PPP: Experience, Memory, and Insight

"Nietzche [sic] said that poetry has turned beasts into men. Perhaps, then, poetry will turn computers into men? Next question."

—Racter

"Writing poetry with a computer is like eating spaghetti with a fork a yard long. It can be done, but it's not easy."

—Kevin McKean

Poetry and computers have been at least vaguely linked since Gulliver’s Travels, which mentioned a machine that allowed "the most ignorant person . . . [to] write books in philosophy, poetry, politics, law, mathematics, and theology without the least assistance from genius or study."

In a couple of previous columns, I've talked about rhyme and meter and how they work from a linguistic point of view. You'd think that given a codification of such ideas, it shouldn't be hard to write a computer program that can write poetry. But you'd be wrong.

Most poetry-related software falls into one of two categories:

  • From a short list of words, the program chooses a word at random and displays it on the screen. This step is repeated several times, with occasional random line breaks and punctuation. This is the approach that many beginning programming students take; in theory, it's supposed to result in e e cummings-like poems. Of course, that idea is held only by people who don't recognize the artistry and skill that went into cummings' work.

  • Alternatively, the program doesn't try to generate poems at all; instead, it simply provides a human poet with ideas. Such programs may provide rhyming dictionaries, and may even provide help with meter; but the human supplies the creativity and the meaning. Although these programs don't "really" create computer-generated poetry, and in some ways it seems like a human who uses one is cheating, I don't see any strong distinction between such programs and other tools that poets use as creative aids, from rhyming dictionaries to magnetic poetry sets to ten-minute timed writings.

What most people are thinking of when they talk about computer-written poetry, though, is something else entirely: good poetry written entirely by a computer. But the closest to that that we're likely to come anytime soon is poetry that's generated according to metrical and phonological and syntactic rules. Computer programs could be written today that would start with a dictionary (containing pronunciations of words and information about their parts of speech) and some rules of syntax, and could generate syntactically correct sentences. Such programs could use the pronunciations given for the words in the dictionary to generate lines that followed metrical rules and rhymed. Such generated verse would satisfy most of the criteria for a certain kind of poem: it would rhyme, it would scan, it would contain syntactically correct English sentences.... And yet it would run aground on the same fundamental problem that those cummings-imitator poetry programs hit: the lack of meaning. No computer program in the world is close to being able to independently say something that humans find meaningful, except by accident.

There are efforts that could lead to a certain kind of apparent sense: for example, a long-running project called WordNet provides some semantic information about words, and there's a project dedicated to codifying millions of tidbits of information that humans classify as "common sense." A computer program that was designed to use such information sources might be able to produce verse that had some sort of coherence—it might stick to a subject by relying on words that have related meanings, for example, and it would be able to avoid such solecisms as having a door walk into a person (because the common-sense database would inform it that doors can't walk). The common-sense database would preclude a certain degree of playfulness and/or surrealism that many human poets aspire to, but never mind that. My point is that even after all that, the computer would still not be "writing" a poem in the sense in which I use the word, because there would be no intelligence behind the composition; if you looked behind the veil, you would see an easily understood mechanical process being followed. It could be argued (by the less metaphysically inclined among us) that a human poet is, at root, following a similarly mechanistic process; that may well be true, but in the case of the human, the process is sufficiently complex that we can't easily follow it, at which point it becomes simplest to assume that the semblance of intelligence implies something we can call actual intelligence.

I should add that I'm not at all in sympathy with those who say that it's objectively impossible for computers to produce art; I merely note that it's currently impossible. I concede—indeed, I hope—that at some point in the future we may produce computer-based intelligences; and I'm confident that if such intelligences are ever created, they will in turn create works that they will consider to be art. (Whether humans will recognize those works as art is a different question.) I'm looking forward with great curiosity to that time, though I now know that if it ever does happen, it will take many years longer than I once naively expected.

Some would argue that if a computer can produce real art, it somehow diminishes human achievement. I find this view incomprehensible. The best response I've seen to it is from Iain M. Banks' science fiction novel The Use of Weapons: "The fact a machine could have done it faster doesn't alter the fact that it was you who actually did it. ... [H]ave you ever been gliding, or swum underwater? ... Yet birds fly better than we do, and fish swim better. Do we stop gliding or swimming because of this?"

In a 1986 review of a couple of books about computers, thought, and souls, Thelma Z. Lavine wrote, "Experience, memory, and insight ... are the sources of our creativity." She went on to note that these capabilities are beyond the reach of current computers; but she added that there was no reason to believe that they would forever remain beyond the reach of computers. It'll be interesting to see what happens if and when computers have those capabilities.


The Lavine quote is from her article "Modern Philosophy in Search of Its Soul," The Washington Post, Sunday, July 13, 1986.

Join the Conversation