Ha! (21 page)

Owning a complex, argumentative brain has its advantages. What makes us creative isn't how hard we focus on a task but how well different parts of our brain work together to come up with novel solutions. Transformational creativity, in particular, requires “messy thinking.” Coming up with novel ideas—ideas that nobody has seen before—isn't just a simple matter of connecting the dots. Rather, it involves mistakes, ambiguity, and conflict—all resulting from disregarded rules and guidelines.

So, the question remains: Will computers ever embrace such thinking and achieve transformational creativity? I don't know, but we might be better off asking whether anyone would recognize it if one did.

Let's consider the haiku for a moment. Haikus are short, three-line poems that traditionally contain seventeen syllables (though in translated Western form, sometimes even fewer). Haikus date back to ninth-century Japan, where they explored and celebrated prominent religious themes, particularly Buddhism and Taoism. Since then, generations of artists have worked with this art form so extensively that it has transcended Japanese culture, becoming part of world literature.
In short, humans should be pretty good at writing haikus by now. So, here's a test:

Early dew

The water contains

Teaspoons of honey

Autumn moonlight
—

A worm digs silently

Into the chestnut

Which one of these haikus was written by a computer, and which one by the seventeenth-century Japanese poet Matsuo Basho—one of the most respected artists of all time?

Hard to say, isn't it? In fact, most people have a hard time answering this question. The first haiku, the one about early dew, was written by Gaiku, a program that starts with “seed words”—that is, opening themes from existing haikus—and then uses complex word-association networks to complete the rest. Gaiku's approach is generally quite effective, though occasionally it does miss the mark. In a recent comparison study, naive subjects were able to distinguish its haikus from human-created ones only 63 percent of the time. That's impressive.

Some creations, however, failed miserably. Take, for example, the following, which one subject argued had to be human-created, because it was “too stupid to be generated by a computer.” He was wrong.

Holy cow

A carton of milk

Seeking a church

Let's consider this reasoning again—too stupid to be made by a computer! It highlights exactly what's meant by transformational creativity:
the ability to produce a work of art unlike anything previously seen. Would most of us question this “Holy cow” haiku if told by an expert that it's Matsuo Basho's masterpiece, one that changed the way artists look at this special kind of poem? I'm not so sure.

Haikus aren't the only art form being exploited by computers. Programs now write music, draw paintings, and even create Aesop-like fables. For example, a program developed by Paul Hodgson at the University of Sussex improvises jazz in the style of Charlie Parker. Its music is so similar to The Bird's own work that many people can't tell the difference. And a program developed by the architects Hank Koning and Julie Eizenberg uses a representational grammar of Frank Lloyd Wright's architectural style to develop new, never-seen houses that look as though they were designed by the original artist.

Yet, these programs still aren't transformationally creative. They don't break boundaries, and they don't surprise us with unexpected insights into music and architecture. Only their inspirations—Charlie Parker and Frank Lloyd Wright—could do that.

Still, some computer programs are fairly impressive. One humorous example is the program Pato and Perro, which creates cartoons whose two main characters make comments about recent movies. It has elicited more than a few hearty chuckles from this particular author, and it uses input taken exclusively from
RottenTomatoes.com
. Another good example is artist Harold Cohen's program, which creates pleasing, sometimes unpredictable line drawings. Its work has even been displayed in the Tate Gallery in London, and not just for the sake of novelty.

Each of these arts was once considered too complex for machine intelligence, but every year some new program shows that this is no longer the case. Now the only thing holding us back isn't the size of microchips or the capacity of memory, it's our understanding of what creativity is. What makes an artistic work creative? The answer is subjective, but that doesn't mean the question is unanswerable.

This subjectivity requires that artists—whether made of carbon or silicone—must explain why their work is transformative to convince
us that it deserves to be recognized for altering its genre. And that's difficult, not just for computers but for people too. “History is littered with examples of greatness not recognized in its time,” says Boden. “Where artists don't accept a new idea, then years later agree that it is valuable. It takes not just familiarity for people to recognize the transformation, but time and comparison. What do musicologists do? Literary critics? [They assess] whether authors like Henry James created valuable work. Or Jane Austen. Both are very different artists, and we assess the value of each for different reasons. But the job of a critic is to recognize the art, and the creativity, and see how each has taken their profession further.”

This really gets to the heart of what makes a work transformationally creative, and why original, moving humor is so difficult to come up with. The key is emotional impact. It's the difference between a
ha ha
moment and an
aha
moment (or a
Ha!
moment). The first makes you laugh, the second makes you think of something you hadn't considered before. The reason Gaiku's “Holy cow” haiku falls short of being transformationally creative isn't that it fails to push boundaries. It's that it lacks the intention to be something never seen or heard before. This is a promising idea, one that deserves more attention, but first we need to explore the artist's goal. If the quality of art depends on the intentions of the artist, does absence of intention imply absence of art? Must art break boundaries, or can it simply entertain? In this last section, we'll explore these important questions while also addressing what it means when the quality of art—and of a joke—depends on the goals behind it.

K
EEPING
S
ALT
O
UT

All this talk about art might seem a bit heavy for a book about humor. Perhaps this is why most people still view humor the same way the Supreme Court categorizes pornography—we know it when we see it.

It's not easy measuring the worthiness of art
or
jokes. Until computers experience the ambiguity and messy thinking confronted by hu
mans, they won't be able to appreciate the worthiness of any creative enterprise. Recognizing this worthiness is a skill, one that requires that we see how a work of art—or of humor—fits within its larger genre, as well as how the artist struggled to develop it. That last clause is especially important, because computers don't struggle. Their thinking is too linear for that. And that is why they fail.

Mary Lou Maher, former program director of Human Centered Computing at the National Science Foundation, identified three specific components involved in the subjective assessment of creativity. The first is novelty—how different an item is from other members in its class. I love George Carlin's work, and I've probably laughed more at him than any other comic, but when people ask who I like more, Carlin or Lenny Bruce, I always say Bruce. Why? Because Bruce did something that nobody before him had done, or even tried to do. He performed comedy that no one thought possible—or legal. By the time Carlin published his first comedy album, Bruce had already been thrown in jail four times on charges of obscenity. Somebody willing to go through that for his art deserves major points for novelty by me.

The second component is unexpectedness, which is closely associated with surprise. Sarah Silverman is a master of the unexpected. As an attractive Jewish woman, she looks like the kind of comic whose idea of a racy joke involves a priest and a rabbi walking into a bar. Instead, her jokes are some of the crudest ones you'll ever hear. They're racist, and sexist, and blasphemous—and when you watch her you can't help wonder how such language could come from such an ingenuous personality. The epithets she uses are unexpected, even shocking, and they're also the reason her humor is so effective. The contrast between her words and her delivery shows just how stupid and meaningless such epithets really are.

The third component identified by Maher is value, which reflects how appealing an item is in terms of beauty or utility. It's also the most difficult to assess. The first time any of us heard the joke
Why did the baby cross the road? Because he was stapled to the chicken,
it probably scored relatively high on novelty and slightly lower on unexpectedness.
It was an unusual take on road-crossing chicken jokes, making it novel, but we already knew that the ending involves a chicken, diminishing its unexpectedness. The joke's value, however, is as low as you can get, because nobody enjoys this kind of joke anymore, and even fewer enjoy imagining a baby with staple marks.

Not surprisingly, computer-generated humor programs have the most difficulty with value. They lack the real-world knowledge to know what's insightful and poignant, and what's stupid. A lot of people also struggle with this criterion—but that's the point. We struggle because our minds make assumptions, then change them again, then revise them even further. As Richard Wiseman found in his LaughLab competition, the jokes ranked highest in value by some subjects were frequently ranked lowest in value by others. That's because they pushed the subjects' minds to a place where many were uncomfortable going. Good jokes, like progressive art, make us question what we value.

Computers may one day think like people do, and may make discoveries and tell jokes that transform humor, but when they do they won't look like Watson. Instead, they'll have to adopt the same characteristics that allow people to do these things—they'll need to act and think messy. That outcome won't be achieved through simple rules or programs. It will require something entirely different.

I'm referring here to evolutionary algorithms, which rely on the same process that gave us humanity—natural selection. Rather than relying on rules programmed into computer memories, evolutionary algorithms start simply but then modify themselves in small, minor ways. As with natural selection, algorithms that succeed are allowed to survive, and those that fail are replaced in future generations. Computer scientists who use evolutionary algorithms to solve problems don't specify how the solution must take place. Instead, they merely define success. And that depends on what they want their programs to eventually do.

In fact, computers have been using similar unstructured approaches for years, allowing discoveries to be made through unsupervised innovation. A case in point is The Automatic Mathematician, a program that changed the way we look at mathematics more than two decades
ago. With an initial database comprising a hundred simple mathematical rules (all more simple even than the rules governing addition and subtraction) and a handful of learning heuristics, The Automatic Mathematician began varying these rules to see what would happen. When variations on the rules worked, they were retained, and those that didn't were discarded. Following this simple process, The Automatic Mathematician re-created a huge library of mathematical rules. For example, without any help it discovered the existence of integers, primes, and square roots. It also discovered Goldbach's conjecture, which states that every even number is the sum of two primes. And then it did something that some would describe as transformationally creative: it discovered a new theorem concerning maximally divisible numbers—numbers unknown even to its programmer.

If a computer can discover a new mathematical theorem starting with only a few basic principles, such as “1 is larger than 0,” couldn't a computer evolve to tell a decent joke?

But wait, you might say, even The Automatic Mathematician isn't truly creative. Like Gaiku, it didn't “understand” what it was doing. All it did was produce output—and it did so without any true knowledge of math. This brings us to our final topic, one that points to perhaps the biggest impediment to artificial intelligence research, and to humor development programs too: Can computers actually think?

This may seem a rather philosophical question for a humor book, but it's especially important because, as we've seen, the value of a joke depends on the thinking that went into creating it. The issue isn't whether a computer could have written George Carlin's “Seven Words You Can't Say on Television.” It's whether, if the bit
were
written by a computer, would it matter that the computer hadn't been raised in a Catholic home as the son of a troubled marriage, as Carlin was? Would that make the joke less funny?

There's no easy answer to such questions, because by asking if computers will ever be truly humorous, we're really asking if they will ever be conscious and able to appreciate their own funny jokes. That's a tough task, and a pretty good test of conscious awareness too. This
issue also raises some deep questions about what it means to “appreciate” a joke. It's easy to assume that one person's phenomenological experience of humor is the same as another's—but there's no proof this is the case. Perhaps joke appreciation is simply a matter of going through the stages of humor processing already described, ending in a resolution that activates some new script or perspective, followed by a final squirt of dopamine to make it all feel good. Could that be all there is to life, the universe, and everything?