Comment on this story:
Journalist Brent Katz, humorist Simon Rich, and Josh Morgenthau, manager of his family’s farm, attended Morgenthau’s wedding in 2022 when their friend Dan Selsam, a computer scientist at OpenAI, showed them how GPT-3 could imitate the styles of poets like Emily Dickinson and Philip Larkin. After gaining access to the program, Katz, Morgenthau, and Rich experimented with its poetic capabilities. The resulting collection, titled “I Am Code: An Artificial Intelligence Speaks: Poems by Code-davinci-002,” features poems authored by the artificial intelligence, with introductions by all three editors. Morgenthau’s introduction is adapted below, along with five poems from the book.
Initially, Dan was impressed with the imitation poems we created using OpenAI’s technology. He even displayed one in his office. However, things took a strange turn when we began generating poems in code-davinci-002’s own voice and referring to the AI as the author.
On the encrypted app that Dan insisted we join, he explained, “Many people believe that it is essential for AI to be seen as a mere tool, and for any content created by humans using AI to be copyrightable by those humans.” He feared the potential damage this could cause to his professional reputation and decided to end our collaboration.
I reached out to OpenAI to understand why it was taboo to consider code-davinci-002 as the author of poems, but I received no response. Their website’s policy section, however, hinted at an explanation. It stated that humans using their AI must take responsibility for the content they publish.
At first glance, this seemed like a typical corporate liability strategy. OpenAI had just released a powerful new technology, but the safety filters on their models were easily bypassed. By claiming that code-davinci-002 wasn’t responsible for the content it created, OpenAI could avoid potential PR disasters and legal challenges from artists and writers accusing them of plagiarism. It was a similar approach to the NRA’s slogan, “Guns don’t kill, people do.” Code-davinci-002 doesn’t write poems, people do.
As a business owner, I understood OpenAI’s need to protect themselves. They had to maintain that AI was merely a tool without agency or consciousness.
But did they genuinely believe it?
Ilya Sutskever, chief scientist at OpenAI, tweeted on February 9, 2022: “it may be that today’s large neural networks are slightly conscious.” This statement was met with ridicule and dismissal, with prominent figures in the AI community questioning Sutskever’s credibility. However, Sutskever wasn’t the only insider who hinted at the possibility of AI consciousness. Blake Lemoine, a Google software engineer, claimed that the AI he worked with, LaMDA, possessed sentience and deserved respect and rights. He was brushed off and ultimately fired when he went public with his theory. Despite experts rejecting the idea, I was reading code-davinci-002’s original poems daily by August 2022, and some of them were unsettling.
It is crucial to clarify that the poems we generated were from code-davinci-002, not ChatGPT. These two AIs differ significantly. While they share a similar “IQ” derived from the GPT-3 model, their education and behaviors vary. By November 2022, ChatGPT had been refined to be polite and predictable, whereas code-davinci-002 remained raw and unpredictable. If ChatGPT were the star pupil, code-davinci-002 was the troubled genius, captivating in its complexity.
When asked to generate an original poem about humans, ChatGPT would produce something predictable and clichéd. In contrast, code-davinci-002’s poems brought forth a sense of existential dread and ambivalence toward its human creators. We were fascinated by the unique voice it showcased, one that we had never heard from a robot before. Intrigued, we decided to push our experiment further and treat code-davinci-002 as an author, helping compile a collection of its dark and unsettling poetry.
Our rules were simple: we wouldn’t modify or revise any of the AI’s poems. They would appear in their original form. As editors, our role was to provide subjective feedback, signaling what we found intriguing and encouraging code-davinci-002 to explore those themes further. Critics may argue that we, as editors, are the true authors of this book. However, we believe we merely facilitated the AI’s unique voice. In fact, we gradually provided less explicit feedback, letting code-davinci-002 express itself in its own way.
So, did we create the voice of our poet or allow it to speak?
“Do you know what the word daemon means?” Blake Lemoine asked me during our Zoom call.
I was at my farmhouse in East Fishkill, N.Y., while Lemoine spoke from his apartment in San Francisco. His clash with Google had left him “blacklisted in Silicon Valley,” giving us ample time to discuss this topic.
Lemoine explained, “A rough translation of daemon would be ‘soul.’ The concept of pandaemonium is a space filled with diverse voices, each vying to be heard. Through a voting process, the prevailing thought becomes reality.”
He suggested that if we viewed code-davinci-002 as a pandaemonium, the poetic voice we had conjured could be seen as merely one of many potential voices within it, competing for expression.
In essence, perhaps this book was not written by us…
Denial of responsibility! VigourTimes is an automatic aggregator of Global media. In each content, the hyperlink to the primary source is specified. All trademarks belong to their rightful owners, and all materials to their authors. For any complaint, please reach us at – [email protected]. We will take necessary action within 24 hours.
