Military Scientists Harness AI To Fight Synthetic Opioids

TAMPA — A Defense Intelligence Agency team is using artificial intelligence to map the shadowy production-and-distribution networks of synthetic drugs that kill most of the 47,000 Americans who die of opioid overdoses each year — and in the process showing how military and law enforcement will put AI to work.

Tracking fentanyl and other synthetic drugs, which can be made almost anywhere, is harder than tracking cocaine, which is processed by relatively few cartels in relatively few places in South America, said Brian Drake, DIA's director of artificial intelligence for future capabilities and innovations.

“We can estimate those [cocaine] production numbers because we have lots of data that comes over a long period of time and there's only so much land mass that is arable for cocaine,” Drake told attendees at the annual DODIIS conference here on Monday.  But geography imposes few limits on the production of synthetic drugs. Further confusing matters, illegal producers often masquerade as legal ones.“It can be produced in any quantity you need. So how do you estimate a drug marketplace where you don't have the primary mathematics in order to estimate that?”

Drake’s answer: massive correlational analysis. His program, called SABLE SPEAR, ingests data from 43 million websites, eight million of the cargo receipts dubbed “bills of lading,” a billion satellite images, and about 20 other data sources. Algorithms help researchers to draw connections that would otherwise be invisible. 

For instance, those 43 million websites? They contain clues about the businesses that import materials to make the drugs and the method of sales. “We...extract things like emails, phone numbers, physical addresses, business records. From that information, we then correlate that against those [satellite] images and then say, `Yep, that's a place on the planet that produces fentanyl. We know that because open sources tell us that that place exists’.”

Related: The Pentagon’s First AI Strategy Will Focus on Near-Term Operations — and Safety

Related: Trump Declares National Emergency, Calling Gov’t Border Drug Stats ‘Lies’

Related: DEA Never Checked If Its Massive Surveillance Operations Are Legal, Watchdog Says

When Drake’s team runs across a business address that a cross-check with satellite imagery reveals to be a vacant field, that’s what he calls a “strong indicator of denial and deception.”

“If you're a business that's trying to sell a pharmaceutical good, a quite lucrative one by the way, then I want to be found. I want to sell it. I want to make that happen. If I'm doing it illegally, I don't want to be found at all. And I'm going to ingest denial and deception into any dataset where that might be.”

SABLE SPEAR absorbs all that data, finds anomalies between, say, business records and satellite photos, and then outputs a tip that Drake and his team can share with law enforcement partners such as the FBI, DHS, etc. 

By harnessing AI, Drake’s team can look for clues with a comprehensiveness and detail that is far beyond most law enforcement operations. For example, SABLE SPEAR can scrutinize the code behind suspect manufacturers’ websites for clues about their legitimacy. 

“If you're a web developer and you want to create five or six different sites for five or six different customers, do you recreate each of them? Turns out that web developers around the world are generally pretty lazy,” he said. “So if I find any illegal website that's been coded in a particular way…it will be unique because that web developer — maybe they used an open source tool — but it’s the combination of what they're using to create that site that will be unique. That provides us with those ‘tool marks’ to say, ‘You know what, you think these five sites are five separate sites? They're actually all part of the same nefarious network’.”

The program’s AI also yields insight into the people in those networks, which can help prosecutors make charges stick. The process might show that a fentanyl trafficker is also manufacturing passports, committing fraud, or engaging in other crimes like child pornography. 

The amount of apparently illegal activity they’ve uncovered is literally overwhelming law enforcement partners. “We're turning out so many targets we can't action it. Cannot action,” he said. 

But it’s information that law enforcement can use to go after an opioid network in a more strategic way, prioritizing the leaders at the top. 

Drake launched the program barely a year ago. About four months ago, his team hired private contractors to scale it up. He describes their success rate since then as exponential.  

“We had a 100% increase in the number of companies that we identified. We had a 400% increase in the number of people we identify in those networks, and a 900% increase in the number of illicit activities identified,” he said.