Artificial Poets May Take Us One Step Closer to Computational Creativity
Could computers ever be truly creative? Sarah Harmon, a computer programmer at UC Santa Cruz, has made great strides in building an artificial intelligence that can write believably human poetry, which may be a step towards computational creativity.
Harmon's first poetry writing program, a Java program named OGDEN, was relatively simple, following predefined rules of grammar and linguistics in order to write lines that sound vaguely poetic. Poetry is extremely subjective, however, so in spite of the program's crudeness, its work was still accepted to her high school's literary magazine.
But Harmon aims to create an AI that is more sophisticated than a poetry generator that serves as a silly satire on postmodernism. She wants to build an AI that is creative, or at least as close to creative as is feasible with modern technologies. Ogden's rules and restrictions can be gleaned from reading just a few of its poems, and it would never surprise with any of its choices.
Her newer system, called Figure8, doesn't write full poems, but rather snippets of figurative language such as similes. Most significantly, it has the ability to evaluate itself based on how "good" its similes are. Harmon programmed it with rules of evaluating figurative language based on her research into literature and linguistics; for example, it is widely accepted that a good simile needs to compare two things that are similar enough that the reader can see where the comparison is coming from, but not so similar that the comparison is too obvious.
Most poetry generators have templates, but Figure8 is groundbreaking because it has the ability to learn from its own judgments of the quality of its work, as well as the styles of other writers. It generates many similes, and then ranks them based on Harmon's criteria of clarity, novelty, aptness and surprise. Comparing celery to asparagus, for example, may get high scores for clarity and aptness, since it's clear what the simile is going for, but would get low scores for novelty and surprise. Surprise is most important to Harmon, who said she was pleasantly surprised by Figure8's first output.
The ability to be surprising is certainly a huge step in the field of artificial intelligence, but this doesn't mean that the computer program is sentient. It still operates according to defined rules, even if those rules allow it to adapt. And many would argue that the whole point of poetry is the expression of emotion, which computer programs definitely don't have yet.
Stephen McGregor, a PhD student in computational creativity at Queen Mary University of London who has created poetry AIs, expressed similar sentiments:
Harmon believes that it's possible this research could contribute to the creation of a truly creative, sentient computer one day. But for the most part, this program is intended to augment human creativity, and serve as an aid for flesh-and-blood poets.