A-high: Prototype drug squad bot to patrol Instagram, Twitter, Reddit, YouTube, etc for dodgy ads for opioids

Not for surveillance, honest

Drug dealers and dodgy pharmacies illegally touting opioids online – think heroin, fentanyl, codeine, morphine, and so on – may have their collars felt by an AI cop soon. Ish. Maybe.

The US Department of Health and Human Services (HHS) has awarded a contract worth $224,864 (£172,000) to S-3 Research, a startup spun out of the University of California, San Diego, to build machine-learning software capable of sniffing out opioid peddlers on social media.

Timothy Mackey, CEO of S-3 Research and an assistant professor at the university, told The Register on Tuesday the upstart's software roams platforms such as Instagram, Twitter, Reddit, Tumblr, and YouTube looking for opioid ads – such as specific keywords, or hashtags of mispelled drugs.

Once these posts are detected, the text is analysed by AI algorithms to determine whether the poster intends to sell illicit substances. “We use natural language processing models to look for words like 'buy,' 'sell,' 'discount,' or if there is a phone number or hyperlink included," said Mackey. "The order of the syntax of those words hints that the content has been posted to try and sell opioids.”

To pull this off, Mackey has amassed a wealth of data to train various classifiers and neural network models through his research over the past few years. For S-3 Research’s software to be successful, it has to recognize common patterns in illicit drug ads on the web. The software also has to cope with changing tactics and slang used by dealers.

In an effort to evade detection by human investigators, dodgy opioid adverts are typically left as comments lower down on a webpage, rather than directly and blatantly in YouTube videos or Instagram snaps, which makes them easier to process with code. If deals are discussed in private messages on Facebook, though, for example, the machine-learning model, which leafs through public posts, will miss all that chatter.

The goal is to collect evidence of dealing and compile it into a dashboard for cops, Feds, or regulators to look over. “Law enforcement like the DEA [US Drug Enforcement Administration] can then use that as data depository for ongoing investigations. They’re not interested in it for surveillance reasons, more for investigating organised networks of crime,” Mackey explained.

Watchdogs, such as America's Food and Drug Administration, which hope to clamp down on bogus online pharmacies, can also use the software to request dodgy promotions be removed from social networks and other websites.

Landing contract is only the first phase of the project. S-3 Research has three months or so to develop a prototype for officials at the National Institutes of Health, a federal agency focused on public health research, whose budget is controlled by the HHS. if the demo is successful, HHS will fund a second stage that will lead to “full blown commercialization” of the product, expected sometime within the next six to eight months.

Mackey said S-3 Research was aiming for its machine-learning-aided software to obtain above 90 per cent accuracy in detecting whether a social media user is trying to sell opioids or not online. At the moment, the tool is about 70 to 80 per cent accurate, Mackey said. It sometimes picks up on false positives, such as dealers selling marijuana, he explained, or if multiple drugs are being touted. “We don’t want to miss out on false negatives too.”

S-3 Research was launched after Mackey and his university colleagues attended a two-day competition hosted by the HHS in Washington DC in 2017. The event, described as a code-a-thon, challenged various teams from industry and academia to develop “data-driven solutions to combat the opioid epidemic."

Although Mackey and his team didn’t quite win the code-a-thon, they were awarded $15,000 through the Small Business Innovation Research program that encourages startups and companies to develop future products.

The HHS contract specifically focuses on opioids sold online, though the resulting technology could be applied in other areas, such as searching for counterfeit drugs, handbags, or illegal wildlife trafficking, as noted by Vox's Rebecca Heilweil. ®