The dark web, despite all the attention police have been giving it lately, is still the wild west of the internet. On the markets where people pseudonymously buy and sell drugs for Bitcoin, vendors can simply disappear overnight, along with everyone's money. It's called an "exit scam," and it happens regularly.
What if an algorithm could predict when a dark web drug dealer is about to make off with untold riches in Bitcoin? Purchasing illegal goods online—arguably kind of a dumb thing to do in light of the high-profile arrests that have gone down—would get a whole lot more intelligent, albeit artificially.
Reddit user "twofeetdown," who described herself* as "a graduate student at a large university in the bay area" in a message to me, is designing a program that can do just that as part of a university research project. According to twofeetdown, AI that can predict when a dealer is about to dip is necessary because "people lose a lot of money to scammers," she wrote, and current methods of alerting people to a scam are ineffective.
Usually, the news that a dark web vendor is scamming their customers comes in fragments: a Reddit post complaining about a missed delivery here, a poor review there. Often, however, nothing is confirmed until it's too late. It certainly doesn't help that misinformation and unwarranted freakouts are at least as common as actual scams, if not moreso.
"All those things are pretty detectable just by scraping the forums and market"
Still, scams tend to follow a particular pattern, twofeetdown said, and training an algorithm on a corpus of data would allow it to scrape the markets and find the telltale signs of a scam. For example, a vendor advertising a big sale and then demanding that buyers send them Bitcoin directly, bypassing the escrow services most markets use, could be a good indicator that they plan on disappearing soon.
"All those things are pretty detectable just by scraping the forums and market," twofeetdown wrote. "Of course, if a vendor just completely disappears, there's not much to go off. There's a lot of extra money to be made if you pump your sales before leaving, so most of the big scams have followed this pattern."
Twofeetdown told me that she's already written code to parse archives of the popular Agora market, collected by an independent dark web researcher called Gwern, who regularly scrapes the markets and releases the archives. After the research project is finished, she plans on making the code available as an open source repository so anyone can use it.
"My dream would be for markets to integrate stuff like this directly into their sites," she wrote, "but I doubt it's going to happen any time soon."
After all, if markets and users haven't learned their lesson by now, with potentially millions of dollars in Bitcoin stolen, they may never.
* twofeetdown would not reveal their identity or gender to me, citing concerns over their real name being associated with the markets. For the purposes of readability alone, twofeetdown is referred to as a woman in this article.