The computer-generated Jane Austen
It’s Independence Day here in the United States. What better occasion to consider the question of how soon our computer overlords will force us to submit to their rule?
You might not think Jane Austen would have much to contribute to this conversation, seeing as how she died two centuries ago. But not so! Recently, Austen herself – or, at least, a version of Austen powered by artificial intelligence – took a starring role in just such a discussion.
Last month, Oxford University’s famed debating society, the Oxford Union, held an event centered on the resolution, “This house believes most of the world’s content will soon be created by AI.” Both sides of the question were argued by an AI system trained to mimic four famous dead people: Winston Churchill, William Shakespeare, Oscar Wilde, and Austen.
From one standpoint, the AI, relying on technology called natural language processing, functioned with extraordinary proficiency, turning out iambic pentameter, witty dialogue, or soaring rhetoric, depending on whether it was portraying a Renaissance poet, a nineteenth-century comic writer, or a twentieth-century orator.
From another standpoint, the AI’s offerings were clunky and vapid. (Sample Jane Austen-style argument, delivered in the voice of Mrs. Bennet from Pride and Prejudice: “You see, there are now programs that can help create content that is more attractive to potential suitors. My husband and I are very excited about this new technology and we believe that it will be a great help to our daughters in finding good husbands.”) Samuel Johnson’s famously misogynist remark about women preaching and dogs walking on their hind legs comes to mind.
The event was intended to spur discussion about the implications of proliferating machine-generated content; unlike human writers, the AI was not engaged in a creative process, event organizer Alex Connock, a business professor at Oxford, noted in a press release issued by the university's business school. The AI was "merely statistically exploring training datasets,” Connock said. “Debate pyrotechnics notwithstanding, AI is nowhere near the finished article. It is still a toddler, though growing up fast.”
Meanwhile, the shortcomings of human-generated content are on vivid display in the Oxford press release, which contains two spelling errors. Say what you will about our computer overlords: At least they’d know better than to refer to “Jane Austin’s Pride and Prejudice.”