In 1859, George Eliot confided in her friend Charles Bray about her literary ambitions: “The only effect I ardently long to produce by my writings, is that those who read them should be able to imagine and to feel the pains and joys of those who differ from themselves.” Eliot, who went on to publish The Mill on the Floss and Middlemarch, knew instinctively that writing can—and should—stir readers, leading them down pathways and into worlds beyond their own.

That spirit runs through Naomi Baron’s new book, Reader Bot: What Happens When AI Reads and Why It Matters. Baron, professor emerita of linguistics in the Department of World Languages and Cultures, centers her work on large language models (LLMs) such as ChatGPT, Meta AI, and Claude. She examines what is gained and, critically, what is lost when these predictive algorithms read on our behalf.

To call what AI does “reading” is itself a stretch. Whereas people might read to understand—reflecting on the ideas presented, drawing on memories, and making connections with other works—LLMs read datasets to see which words appear most often and in what order. The writing that results is a string of text, Baron explains, pieced together based on predictions rather than an underlying thought or argument. While AI writing can mirror a human’s, and some, Baron concedes, is “astoundingly good,” the tools raise important questions about the value of reading now and in the years to come.

“In answering AI’s siren call to do the reading for us,” Baron writes, “we risk weakening our own reading skills.”

Among the reasons we read is to know ourselves—to understand, Baron says, “who we want to be as people.” Take The Adventures of Huckleberry Finn. To immerse ourselves in Mark Twain’s masterwork is to grapple with the character of Huck, with the abuse he faces and the choices he makes. “How did he decide what to believe?” Baron asks. “How do I?” The act of reading turns on the time we devote to it, the comparisons drawn, the memories it brings to the fore. To read closely is to see characters not as two-dimensional representations but as people with rich inner lives, with problems like our own.

There’s no single way to interpret a piece of writing, Baron says. Imagine two people reading the same book, she suggests. They will most likely come to different conclusions, she says, because “they’ve had different lives.” ChatGPT might be able to summarize a book, arriving at a accurately constructed storyline, but it can’t re-create the experience of being absorbed by a work of literature, of lively details and colorful turns of phrase. As Baron writes, “Do you want to know the plot or do you want to experience human struggles?”

To read closely, to give ourselves over to the written word, is to see the world anew, “to imagine and to feel the pains and joys” of others, as Eliot muses—to go somewhere we’ve never gone before.