Special Operators Want AI to Help Discern Public Opinion

TAMPA, Florida—U.S. special operators have long understood that military advantage depends on how the public perceives the players. In the next few years, they hope to develop AI tools that can measure those perceptions better than ever—perhaps even well enough to vault U.S. influence operations past the larger efforts of China and Russia.

“In the information space, I still don't think that we have all the tools that we need, and we need to…continue to develop that speed,” said Gen. Richard Clark, who leads U.S. Special Operations Command. 

“The one thing that we're working on very hard at is just the sentiment analysis capability,” said Clark, who spoke Monday at the NDIA SOFIC conference here. “I think if we can apply big data, along with artificial intelligence and machine learning, I think it will give our people that are working in this space an advantage…I think we've got to continue to look at the [legal] authorities that go with that.” 

Clark praised the Ukrainian government’s use of information operations to “bolster morale of the Ukrainian forces to ensure that the truth is getting out about what the Russian forces are doing.” 

But Ukraine’s information environment is an “easy” one “because everybody can see what is happening," he said. "How are we going to do this against the near-peer adversaries who may not be as open or the whole world may not be pointing or it's a slow boil where we need to go?”

Too often, he added, the U.S. government is reluctant to operate in the information space. 

Sentiment analysis isn’t new. Today’s practice derives from public-opinion polling following the Cold War. But the rise of information technology in the 1990s and early 2000s enabled computational sentiment analysis. A seminal 1990 paper showed how crunching large amounts of scanned text to count how frequently words appear with other words can generate a sense whether people are feeling good or bad about a given subject.

The practice has exploded in recent years as the spread of social media has created new and easily structured data sets that reflect public mood. More than 99% of the peer-reviewed papers on the subject have appeared since 2004.

James Smith, the head of acquisition for SOCOM, said special operators “need to understand what the environment is where they're operating. What is the sentiment? As the [commanding general] said, we need to be able to message and counter-message in that environment at a speed and at a scale” that matches the pace of adversaries.”

Various heads of theater special operations command reiterated that message in a later panel. 

Rear Adm. Milton Sands, who leads U.S. Special Operations Command Africa, said that the presence of Russian Wagner mercenaries in Africa has made it harder for the United States to operate in Mali. 

“That's largely because of information operations,” Sands said. “You see a connection between both what's happening for example in the UN, when, in the UN, the ambassador from Russia is talking about Mali's airspace ‘being violated’—basically a false claim.” 

That will likely lead the Malian government to ask Russia for some sort of air defenses, he said. 

“So it gets very, very interesting, very complex, and it's very active,” he said. But U.S. troops are working hard as well: “We'll have soldiers tweeting from the field in firefights in order to stay ahead of Al Shabaab.”

Rear Adm. Keith Davids, who leads U.S. Special Operations Command South, said, “I would say that we are getting better, but [are] still a little bit slow as an intergovernmental team” at identifying misinformation and disinformation. “I will tell you Russia is very active,” in disinformation operations in Latin and South America, he said. 

A tactical battlefield victory is moot if the public perception of it isn’t good, Davids said. So sentiment analysis will be essential to help special operators assess how well they are doing relative to China and Russia in places like South America. 

“I think there's some cutting-edge work being done with the use of sentiment analysis and AI to get at those things,” he said. 

One SOCOM official told Defense One on background that money had not yet been allocated to the effort and he expected the program to formally start in 2024.