‘Candid,’ the anonymous chat app, enforces civility with AI

If Twitter, 4Chan and Reddit are any indication, people will exploit even the barest of anonymization to be complete and utter asshats online. Just look at the recent attacks against Ghostbusters star Leslie Jones, or Reddit’s self-induced purge of racist, xenophobic and bigoted chat groups. Fully anonymized social services like Whisper or Yik Yak, where unidentifiable mobs can unceasingly bully other users, are no better. An app called Candid, which launches today, is designed to fight this unacceptable online behavior with artificial intelligence.

Developed by a pair of former Google project leads, Candid (the company) has developed a natural language processing system that analyzes every piece of posted content and flags inflammatory items for removal — things like hate speech, threats and slander. Off-topic posts are moved to more appropriate sections, so you won’t have to dig through political discussions to read about NASA’s mission to Mars.

Candid (the app) is available free on both iOS and Android. Similar to Yik Yak’s location-based function, each Candid user will see a personalized feed of posts and content based on their “education, employment, interests and neighborhood,” according to the company’s press release.

Interestingly, you can even sign up using your Facebook account to seed the app’s Group suggestions. That seems counterintuitive, though Candid’s stringent privacy rules account for this. All personally identifiable data — including IP address, precise location data and Facebook contacts — are encrypted with a one-way hash before they reach the company’s servers. That hash cannot be decrypted by Candid — or anyone else, for that matter. What’s more, closing or abandoning an account permanently deletes all that encrypted data from Candid’s servers.

Oddly, though, the app also requires users to provide their phone number during the initial account setup. The app even sends a two-step authentication code to confirm that the number is real. That data are encrypted like everything else Candid collects but it feels strange and out-of-place to request that right off the bat, especially from an app that sells itself on anonymity.

Once you get through the initial setup, Candid offers a number of mechanisms to maintain the user’s privacy. For example, the app applies a new, continually randomized username like “Curious Rabbit” or “Creative Lemur” to every new post. The app also gently coerces users to be polite by awarding various badges like “Explorer,” “Giver” or “Gossip” for positive posts, but will slap a “Hater” tag on people who are consistently negative.

This process, according to Candid’s PR team, is entirely automated. A “lot of factors contribute to getting the hater badge,” a rep told me, “including the number of negative comments and posts based on sentiment analysis, number of down votes a user gets and the number of posts from the user that were taken down. Posts that are taken down by the system are reviewed by a human.”

Additionally, Candid has a system in place to first identify potentially unsubstantiated rumors through its algorithmic AI, which are then verified by a person using web and Twitter results. Any rumors deemed to be false are quickly removed, while true statements — such as news leaks — remain. Similarly, if the system sees that a poster is threatening self-harm, it will issue a push notification to him or her with the number for a local crisis helpline.

So what do you get when an AI automatically scrubs your internet forum of all offensive content? Turns out, it’s banality. The test feeds that I created during my time using the service felt like a disembodied comments section, regardless of the groups that I subscribed to. While you can add external links to posts, very few of the 600-plus beta testers appear to do so. This leaves you reading strings of random, disconnected thoughts with very little context. You won’t find inflammatory content (kudos for that) but the discussions filling that void are far from riveting. Most posts echo the same shallow hot takes you’d find on YouTube or Reddit, just without the overt xenophobia and misogyny.

What’s more, there isn’t much actual discussion going on within these posts. Granted, that may be because there are only a few hundred beta testers, but most replies to posts resemble those in a YT comment section — people talk at each other, often in non sequiturs, rather than with each other. Or perhaps it’s like The New York Times comment section: There’s plenty of civility, sure, but it’s generally devoid of real interaction. Groupthink is also an issue once you delve beyond the broadest of groups. The overarching Politics Group offers a variety of thoughts and opinions but once you get into the Republican and Democrat groups or the Sanders, Clinton and Trump groups, views expressed within them become increasingly myopic.

At a more basic level, I can’t figure out the intrinsic value or benefit this app is really supposed to provide its users. Outside divulging state secrets or posting the details of a damning business deaI — which I am willing to bet nobody reading this post has ever been in the position to do — I fail to see why one would need to go to these anonymizing lengths. If anything, this app encourages disingenuous behavior. I could go on there and make threats against the president’s life if I felt like it, with virtually no reprisal outside having the system eventually flag and delete the post. I mean, if you feel you have to shield yourself behind multiple digital walls just to toss that gem of an opinion out onto the internet, write it down in a journal instead, scream it into a pillow — or maybe just keep it to yourself.