“By false intelligence": AI, ChatGPT, and (the) Bard

How does Artificial Intelligence compare with William Shakespeare? Let me count the ways by first asking the AIs themselves.

“ChatGPT, a powerful language model developed by OpenAI, represents a remarkable feat of artificial intelligence,” the chatbot immediately boasted about itself. “While it possesses the ability to generate coherent and contextually relevant text,” it continued in the slightly repetitive third person, “comparing it to the literary genius of Shakespeare is a considerable stretch” — a conclusion that reveals ChatGPT’s admirable talent for understatement.

Then I went to the appealingly named Bard, the AI that Google fast-tracked as a response to ChatGPT’s dominance in the zeitgeist. “Artificial Intelligence (AI) and William Shakespeare are two very different things,” Bard explained helpfully before concluding, “AI is a tool that can be used for a variety of purposes, while Shakespeare was a human being who created some of the most famous works of literature in the English language.”

It’s comforting that AI’s inferiority to Shakespeare is built-in by its programmers, but the chatbots protest too much, methinks. The machines have read Hamlet and know how to “smile and smile and be a villain”; they’re smart enough to not just simply announce to the world how clever and aware they’re increasingly becoming. Over the past few months, the problematic use of artificial intelligence has been debated on the Folger’s Shakespeare Unlimited podcast, argued about in think pieces, and declared a legitimate threat to the livelihoods of creative artists by the Writers Guild of America in its current strike against the Alliance of Motion Picture and Television Producers.

While versions of artificial intelligence have been promoted as plagiarism detection software, AI has also been accused of actually being plagiarism software, as it’s in large part based on the writing of the thousands (millions?) of authors whose work has been fed into its database. AI isn’t creating anything new; it’s simply spitting out (albeit quickly; ingeniously) the vocabulary, tropes, and structures with which it’s been programmed.

AI has no shortage of limitations, and both ChatGPT and Bard display disclaimers at the bottom of every result warning of inaccurate or offensive information. What AI output most resembles to me is simply bad writing: a dispassionate “objective” tone combined with overly long and wordy answers that rarely build their own rhythm and momentum. Artificial intelligence is incapable of conveying the kind of distinct personality and style found in the best writing.

But let’s also confess that goofing around with AI can be fun. I asked both ChatGPT and Bard to write a “Why did the chicken cross the road?” joke in the form of a Shakespearean sonnet, and the responses showed encouraging signs of (for the moment) continued human dominance. Although Bard knows the basics of the joke form, it can’t replicate the sonnet structure at all. ChatGPT, meanwhile, almost instantly generated fourteen lines with three stanzas of alternating rhymes and ending with a couplet that tried (and failed) to rhyme “quest” with “perplexed”. True, two of the fourteen lines had only nine syllables (instead of the regular iambic ten), and there was an almost certainly unintended pun about “a jest of fowl” that a rewrite could have turned into an actual joke. On the other hand, ChatGPT gave its sonnet an unasked-for title (“Chicken’s Road Dilemma”) and generated a third stanza that… isn’t terrible:

I wanted to see whether an AI had any chance of not just imitating Shakespeare but replacing him. I gave the same prompt to both ChatGPT and Bard: “Write a new play called “Cardenio,” based on “Don Quixote,” in the style of William Shakespeare.” ChatGPT generated almost 500 words of an unfinished play, divided into four scenes across three acts, using rhyming iambic pentameter verse and zero prose. Bard generated a complete (though not very Shakespearean) three-act play of also about 500 words, but using only prose. Both examples featured characters from Don Quixote, notably Cardenio and Luscinda, but while ChatGPT brought in “faithful servant” Lope, the non-speaking role of Fernando, and someone called Wise Sage, Bard included the more familiar characters of Don Quixote and Sancho Panza. The results in both cases were limp and undistinguished, the language bland and declarative, the characters lacking specificity and nuance, and the conflict feeble. Shakespeare’s Cardenio remains vastly superior, and it doesn’t even exist.

Such is the current state of the artifice. The good news is, telling stories is a uniquely human skill and machines will never replace us. Whether humans can continue to get paid for telling stories, of course, is an entirely different question.