Do androids dream of writing the great American novel? 🤖

Hi readers,

Artificial intelligence isn’t just the stuff of sci-fi—it’s everywhere in our lives today, from natural disaster warnings to personality assessment to, well, almost everything that Google does. It’s not sentient robots (or a smartphone assistant you can really talk to)… yet. Instead, we’re talking machine learning, a kind of AI in which a computer “learns” how to do something by practice and iteration, or, in the case of neural networks, by being shown a large pool of examples, and extrapolating from there.

A simple, practical kind of neural net learning is facial recognition. Instead of programmers needing to write an extensive set of instructions defining what a face is and isn’t, they can show a neural net thousands of images of faces, and the computer extrapolates from there.

You’ve probably seen neural nets doing much less dignified work on Twitter, where they’re a favorite of amateur programmers and jokesters. You can ask a neural net to generate Goop product names or bird species—any kind of thing you can collect into a big list. Beth Skwarecki at Lifehacker wrote about how to make your own neural net earlier this summer, highlighting the strengths and limitations of the process: “The computer doesn’t actually understand the rules of, say, making recipes. It knows that beer can be an ingredient, and that things can be cut into cubes, but nobody has ever told it that beer is not one of those things. The outputs that look almost right, but misunderstand some fundamental rule, are often the most hilarious.”

These just-off-the-mark outputs reveal how neural nets are great at highlighting hidden rules, patterns, and conventions, unconstrained by human logic or restraint. And what’s more packed with rules, patterns, and conventions than stories? How far can today’s AI go? Well…

[Scene: Monica and Rachel’s, Monica and Phoebe are dancing.]

Van Damme: I’ll go in a crap.

Monica: Keep talking!

Phoebe: Wow lady! You’re just gonna come over to him jumpy. (They start to cry.)

Chandler: So, Phoebe likes my pants.

Monica: Chicken Bob!

Chandler: (in a muffin) (Runs to the girls to cry) Can I get some presents.

Yeah, no one’s passing the Turing Test there. That’s an excerpt from a neural net–generated Friends script. It’s funny, but in a Dadaist, nonsensical way.

But what about this bot-generated Saw movie script? Or this Olive Garden commercial? The short answer: Gizmodo has a guide for spotting fakes like these, though as long as you’re not being tricked, there’s plenty to enjoy in these human-generated scripts, too. When a comedian (it’s usually a comedian) writes a fake neural net output, they’re basically doing the same thing a neural net does: working on a bank of knowledge about a TV show’s (or movie’s, or commercial’s) tropes and patterns, and generating something that reflects those conventions and skewers them in surprising ways.

Real AI is developing its storytelling chops, though. A team of researchers from UC Santa Barbara taught a neural net called AREL (Adversarial REward Learning framework) to tell stories about images its shown, and, according to The Next Web, it did a pretty damn good job, passing a Turing Test 3/5ths of the time. We’re not at the point where computers are ready to write our novels or movie scripts, and this is more about seeing what AI can do than replacing human writers. But I bet computers don’t get writer’s block…

Happy reading,

Serial Box

< >