- In the wake of the Cambridge Analytica scandal we asked an ex-Facebook engineer how tech can be used to prevent internet sites like Facebook from being abused.
- Pete Hunt is the CEO founder of a startup called Smyte which was formed to tackle exactly this problem.
- He says that finding bad actors is relatively easy by searching for signals. The hard part is to not accidentally catch the good guys “in the net.”
In the wake of Mark Zuckerberg’s exhaustive Congressional testimony over the Cambridge Analytica scandal, we caught up with former Facebook engineer Pete Hunt, the CEO founder of a startup called Smyte.
We asked him how technology can be used to find bad actors online, such as Russian hackers and trolls intent on influencing elections, and prevent them from using internet sites to spread misinformation or do other misdeeds.
Smyte was formed in 2015 to tackle exactly this sort of problem.
It was founded by two former Facebook engineers (Hunt and his cofounder Josh Yudaken) and an ex Google engineer (co-founder Julian Tempelsman). The company uses technology to find the abusers so they can be booted off of internet sites such as Facebook, Twitter, YouTube and Reddit.
Smyte can also be used to protect corporate websites and apps from being hacked by bad guys using trickery like spam, manipulating customer support agents and so on.
“Technology can enable lots of harm if you don’t think about abuse,” Hunt said.
In his recent testimony on Capitol Hill, Zuckerberg talked about hiring more people to monitor content and help it police its website.
But even when Facebook builds that team out to 20,000 people, that won’t be enough. Facebook’s 2 billion users upload 100 billion bits of content like links, status check-ins and photos every day. Humans just can’t watch all that stuff. So Facebook uses monitoring technology to be its eyes and ears and its working on making that tech smarter using artificial intelligence, Zuckerberg said.
We asked Hunt how technology can be used to spot the bad actors, fake news, malicious links and the like.
“The thing about abuse and these kind of adversaries in general is that it’s a business just like any other business. They have a sales funnel just like we have a sales funnel: you start with a list of prospects who seem like high-value targets,” Hunt explained.
“And the way you do that in a technology age, is you use all these great machine learning technologies and search and crawling technologies that can be used for good, and you can can use them to identify the most vulnerable and highest-value targets,” he said.
For instance, Cambridge Analytica gathered data on 87 million Facebook users to determine their beliefs and leanings, then fed them misinformation to influence their opinions and ultimately their actions.
How to stop it
Smyte says its team has figured out how to stop these bad guys by looking at four specific types of “signals.”
1. Content signals. Does a post or ad have a photo and if so, it uses machine learning to determine what the photo is, such as a photo of a political figure like Hillary Clinton or Donald Trump.
2. Behavior signals. For instance, when someone signs up for a new account and instantly copies and pastes a description about themselves in, as opposed to taking time and care.
3. Reputational signals. Is the IP address coming from say, the Ukraine. If so, why is someone from the Ukraine buying ads on US political figures or issues?
4. Relationship signals. Who is the account connected to and are there clusters of suspicious behavior among them?
After content is flagged, a company can’t just boot the person off. Bad behavior isn’t black and white and the tech will make mistakes in the gray areas.
“It’s really easy to catch bad guys, if you don’t care about the good guys who get caught in the net,” Hunt says.
In fact, Zuckerberg was grilled by several conservative politicians over instances where content was banned, prowling for evidence that internet companies are censoring conservative viewpoints.
But it’s more accurate to say that mistakes are just a numbers game, Hunt point out.
With 2 billion users, even if Facebook’s monitoring tech was 99% accurate, that’s still 2 million people that get impacted. And there’s a network effect, of all those people’s connections learning about the mistake, spreading the impact to millions more.
Rather than cutting people off, Hunt advocates putting suspicious players in a gray-area where they are monitored for more bad behavior, warned and educated, before they are cut off.
Obviously, his hope, and the hope of his startup, is that this Facebook scandal causes more companies to start taking abuse more seriously.
He imagines a day when companies have a VP of Anti-abuse job in-house and in which the focus shifts “from security that’s focused on protecting the company’s assets” to security that can “stand up and protect the users.”