Army of Twitter robots could unveil answers into Russian hacking

BERKELEY, Calif. (KGO) -- As questions swirl on Capitol Hill over Russian involvement in the U.S. Presidential election, two U.C. Berkeley students are proposing an answer to one part of that puzzle -- an answer that has them getting attention from all corners the Internet.

Ash Bhat and Rohan Phadte call their venture RoBhat labs -- the latest collaboration of a pair that's known each other since middle school. Now, as college roommates, they're taking on what they say the executives of social networks like Twitter won't: an army of "bots" tweeting and retweeting political propaganda in efforts to sway public opinion.

"This is a huge problem, and we need to try to find a solution," Phadte said.

The two say they first discovered "propaganda bots" after building a machine learning tool that analyzes news articles posted on Facebook for truth and bias. When attempting to build a similar tool for Twitter, they came across a problem: many of the articles were being posted and shared by users who behaved strangely.

"These accounts don't really exhibit very human behavior," explained Phadte. "They're constantly retweeting, they're interacting with each other."

After protests got out of hand on campus during a planned visit by conservative speaker Milo Yiannopoulos, Bhat and Phadte decided to take a closer look at the problem. They began building a machine learning model that measures over a hundred factors to determine the likelihood that a Twitter account is fake, and run by an algorithm that spews propaganda. Many bots are following thousands of users -- many of them other bots -- and tweet at all hours of the day and night, the two explained.

The resulting computer program, Botcheck.me, is available as a free Chrome browser plug-in. It adds a button to every tweet and Twitter account that instantly displays whether a user is likely to be a bot -- or not. In the 7 percent of cases where the app gets it wrong, a button to "disagree" with its prediction allows Bhat and Phadte to feed the error back into the machine learning model and make it better.

Botcheck.me had no problem identifying a user dubbed "Tina 4 Trump" as a likely robot. Bhat read off some of the account's recent tweets: "Talking about Bernie and Hillary, retweets Donald Trump... talking about how they're boycotting the NFL."

Bhat says it's possible Tina is a real person whose Twitter account was hacked. During their research, the two found at least one case in which a user's account began spewing propaganda after tweets that read, "Someone has hacked my Twitter account," and, "I didn't post this -- what is happening to my Twitter?"

But the pair says the most troubling discovery they've made is about how these bots behave in swarms -- an army of them, often retweeting the same fake news, in a coordinated attack on public opinion. In many cases, the original post comes from an account that impersonates a real politician, they said. Those accounts are listed as "parody accounts," a type of speech that's protected by the First Amendment, but users may not realize that.

Bhat said he's read replies by users who appear to be "actually believing that this account is the real (Vice President) Mike Pence."

Bhat and Phadte say the response from users and journalists has been overwhelming in the few days since Botcheck.me went live. But they say they wish their tool weren't necessary.

"It's not the responsibility of the users," said Bhat. "It should be the responsibility of Twitter."

Twitter did not respond to our requests for comment on this story.

Visit RoBhat's website and view live statistics on what propaganda bots are tweeting about.

Click here for more stories, photos, and video on Twitter.