Human on the Loop: Hrytseniuk on Ukraine's Autonomous Drones

Human on the loop means the AI does the flying and a person only monitors, helps, and confirms or cancels its wrong decisions, says Andrii Hrytseniuk, then Brave1's CEO. Ukraine's drones have already started that stage, he says, and he is "definitely sure" that next year the majority of hits will be done in fully automatic mode.

Hrytseniuk led Brave1, Ukraine's state defence innovation platform, which tests and funds defence products for the front. That makes his account of autonomy a description of a programme, not a forecast from outside it. It also carries a warning: Russia, he says, has already removed the human too early, and its automated drones are hitting the wrong targets.

What does human on the loop mean for Ukraine's drones?

Hrytseniuk describes three stages, and places Ukraine between the second and the third.

• Manually operated drones. The pilot does everything.
• Human in the loop. Some activities, mostly the routine ones, are automated by artificial intelligence, and the AI advises the soldier to take better decisions. Ukraine, he says, has successfully moved to this level.
• Human on the loop. The next level, which he says Ukraine has started. "There is no need to be an extremely effective and skilled pilot. You can just help the AI to be effective and skilled pilot."

His reasons are specific. Computer vision sees better than a human. Artificial intelligence takes decisions faster than a human. Where there is a big amount of data, a computer works better. The skills, as he puts it, "are already transferred to AI."

What is left for the person is to monitor, help, and have the possibility to cancel the AI's wrong decisions. He keeps the reason for that last role in plain view: "because AI is not perfect."

Behind all of it is his reading of the war itself. More and more, he says, it is not human against human, or human against drone, but drone against drone, and wherever technologies fight technologies, automation is needed.

Hear Hrytseniuk define human on the loop at 30:09

Will most drone strikes be fully automatic next year?

Hrytseniuk thinks so. He believes that next year, on both sides and across the defence industry, fully automated drones will be in wide use. "It's scary when drones are taking decisions themselves, but this is the new reality."

Samuel Cook asked him to spell out what that means: a drone that goes up, identifies targets on its own and engages them without a human in the loop. Hrytseniuk's answer was the strongest claim he made: "I'm definitely sure that next year, the majority of hits will be done in fully automatic mode." And if Ukraine does not implement it, he added, "we will be late in this run."

His prescription is legal rather than technical. Legislation, he says, has to be adopted that acknowledges this cannot be avoided, and that drones will be fighting drones in fully automatic mode. The law, in his framing, catches up with the stage that is coming. It does not decide whether that stage comes.

What happens when the human is removed too early?

Rob Lee raised a report of a Russian Molniya that used AI for a strike, and asked whether Russia was doing this a lot. Hrytseniuk confirmed it, and went further.

Russia, he says, is running a lot of experiments with fully automated drones even when it does not have good enough AI technologies and decision-making solutions. To do it, Russia has started "to close eyes on the rules that a human should confirm the target."

The result, in his words: "we have a lot of cases when fully automated drones from Russia are identifying wrong targets and destroying civil objects." He gives the cause as "too early implementation of AI." Not a targeting policy, in his account, but a system fielded before its judgement was good enough, with the person who would have caught the error removed.

Ukraine's line is the one that separates the two stages. "We are much more respectful to legislation and we are using human to avoid such mistakes by AI." Read against his own stages, that is the difference between human on the loop, where a person can still cancel, and the fully automatic mode Russia is already experimenting with.

There is a tension in his position, and he does not hide it. He criticises Russia for automating early, and he predicts that most hits will be fully automatic within a year. What separates the two, in his account, is whether the AI is ready, not whether the human should eventually step back.

How is Ukraine answering Russia's automated drones?

Russia playing with fully automated drones is, Hrytseniuk says, "the reality where we need to find countermeasures", and Ukraine has already implemented some countermeasures for such technologies.

The longer answer is about partners. To have parity, or more advanced technologies, Ukraine has to work with allied countries and foreign companies on solutions that can take effective decisions in a fully automated mode. The head of Ukraine's defence innovation platform was, in effect, asking allies to build the same category of system together.

Speed is the other half. Among his lessons for Western militaries, he says four years of war have shown that new technologies work effectively for the first three to six months. Deploy faster than the enemy and you hold a temporary advantage. If the enemy implements something faster, you have to be faster still at inventing countermeasures.

Automation is also arriving on the ground. Hrytseniuk calls automatically managed turrets, such as Sky Sentinel, which Brave1 funded at a very early stage, a high priority, because "it's impossible to scale people and scale soldiers." A turret is, in his words, "maybe not perfect, but good enough substitution of soldiers on some level." Artificial intelligence is already used in Ukraine's ground robots too, as set out in what Ukraine's ground robots do.

The depth these drones now have to cover, and why it breaks older radio links, is in how deep Ukraine's gray zone runs.