How an AI company parsed misinformation early in Israel-Hamas war

Colin Demarest

Source Link

WASHINGTON — When Hamas militants raided Israel in early October, killing and abducting more than 1,000 people, videos, images and text flooded social media. Rumors and shoddy information proliferated, blurring the line between fact and fiction.

Artificial intelligence and data analysis firm Primer monitored the situation from afar using its Command software. It demonstrated its AI-enabled parsing capabilities at the Association of the U.S. Army’s annual convention in Washington days later, promising to identify kernels of truth among the chaos in the Middle East.

“Just aggregating lots of data, particularly if it’s a really noisy environment and the facts have yet to be established, can be really problematic because you’re just making a big pile for the user to go through,” Primer CEO Sean Moriarty told C4ISRNET on the show floor. “As you might imagine, data is all over the place. There’s all sorts of open-source intelligence data. The question is: What can a professional do with it, using their knowledge and experience? And that comes down to speed, power and accuracy.”

The Command software is designed with the single-pane-of-glass motif in mind. The program takes queries from users, much like a Google search; pulls vast amounts of data, namely social media feeds and news articles; and populates the results with summaries, context and name-entity recognition. It extracts people, places and things of note, handles translations, and presents sources that explain the process like math homework.

In a demo at the AUSA event, the software sorted through information related to the Israel-Hamas war and then produced a continuously refreshed timeline of events. Some of the points were geolocated, generating a heat map of posts and interactions.

“What it’s actually doing is interrogating these disparate sources, identifying anomalies where information is in conflict, and scoring it,” said Moriarty, who previously led Ticketmaster. “Anywhere there is a hot spot, our folks are looking to see what signal we can pull.”

The Command software is tailored for defense applications, according to the company, whose advisers include a former principal deputy director of national intelligence and former leaders of U.S. Special Operations Command and Africa Command.

Primer in June announced a $69 million funding round to help accelerate product delivery to government and commercial customers.

Mark Brunner, the president of the company’s federal team, said the goal of Command is to reduce “the latency between sensor and shooter.” That language is often used to describe the Pentagon’s Combined Joint All-Domain Command and Control effort, or CJADC2, which seeks to seamlessly link forces across land, air, sea, space and cyberspace.

“Sense, make sense, decide and act — we’re the ‘make sense’ people in that loop,” Brunner said. “If you’re a customer and you’re on the watch floor at SOCOM or the Army intel fusion center, our platform gives you the ability to ingest not only [open-source intelligence] from these various sources, but in over 100 languages.”

Military analysts and others in the intelligence community must sift through sometimes overwhelming volumes of information. Some of it can be correct, some of it can be incorrect and some can be purposely deceptive.

The U.S. Defense Department is spending money on AI and machine learning to augment such workloads and pick out patterns that may have been previously missed.

“What we have, though our product here, is the ability to ingest vast amounts of data, run it through our GPUs, and actually summarize and contextualize that data,” said Chris Lacy, who led the Command demonstration. “The key things an analyst is immediately going to look for — it is categorizing that and showing that [in] real time to the analyst as soon as they walk in the door.”

“As soon as the breaking news happened,” he added, “I put in a monitor and just started tracking it.”

The Defense Department sought $1.4 billion for artificial intelligence in fiscal 2024, which began Oct. 1. A continuing resolution that maintains funding rates from the prior year is in place until mid-November.