PicoZ

There is a sense of uneasiness when the screen lights up. Excitement, yes, because you’re being shown a new way to fight a war, having gained access to a perspective until now closed to human perception. But also modesty because the action is down below, a thousand miles below, and all the courage and suffering of the battle are so distant as to almost lose their human meaning. In a recent visit to Palantir’s offices in London, I was able to witness first-hand how the firm’s superior data technology really works. I have not been able to stop thinking about the experience ever since.

Palantir’s founding team, led by investor Peter Thiel and Alex Karp, wanted to create a company capable of using new data integration and data analytics technology — some of it developed to fight online payments fraud — to solve problems of law enforcement, national security, military tactics, and warfare. They called it Palantir, after the magical stones in The Lord of the Rings. Palantir, founded in 2003, developed its tools fighting terrorism after September 11, and has done extensive work for government agencies and corporations though much of its work is secret. It went public in 2020. But through its 20 years in business, the question has been just how capable are its systems and what could it achieve on a large scale conflict. Can it deliver in a war between large armies and with greater firepower?

The following scenario has been described to me by a Ukrainian commander rather than Palantir, which does not comment on operational details. As Ukraine launches several and potentially dozens of thrusts across the full frontline, Russia has to bring reserve forces out of hiding, which Ukraine can track in real time using satellite and drone imagery, as well as visual recognition algorithms, and target with unprecedented speed and precision. Before the start of the current offensive, and now in its early stages, Ukraine has spent weeks shaping the battlefield by destroying command and control centers, logistics, and munition depots deep behind enemy lines. In some cases it used the newly available and long range Storm Shadow missiles.

More from TIME

In my recent visit to the London office, Palantir’s largest and its main research hub outside the U.S., I was first shown the earth’s orbit. The screen on the wall comes alive with images of the thousands of satellites in low-earth orbit. Palantir’s MetaConstellation platform allows the user to task these satellites to answer a specific query. Imagine you want to know what is happening in a certain location and time in the Arctic. Click on a button and MetaConstelation will schedule the right combination of satellites to survey the designated area. Remarkably, the software can deploy algorithms at the source so that only those images where the algorithms find valuable information are downloaded, saving time. It’s a cat and mouse game: the same software can help you find a blind interval when no satellite is covering a particular area. Or perhaps you can fool your opponent into thinking they have found a blind interval by using stealth satellite technology or hacking their systems. Increasingly, the real battle will take place not on the ground and not in space but inside computer code,

The target coordination cycle: find, track, target, and prosecute. As we enter the algorithmic age, time is compressed. From the moment the algorithms set to work detecting their targets until these targets are prosecuted — a term of art in the field — no more than two or three minutes elapse. In the old world, the process might take six hours. As I look at the screen, it occurs to me that armies today are still operating in that world, following military doctrines that were tested and developed in a world where the cycle was measured in hours. Their equipment and movements are too large, too visible, too slow and ultimately too vulnerable. The U.S. military will face this problem if and when it has to fight a peer competitor such as China.

If a particular outcome falls short, the whole process is automatically corrected and improved, from target identification to effector pairing, a term referring to the selection of the appropriate weapons system for a given target and battlefield. Simultaneously, feedback from deployed models may then be used to refine, recalibrate, or develop new models to replace previously deployed ones. The platform is able to integrate data from multiple and disparate sources — think satellites, drones, and open-source intelligence — while allowing a new level of decentralised decision-making.

As one of Palantir’s U.K. engineers explained during my visit, this is “not a relevant domain for a human to exist in.” He was talking primarily of the correlation between signals intelligence and satellite imagery. The former is very good at telling you what something is, the latter can tell you where it is. The goal is to correlate the two, but no human being would be able to compare thousands of images with thousands of hours of intercepted communications. The software does it more or less instantaneously.

Some tranches of the targeting process can be considerably automated. Just as a deep learning algorithm knows how to recognise a picture of a dog after some hours of supervised learning, the Palantir algorithms can become extraordinarily apt at identifying an enemy command and control centre. Presumably, there are hundreds or thousands of indicators for such a target, which can be graded according to relevance. Machine learning algorithms are particularly suited for warfare because their very high number of variables, most invisible to the human eye, make them very difficult to deceive or evade. Nevertheless, the idea that the full cycle can be automated is pretty extreme and still very far from where the technology or doctrine are today. The Palantir strategists I spoke to disowned any interest in transforming their software from a tool into an autonomous agent.

Next step: target. Here Ukraine has made some remarkable breakthroughs on its own: the Kropyva software allows commanders to enter target coordinates into a tablet, and then the direction of firing and the distance to the target are calculated automatically. Kropyva was created back in 2014 after the first Russian invasion. The software is reminiscent of Uber in that it assigns targets to the nearest artillery battery or missile launcher.

The Palantir platform will consider the full range of weapons with the right capacities and range to prosecute the target. It will suggest the best available option, a suggestion that needs to be confirmed by the user. The last step is, of course, the battle damage assessment, or estimate of the outcome, whose results are fed back to the algorithm. That’s where the self-learning happens.

How much could the Palantir software achieve in a place like Bakhmut where street warfare has been dominant? In a way the question may miss the point. If Ukraine were able to deploy the full force of algorithmic warfare, Bakhmut might never become necessary. Could Russia have moved its large weapons systems to Bakhmut if Ukraine had from the beginning had access to all the components of a software system like Palantir’s? Even today it lacks some critical elements, such as ready access to the most classified intelligence, aviation or long-range missiles. Increasingly, the goal must be to assemble all the pieces of the target coordination cycle.

Alex Karp, Palantir’s CEO, has argued that “the power of advanced algorithmic warfare systems is now so great that it equates to having tactical nuclear weapons against an adversary with only conventional ones.” Palantir would not have invested billions in developing its system if it did not believe in its transformational power. Karp was the first executive of a major Western company to visit Kyiv and meet with Zelensky after the Russian invasion.

One idea I heard in London is that warfare may increasingly take place as a complex simulation within algorithmic systems. The process may have some deterrence powers: two opponents might reach the same conclusion about the outcome, preempting any need to trigger a conflict in the physical world. Is this utopian? Probably. The most likely scenario is an algorithmic arms race happening at superhuman speed. Here China rather than Russia is the real opponent. Taiwan rather than Ukraine is where the algorithm takes over.

Once two or more actors acquire precision-warfare capabilities, the battlefield is once again contested and new technologies might no longer favour the offence. Forward bases or assets used to scout and strike enemy positions become themselves vulnerable to attack over long distances. Chinese strategists increasingly see military conflict as “algorithmic warfare.” In this context, advantages of speed and persistence might only be accessible to fully autonomous systems, with artificial intelligence forming the core of the coming revolution in military affairs. In a speech on June 30, the chairman of the Joint Chiefs of Staff Mark Milley argued that rapidly advancing technology is causing the most significant fundamental change in the character of war ever recorded in history.

Ukraine might be the last large war fought primarily in the physical world. We should be getting ready for the moment when the physical and virtual worlds swap places, where everything happening in the former may well feel tangible and real only from the perspective of those unable to climb to the higher virtual plane. Virtual war is not a war between soldiers, tanks or airplanes, but a clash between algorithms. Here victory means the ability to build the basic rules determining how the world works. Inferior algorithms will simply operate according to the rules and outcomes set by more foundational software. Geopolitics used to mean the struggle to control the physical world. In the future it will be about the struggle to build a virtual one.

ncG1vNJzZmismaKyb6%2FOpmZvamlogHqEjqmYpZmeqbazecWuq66qlWK8p3nWmqmfmaKaera30Zqgp51f