it’s already happening – everyone will be able to do this. That’s what mutually assured destruction has been about. The day that everyone gets the bomb, or so the thinking has gone, nobody will want to use it because everybody else will use their bombs and everyone will be destroyed.
We’re there, folks, and not just nuclear weapons. When everyone can destroy each other by deploying made up social media AI shit depicting one another as thieves or sex criminals or murderers, what’s the use? Nobody will believe any of it if it’s all taken to be fake.
Does any of it matter? When it comes to weapons, it does, because the end result is dead people, and presumably, nobody wants to be dead.
But notice the “body” in “nobody.” That’s a human being. AI isn’t human. What do AI systems think? Will AI come to the same conclusion that humans came did with nuclear weapons about mutual assured destruction? Will AI systems save themselves from each other? Do machines have a soul?
We’re either going to figure out how to give them a soul, or they’re going to take ours.
