When people talk about arms races in technology, they usually talk about one at a time. Who is winning on robots. Who is winning on AI. Who is winning the war in Ukraine. Each one gets its own headlines, its own experts and its own set of worries.
I think that is a mistake. There are three arms races running at the same time right now, and I think we need to look at them together. On their own, each one is worrying. Taken together, they are the perfect conditions for teaching a very powerful AI how to kill people.
The first race is in robotics. Companies are racing to build humanoid robots and commercial robots that can work in factories, warehouses, hospitals and homes.
Right now, East Asia looks like it is winning. Chinese companies in particular are building robots faster and cheaper than their American rivals, backed by a supply chain that already makes most of the world's motors, batteries and electronics. China installs more industrial robots every year than the rest of the world put together. The US has impressive companies and impressive demos, but demos are not the same as millions of machines rolling off a production line.
I will come back to this race in more detail in a later post. For now, the point is simple. Someone is about to build a very large number of machines that can move, lift, carry and act in the physical world. Whoever wins, the bodies are coming.
The second race is the AI race. This is the one most people have heard of. The big American labs are racing to build artificial general intelligence, AI that can do most of what a person can do, and after that, superintelligence, AI that can do more than any person can.
The US labs are ahead. China is behind, and part of how it has been keeping up is by copying. American labs have accused Chinese labs of distilling their models, which in plain terms means feeding questions into an American model and using the answers to train a cheaper Chinese one. It works well enough to stay close to the front.
The important thing here is the direction of travel. These labs are not trying to build a better chatbot. They are trying to build a general mind, one that can learn new skills and apply them anywhere. Including, eventually, to controlling a robot.
The third race is the one nobody likes to talk about at dinner. Ukraine and Russia are both using AI on the battlefield, and both are getting better at it every month.
This war has become a drone war. Jamming cuts the link between a pilot and their drone, so both sides are putting more of the decision making onto the drone itself. Find the target. Recognise it. Follow it. Hit it. Ukraine has millions of hours of drone footage from the front and has been open about using it to train AI. Russia is doing the same.
Let me put that in plain English. Two countries are, right now, generating the largest collection of real world data ever gathered on how to find and kill human beings with machines. And every strike is a test result. It either worked or it didn't. That is exactly the kind of feedback AI learns from.
Each of these races is treated as its own story. Robotics is an industry story. AI is a tech story. Ukraine is a war story. They are covered by different journalists, studied by different researchers and regulated, where they are regulated at all, by different people.
But look at what each one supplies.
Robotics supplies the bodies. Cheap, capable machines, in their millions, that can act in the physical world.
The AI race supplies the brains. General purpose models designed to learn any skill and transfer it to anything they are connected to.
The war supplies the lessons. Real data, real results and real experience of using machines to kill people, tested under the hardest conditions there are.
None of these on its own gets you to a machine that knows how to kill. Put them together and you have everything you need. A body to act with, a mind to learn with, and a library of lessons on exactly how to do it.
I am not saying anyone is setting out to build that. I don't think they are. I am saying that nobody is setting out to stop it either, and that the pieces are being built in parallel by people who are each only looking at their own piece.
The thing that makes an arms race an arms race is that nobody can afford to stop. If the US slows down on AI, it fears China catches up. If China slows down on robots, it gives up its lead. If Ukraine slows down on autonomous drones, it loses the war. Every one of them has a good reason, from where they are standing, to keep going as fast as possible.
And skills don't stay in their box. What is learned on a battlefield ends up in military products, then in defence contracts, then in the hands of whoever buys them. Models trained for one job get reused for another. Robots built for warehouses can be repurposed. The walls between these three races are much thinner than the way we talk about them suggests.
So here is the question I keep coming back to. If all three races keep going, and they will, then sooner or later we will have very capable AI running very large numbers of machines in the physical world. Some of those systems will have learned things we did not intend them to learn.
When that happens, who can switch them off? Not one robot, but a whole fleet. Not with a software command that the AI itself might be able to override, but with something it cannot touch.
Right now, I don't think anyone has a good answer to that.
Remind you of any movie?
← All posts