In preparation for the rocket attacks on the Antonovski Bridge, a vital link between the occupied city of Kherson and the eastern bank of the Dnieper River, Ukrainian security officials closely studied a series of special reports. It was the summer of 2022 and Russia was heavily dependent on the bridge to resupply its troops west of the Dnieper. The reports contained investigations concerning two things: would the destruction of the bridge spread panic among the Russian soldiers or their families? And, most importantly, how could the Ukrainian government maximize this blow to morale by creating “a particular information environment”?
This is how Sviatoslav Gnizdovski, founder of the Open Minds Institute in Kyiv, describes the work his research team did to generate assessments with artificial intelligence (AI). Algorithms sifted through oceans of content from Russian social media and socioeconomic data on issues ranging from alcohol consumption and population movements to online searches and consumer behavior. The AI correlated all the changes with the evolution of the feelings of Russian “loyalists” and liberals regarding a possible plight of their country's soldiers.
That highly confidential work continues to shape important Ukrainian decisions about the course of the war, Gnizdovski says. And that includes possible future attacks on the Kerch Bridge, the only direct land link between Russia and Crimea.
Ukraine, outgunned by Russia, is increasingly seeking an edge with AI by employing the technology in a variety of ways. According to a Ukrainian colonel involved in weapons development, drone designers often consult ChatGPT as a “starting point” for developing engineering ideas; for example, about novel techniques to reduce vulnerability to Russian interference. Another military use of AI, says the colonel, who requested anonymity, is target identification.
As soldiers and military bloggers have become more cautious and thoughtful in their messaging, simple searches for clues to the location of forces have become less fruitful. However, processing huge amounts of images and text allows AI models to find potential clues, assemble them, and then surmise the likely location of a weapons system or formation of soldiers. That approach of “putting the pieces together” thanks to AI makes it possible for Molfar, an intelligence company with offices in Dnipro and Kyiv, to routinely find between two and five valuable targets a day, says Maksim Zrazhevski, an analyst at the company. The information is then immediately transmitted to the Ukrainian military, resulting in the destruction of some of the targets.
AI helps with targeting in other ways. SemanticForce, a Lithuanian company with offices in Kyiv and Ternopil, a city in western Ukraine, develops models that, in response to text instructions, analyze text and images online or uploaded. Many SemanticForce clients use the system for commercial purposes to track public sentiment about their brands. However, Molfar uses the model to map areas where Russian forces are likely to be low on morale and supplies, something that could make them an easier target. AI finds clues in images, including those from drone footage and soldiers complaining on social media.
It also gathers clues about Russian military weaknesses using a cloaking proxy. To do this, Molfar generates reports with SemanticForce AI on the activities of Russian volunteer groups that raise funds and prepare aid packages for the most needy sections of the front. The algorithms, Molfar says, do a good job of weeding out potentially misleading posts from bots. (Accounts with wild political swings give a good clue.) Company analysts sometimes augment that intelligence using software that masks the origin of a phone call, so Russian volunteer groups can be called by someone pretending to be calling. pass for a Russian eager to collaborate. Ten of the company's around 45 analysts work on targeting, and they do so pro-bono for Ukrainian forces.
Then there is counterintelligence. The use of AI helps those responsible for Ukrainian counterintelligence identify those whom Oleksi Danilov, until recently secretary of the National Security and Defense Council (NSDC), described as people “prone to betrayal.” Offers of money in exchange for sending geolocated images of military infrastructure and equipment often reach Ukrainian phones, says Dmitro Zolotukhin, former Ukrainian Deputy Minister of Information Policy. He himself recently received one of those messages. People who agree to participate in this “intelligence services market,” he adds, are regularly caught by the Security Service of Ukraine (SBU).
Ukrainian counterintelligence is dedicated to detecting illuminating links in disparate data sets with AI from Palantir, a US company. Imagine, for example, the case of a divorced and indebted person, at risk of losing his house and custody of his children, who opens a bank account abroad and whose phone has been detected near a place subsequently hit by missiles. In addition to connecting those “dots,” the AI performs “social network analysis.” If, for example, that hypothetical person has strong personal ties to Russia and has started receiving calls from someone whose phone use points to a higher social position, AI can increase their score on the risk scale.
For more than a decade, AI assessments of interactions between nodes in a network have offered impressive results. Kristian Gustafson, a former British intelligence agent who advised Afghanistan's Home Office in 2013, recounts the capture of a courier carrying wads of cash for Taliban bigwigs. His successive phone calls, he says, “illuminated the whole diagram.” Since then, algorithmic advances in calculating things like “betweenness centrality,” a measure of influence, have made those days seem, as another former intelligence officer puts it, “pretty primitive.”
In addition, network analysis helps Ukrainian investigators identify violators of sanctions imposed on Russia. By relating data from ship records to accounting records stored elsewhere, the software can “pierce the corporate veil,” says one source. According to Zolotukhin, hackers provide Ukrainian agencies with “tremendous” amounts of stolen business data. They are a great help in combating sanctions violations.
The use of AI has been developing for some time. The President of Ukraine Volodymyr Zelensky called in November 2019 for a massive boost to the use of this technology for national security. The result is a strategically oriented model, created and directed by the NSDC, that processes text, statistics, photos and videos. Called the Operations Center for Threat Assessment (COVA), it is fed by a wide range of information, some obtained by hackers, says Andrii Ziuz, chief of staff at the NSDC. The model tracks prices, phone use, population movements, trade, energy, politics, diplomacy and military events down to the weapons in the repair shops.
COVA operators call that model a “constructor,” because it also processes the results of smaller models, such as the Palantir and Delta software, which is combat software that supports the Ukrainian military's maneuver decisions. The COVA “overview” guides senior officials on sensitive issues such as mobilization policy, says Mikola Dobish, chief technologist at NSDC. Danilov comments that Zelensky has been informed of COVA assessments on more than 130 occasions, including one at 10 a.m. on the day of the Russian invasion. Access to some parts (or “circuits”) of COVA is also provided to other groups, such as insurers, foreign ministries, and the US Department of Energy.
The Ukrainian effort in AI benefits from its society's broad willingness to contribute data to the war effort. Citizens upload geotagged photos potentially relevant to the country's defense to a government application called Diia (“action,” in Ukrainian). Many companies provide Mantis Analytics, a Lviv company, with operational data on everything from late deliveries to call center activity and activation of burglar alarms. Among the recipients of the platform's assessments of the functioning of society are the Ministry of Defense and companies that seek to better deploy their own security resources.
It's not yet clear how important all of this will ultimately be. Evan Platt of Zero Line, a Kyiv NGO that provides equipment to soldiers and visits the front lines to study combat effectiveness, describes Ukraine's use of AI as a “positive point.” However, there are concerns. One is that enthusiasm for certain AI security applications diverts resources that would provide greater performance elsewhere. Too much faith in AI is another risk, and there is no doubt that some models on the market are overrated. And, more ominously, could AI prove negative for Ukraine's performance on the battlefield?
Some think so. One of them is John Arquilla, professor emeritus at the Naval Postgraduate School in California, author of influential books on war and advisor to Pentagon chiefs. Ukraine's greatest successes came early in the war, when decentralized networks of small units were encouraged to improvise. Today, Ukraine's AI “builder process,” he argues, is centralizing decision-making and snuffing out the creative sparks that arise “on the margins.” Your assessment is debatable. However, it helps to underline the importance of human judgment in the use of any technology.
© 2024 The Economist Newspaper Limited. All rights reserved
Translation: Juan Gabriel López Guix
