Counter-drone expert Johannes Pinl: “We are entering an age in which machine fights machine”

MARSS develops command-and-control systems for various purposes. By its own account, it is currently the market leader in counter-drone solutions. Based on AI, MARSS’s command-and-control solution automates the “kill chain”, from detection through classification to the engagement of a drone.

The company was founded 20 years ago by Austrian Johannes Pinl and was recently acquired by the Australian company Electro Optic Systems (EOS). Johannes Pinl was interviewed at the Eurosatory defence exhibition in Paris about the future role of the human in warfare.

Mr Pinl, where are your systems in use?
Above all in the Middle East at the moment. The last drone, hopefully the last drone, that came from Iran towards Saudi Arabia was neutralised by our system. A lot is happening there, of course.

How high is the degree of automation?
There are two answers to that. First: what is technically possible? Second: what can really be done, also with regard to the Geneva Conventions, among other things? Technically, the big picture is that the drone threat is becoming ever more automated. On the battlefield we are seeing more and more attacks by first-person view (FPV) drones with automatic control, automatic target recognition and automatic engagement. We build defence systems and protect critical infrastructure. To do that, you have to use the same technology in order to stand a chance against such automated machines. We are entering an age in which machine really fights machine. We are just at the beginning of this new phase.

What does that look like in practice?
Everything is automated as far as possible. We still have the human in the loop. The system detects a drone with radar systems. AI automatically recognises how this object is moving and classifies it as a drone or a bird. If it is a bird, it is filtered out and the operator is not alerted. If it is a drone, a double verification follows with infrared cameras at a distance of 15 to 20 kilometres. There, too, AI comes into play again and checks once more whether it really is a drone. The system can then say whether it is, for example, a category 1 drone, such as a DJI drone, or a larger drone, such as a Shahed drone. Only then is the operator alerted.

How does the operator then act?
The operator receives a live video and then confirms that it actually is a drone. The system then proposes which countermeasure is most suitable. Normally this begins with jamming, that is, signal interference. But that no longer works very well in Ukraine and in Iran. In the Iran conflict in particular, the systems fly with Starlink or are so automated that no interference is possible during the flight. GPS jamming during the flight is often difficult as well.

What happens next?
The second layer is the kinetic solution. For this we use missiles, i.e. medium-range missile systems, such as the Thales LMM. That works very well in Ukraine and very well in the Middle East. But it is very expensive. The system only proposes something like that if it is a larger drone and a real threat. Otherwise, you always calculate the cost per kill. If a Shahed drone costs 30,000 to 40,000 dollars, you do not want to use a missile costing one million dollars. That would not be effective. Instead, you can also use a small APKWS rocket costing 25,000 to 30,000 dollars. Then you achieve cost parity. With this system we automate the kill chain for the operator as far as possible.

How does the operator then work in concrete terms?
The operator only has to say: yes, that really is a drone, and yes, I want to use the proposed effector — or no. That is the only input the operator needs to have. As a result, the operator can deal not just with one drone, but with several at the same time. That is exactly where artificial intelligence plays a real role.

What happens with a drone swarm?
The technology is so far advanced that it can classify which drone you are facing and whether it is, for example, a commercial aircraft that poses no threat. According to our tests, the operator can deal with four to five drones at the same time, no more. That is not yet a swarm attack. Because one speaks of a swarm when 30, 40, 50 or 100 drones approach at the same time. Humans cannot process that many threats at once. That is why AI comes into play and sets priorities for the operator: focus on these drones, which are coming in first and are the most dangerous.

And AI then takes over the engagement of the rest of the swarm?
Technically speaking, yes, to a certain extent. If the system knows when it is almost too late and the operator is becoming the bottleneck, AI could theoretically make the remaining decisions itself. At present, however, this is a purely technical question, because it is not permitted under the Geneva Conventions. The Western world in particular does not allow it. The other side does not abide by this and uses the technology. We do not. This puts us at a severe disadvantage, and that is a big question for the future.

Where is this development leading?
Our simulations clearly show that future combat will be between machines and machines. Technically, this will probably take about another five years. Attack is always easier; defence is always one step behind. A lot is already automated in attack; in defence we are about 80 per cent of the way there, but not yet 100 per cent. In five years it will probably really be the case that machine works against machine. And that will not only be in the air, but also on water, on land and under water. It will be multimedia, or rather multi-domain.

Are there nevertheless areas in which humans cannot be replaced so quickly?
Of course. We try to replace humans as far as possible, simply because humans are too slow for these tasks. The biggest problem for the machine is the final decision: is this really a drone, really a threat or not? For that you still need the human, the human eye. It could also be a Cessna that happens to look like a drone, has no transponder and is not registered. That is exactly where you still need the human in the loop.