For an AI to understand “thirst,” you must first teach it about Drake


The thirst is real, but it can also be real confusing.
On Thursday (March 9), Facebook Messenger launched its “Messenger Day” feature, an overt Snapchat ripoff with one distinguishing characteristic: It simplifies sharing on-the-fly photos and videos with a custom group of people (versus Snapchat, which allows for public sharing, direct-message sharing, or both). In a guide to using Messenger Day for flirting, Mashable characterized it as “perfect for laying down thirst traps for crushes,” i.e. for sending provocative or attention-seeking photos to a bespoke group of would-be suitors, who wouldn’t know that they’re part of a group. (Don’t hate the player; hate the game.)
If the idea of sending anyone multiple daily photos covered in digital stickers doesn’t make you feel old, then the phrase ”thirst trap” may do the trick. Urban Dictionary defines a “thirst trap” as “any statement or picture used to intentionally create attention or ‘thirst,’ ” and defines “thirsty” as “desperate,” “very eager,” and “constantly looking for a person to date.” But thirsty these days goes beyond personal characteristic. One of Vogue’s most overused words of 2016, it has morphed into a multi-purpose adjective.
That got me thinking: How does one identify thirst? Better yet, could a computer? What if Messenger Day automatically decided which groups received your photos based on an AI’s understanding of what thirst looks like, combined with its knowledge of who should receive thirsty posts? More ambitious still: Could AI power a bot that tells you whether an image or word is associated with being thirsty? A Thirstbot™?
“If you had examples of images considered thirst traps, I don’t see a reason why you couldn’t train a machine-learning algorithm to recognize the markers of thirst,” Quartz AI reporter Dave Gershgorn told me when I put some of these very important questions to him. “An easy way to think about machine learning is as advanced pattern recognition.”
Dave also flagged to me some precedent for this idea: “In 2015, then-Stanford Ph.D candidate (and now OpenAI researcher) Andrej Karpathy built a machine that judged how ‘good’ selfies were. The data were public Instagram photos tagged #selfie, and the quality of the selfie was judged by the amount of likes it received. (Thirst could also arguably be measured by likes.) By refining the data shown to the algorithm, it would probably be relatively easy to use open-source code and make a Thirst Discriminator, discriminator being a CompSci word for something that makes a decision about a new piece of data.”
What about thirst as a more general concept, versus just a visual one?
Dave: “So here we get to AI’s big problem. Today, we’re really good at applying AI to specific domains, like pictures of people or pictures of cars or words. When all those are mixed together, the web of patterns that the AI has to make gets too messy and confusing. That’s why artificial general intelligence, or an AI that can understand a wide variety of problems and ideas, is considered the Holy Grail of AI research.”
Sounds challenging. Something that might help though (Dave: “Not really”) is a database of the “thirstiest” things. So for posterity and science, I’ve started that list here. (The results are based on the internet, the only true arbiter of thirst.)
Fiat Chrysler: “the thirstiest carmaker in the world”
(Related: Fiat Chrysler chief Sergio Marchionne is the “thirstiest man in cars.” “[Ask him] about anything, really just anything, at this point,” Jalopnik wrote in 2015, “and all he seems to want to talk about is how he can unload FCA onto someone else.” Incidentally, Fiat Chrysler is still looking for suitors.)
James Corden’s Carpool Karaoke bit at the Grammys: “the thirstiest of awards-show stunts”
Verizon: “the thirstiest telecom company in the history of business”
A PETA press release encouraging “solidarity” with “sexually exploited cows and hens” and a dog food commercial about “woman’s best friends”: “the thirstiest International Women’s Day tie-ins”
Color Me Badd‘s ”I Wanna Sex You Up”: one of the “thirstiest songs ever”
(Related: GQ named ”Santa Baby” the thirstiest Christmas song, asking: “Has anyone ever run the numbers on how many inflation-adjusted dollars the ‘Santa Baby’ woman’s proposed Christmas haul is worth?”)
Barack Obama: “the thirstiest president ever”
A direct message on Twitter: “the thirstiest way to meet someone”
Winter: “the thirstiest time of year”
(Related: August is the thirstiest month of the year. “The summer is almost over, the temperature keeps rising, and swimmer abs have us feeling some type of way.”)
US president Donald Trump or New Jersey senator Cory Booker: the “thirstiest man in US politics”
And then there’s Drake. The thirstiest rapper in the game. The thirstiest man in music. The thirstiest drama queen in Hollywood. The celebrity with the most thirst-inducing beard and the thirstiest fans (Justin Bieber the thirstiest among them). “Drake embodies the rom-com trope of the girl who suffocates her man by taking him engagement ring shopping on the second date,” Amy Zimmerman wrote in The Daily Beast. “Like a particularly smooth sofa, Drake is his best self when supporting women and letting them use him.”
If thirst follows Drake like a shadow, perhaps the best way to teach an AI about the concept is by pairing them up. Thirstbot, meet Drizzy. (Science, you’re welcome.)