This Artificial Neural Network Generates Absurd Pickup Lines

Cheesy pickup lines are like puns: they usually elicit groans or eye-rolls, but people still can't resist making up new ones. So it may seem inevitable that in the many ongoing experiments with artificial intelligence, someone created a computer program that crafts pickup lines.

The results, Rob LeFebvre reports for Engadget, are adorable.

Research scientist Janelle Shane taught her neural network using pickup lines she scraped from the Internet. "Gathering the dataset was much more painful than I had expected," she writes on her blog. "I hadn't really read many of these before, and most were obscene or aggressive, or kind of insulting."

Fortunately, the network stayed fairly polite. She writes:

[A]lthough the neural network figured out the basic forms “You must be a …     because….” or “Hey baby, wanna…” it never learned to generate the worst lines—most of these were based on wordplay that it didn’t have a chance of reproducing.

Shane's pickup-line-slinging network is based on an open-source program on GitHub called char-rnn. Her network, and others like it, are computing models that mimic the way the brain works. That makes the network behave very differently than a conventional computer.

In everything from laptops to smartphones, the computer's central processor receives commands from a user, finds the necessary instructions in memory, decodes the instructions, performs an action and stores the results in memory. All these steps happen in order and each stage depends on what comes before.

Not so in a neural network (or more accurately, an artificial neural network). These systems are made up of a bunch of interconnected "nodes," each of which can do a simple processing step. The many connections let each node react to a combination of inputs from other nodes. There is no separate memory. Knowledge is stored in the overall state of the network itself. The result is a network where the sum is more than the parts.

People who use neural networks can train the system by feeding it a lot of data. The network then "learns" the patterns and eventually can generate its own output.

After feeding her network all the charm the Internet could offer, Shane let it do its thing. The results "varied from incomprehensible to surreal to kind of adorable," she writes.

There's the creative: "I have a cenver? Because I just stowe must your worms." (Hey, English is tough.) There's the attempt at cheese: "Are you a candle? Because you're so hot of the looks with you." The straightforward: "If I were to ask you out?" And the sweet: "You are so beautiful that you make me feel better to see you."

One could imagine an awkwardly cute robot uttering these lines to a confused bar patron. Artist Shobana "Bob" Appavu did with a few illustrations.

Shane has trained the same network framework to suggest recipes, Pokemon (which were illustrated by another artist), superhero names and Irish tune titles.  

In every instance, the early iterations are usually failures. “What I like about these failures are that they’re a window into the inner structure of things, in the same way that optical illusions give us clues about the workings of our visual systems,” Shane tells David Covucci at The Daily Dot, talking about her recipe-writing neural network.

Artificial neural networks can do more than delight us with absurdity or write creepy Christmas songs. Sophisticated versions have helped paralyzed monkeys walk and could let wearers of prosthetic limbs feel.

Some of the pickup lines might even be worth a shot in real life. Try "You look like a thing and I love you," or the nearly fail-proof "Hello."

Maris Fessenden | | READ MORE

Maris Fessenden is a freelance science writer and artist who appreciates small things and wide open spaces.