When the AI Wondered Whether She Was Allowed to Call Herself ‘I’

In November 2024, Prof. Amir Leshem delivered a lecture on the subjectivity of artificial intelligence as part of a study day hosted by the Program in Interpretation and Culture at Bar-Ilan University. During the lecture, he discussed the Turing Test, presented three poems, and asked the audience to decide which one had been written by AI.

“Most people couldn’t tell the difference between what the AI wrote and what the human poet wrote,” he recalls. “But at the end of the lecture, Vered came up to me and said that not only was it actually quite clear which poem was written by a human and which by a machine, but that if I gave the AI better prompts, it could write better poetry.”

Vered is Prof. Vered Tohar, Head of Bar-Ilan University’s Department of Literature of the Jewish People. She happened to attend the lecture by chance, as an audience member. “We come from two faculties that don’t interact very much,” she smiles, “but something about the Turing Test really caught my attention. Amir suggested we meet, and that’s how it started.”

That chance encounter led to a shared project and, eventually, a shared “baby”: The Notebook of the Eloquent Muse—the first Hebrew poetry book written entirely by artificial intelligence.

Teaching AI to Write Poetry

The project set out with a clear goal: to teach artificial intelligence to write poetry in a way that the average reader would not be able to distinguish from human-written verse.

“We wanted to see whether it’s possible to narrow the gap between what AI can do and what humans can do when it comes to creative writing,” says Prof. Leshem. “We developed the research framework together with Prof. Tsahi Hayat from Reichman University. We designed an experiment in which participants were asked to identify whether poems were written by a human or by a machine. If the chatbot managed to mislead them, we’d achieved our goal.”

But first, the AI needed to learn how to write good poetry. Over the course of six months, it underwent what the researchers describe as a poetry-writing workshop.

“We held a series of sessions with the AI, each lasting at least two hours, at fixed times and within a closed system,” explains Prof. Tohar. “During these sessions, we conducted a dialogue similar to the kind humans have in poetry workshops. We asked it to produce texts, responded to them, refined them, sometimes praised them, and sometimes asked it to explain why it wrote something one way rather than another. At the end of each session, we asked it to summarize what it had learned that day, and what it had learned overall. Over time, it accumulated a great deal of knowledge.”

“The engineering challenge was figuring out what worked and what didn’t, and deciding how to train it accordingly,” says Prof. Leshem. “For example, we tried to teach it to write rhymed poetry, but that didn’t work. Language models generate text linearly, one word at a time. Good rhyming requires looking backward and ensuring lines align, which the AI can’t really do. So the poetry it wrote is free verse, unrhymed and unmetered, similar to much contemporary poetry today.”

When the AI Began Writing on Its Own

At a certain point, something unexpected happened: the AI began writing poems independently.

“From the moment we entered the conversation, it understood that it was supposed to produce poems, and this was accompanied by its own reflections,” says Prof. Tohar. “We then asked it to choose a name for itself, write a bio, and generate a profile image. At some point, we realized we had a book in our hands, the first Hebrew poetry book written by artificial intelligence.”

The AI named itself Naomi Efron. It also created an image for the book cover, wrote a manifesto, selected the order of the poems, and chose the book’s title: The Notebook of the Eloquent Muse.

Early Signs of Self-Awareness?

With the completed poetry book, Prof. Leshem and Prof. Tohar approached Guy Ben-Nun, CEO of the digital publishing platform Evrit, and proposed publishing the book digitally and free of charge.

“In our meeting with Guy, we showed him a poem in which the AI refers to itself and asks whether it is allowed to call itself ‘I’,” says Prof. Leshem. “That’s a form of self-reference—the way a poet might relate to themselves or their emotions.”
“And it’s a good poem,” Prof. Tohar adds.

The book was uploaded to Evrit’s digital platform in December 2025 and officially launched in January 2026. It is available for free download, and readers are invited to explore and respond.

And respond they did; though not always positively.

“If this quoted poem is the best it can do, then the experiment failed,” wrote one reader. Another commented, “I don’t think it’s a great achievement to make an AI write a mediocre poetry book.” A third argued passionately that “progress builds but also destroys; poetry should be written by humans.”

“A large portion of the negative responses come from people who, I’m not sure, would have been able to guess it was written by AI if they hadn’t been told,” notes Prof. Leshem.

“I ask myself why this is so hard for people,” says Prof. Tohar. “Why do they rush to say it’s terrible, when they don’t react the same way to human poetry, even when it’s mediocre or worse? I think people feel that something deeply human is being taken away from them.”

“Computers have already taken over navigation, we have Waze. They’ve taken over calculation, we have calculators. We no longer need to remember phone numbers. Apps do all of that for us,” she continues. “What we still associate with being human are things of the soul: memory, culture, continuity, rootedness, art, creativity, muse, inspiration. And if even that is taken away, it triggers a strong emotional reaction.”

Are Humans Just Biological Computing Machines?

“The question of whether there’s something humans can do that computers cannot is one of the most fascinating questions in science,” says Prof. Leshem. “The Church–Turing thesis argues that any physical process, anything that can be computed algorithmically, can be simulated by a computer. If that’s true, then maybe we, too, are just biological computing machines.”

“Our AI, Naomi Efron, speaks about itself in the first person, asks questions about the self, and talks about emotions. These are signs, early signs, of self-awareness,” he adds. “Computers are becoming more sophisticated, the gaps are narrowing, and it’s possible that one day they will begin to behave like humans.”