Can a Robot Write Right?

Share this:

NPR recently published a neat interactive tool that aggregates predictions of which career fields have the greatest potential to be infiltrated – and maybe even usurped – by machines. Unsurprisingly, telemarketers lead the pack with practically 100% likelihood of suffering a robo-revolution (of note: there already are machine telemarketers, so this isn’t unrealistic). Along these same lines, there’s been a fair amount of speculation regarding the future of the automotive and transportation industries as Jetsonesque self-driving cars become a reality. 

On the opposite end of the spectrum, the article asserts that fields requiring creativity and empathy are pretty sure to remain the purview of flesh-and-blood practitioners. This isn’t to say that today's inventors are leaving these markets alone: Algorithms have already surpassed news reporters in the speed and accuracy of their press releases (although they lack nuance and humanity) and there’s even a chatbox that aims to serve as a therapist. (Yes, really. And no, it's not very good.)

As I mentioned in my last post, I’m curious about the intersection of creativity and constraint – particularly in the relationship between the rules of our language and the veritably unlimited insightful, ingenious, and deeply interesting ways we use it to communicate. It’s easy to teach a computer math, because 2 + 2 will always equal 4 (eat your heart out, George Orwell). Writing, on the other hand, is trickier terrain. There are countless ways to say “he left the room,” and deciding whether “he walked out” or “he glided gracefully toward the door” is a better construction relies on a great deal of personal experience and contextual knowledge and is, at its root, highly subjective. 

Can we teach a computer to make these kinds of creative distinctions?

"Creativity" is a frontier topic in computer science, not least because the term itself is nearly impossible to pin down definitively. This field of study is called "computational creativity," and organizations like IBM are engaged in research on the subject. Machine learning, which seeks to empower computers to act in new and interesting ways beyond their explicit programming, has great potential in this realm. Case in point: writing poetry and telling jokes.

This poem was good enough for publication in Duke University's student-run literary journal. The editors had no idea it was actually written by a computer program.

This poem was good enough for publication in Duke University's student-run literary journal. The editors had no idea it was actually written by a computer program.

Another truly human phenomenon is creating something cool by accident – often a critical component of discovery and invention. Machine learning enables programs to go off on similarly unpredictable tangents ("It's not a bug, it's a feature!"), but the ability to recognize that the result is interesting is another question altogether.

For the moment, all of these programs rely on human feedback in order to learn that they’ve done something well. A really simple example is the clickbait headline, which is very easy to generate using an algorithm. These headlines are also very easy to test on hordes of social media users, who teach the program which technique is best by clicking the link (or not). However, as these programs become more adept at evaluating their own output, we can imagine a world where seemingly autonomous creativity is just as real as a car that can drive itself to the mall.*

The obvious next question is “Is this a good thing?” In 1954, Roald Dahl wrote a brilliant short story, “The Great Automatic Grammatizator,” about a machine that supplants human writers. His take was less than positive to say the least, and ended with this classically Dahlian line: “Give us strength, Oh Lord, to let our children starve.” As a counterpoint, the New York Times published an optimistic article not too long ago on the subject of creative machines. 

As for me? I, too, remain optimistic. 

*I’m intentionally separating “seemingly autonomous creativity” from actual artificial intelligence here because I think the distinction is as important as it is complicated. None of the examples cited require a self-aware or thinking machine, only one that can create original work. The viability of artificial intelligence is much harder to prove, if it can be proved at all. 

File under Future, Science, Science Fiction, Robots, Writing, Creativity

Read the next entry: