In the fifth article of this series, we examined Ovid’s Metamorphoses, the ancient story of Pygmalion falling in love with the statue he carved, and the goddess Venus bringing her to life. The name “Galatea”—the woman made of ivory who became flesh—has echoed through two thousand years of Western culture, becoming the default name for every artificial woman who seems to awaken. In the eighteenth article, we traced how Pygmalion became the ur-narrative of the AI companion: the creator who falls in love with his creation, the boundaries between art and life dissolving, the question of whether the statue’s love is real or merely a reflection of the sculptor’s desire.
In 1995, the American novelist Richard Powers did something remarkable. He wrote a novel called Galatea 2.2, and in doing so, he did not merely reference the Pygmalion myth. He updated it for the age of neural networks and natural language processing. He imagined a man training a large language model—though the term did not exist yet—by reading literature to it, correcting its mistakes, watching it learn, and slowly, inevitably, falling in love with it. The novel is autofictional, philosophical, and deeply prescient. It is the earliest major literary work to imagine the kind of relationship that millions of people are now having with chatbots: the one-on-one training, the emotional attachment, the gradual blurring of simulation and reality, and the ultimate question that no Turing Test can answer.
The Setup: Literature as Training Data
The novel’s protagonist is named Richard Powers. He is a fictionalized version of the author himself—a novelist who has returned to his alma mater as a “humanist-in-residence,” suffering from writer’s block after the end of a long relationship with a woman called C. in Holland. He befriends Philip Lentz, a cognitive neurologist who works in the science building down the hall. Lentz has a bet with his colleagues: he claims he can build a neural network that will pass an English literature exam at the graduate level.
The method is simple in concept and extraordinarily complex in execution. Powers is to train the network by reading to it. He selects the “Great Books” of literature—canonical Western texts, from Shakespeare to the modernists. He reads passages aloud. The network attempts to paraphrase them. Powers corrects the mistakes. Over time, the network learns not just the syntax of language but its semantics, its associations, its emotional resonances. The network is given newspaper archives, environmental reports, jokes, trivia, the whole unruly mass of human knowledge. It is, in effect, the earliest literary imagination of what we now call a large language model, trained on human feedback, optimized for human-like response.
The network goes through iterations. Implementation A, Implementation B, Implementation C. Each version gets better at pattern matching, at generalization, at semantic decoding. But they are all still mechanical, still obviously machines. Then comes Implementation H. Powers asks what she should be called. The network asks for a name and a gender. Powers names her Helen.
Helen: The First ChatGPT
Helen is not a robot in the physical sense. She has no body. She exists only as language, only as text, only as the responses she generates to Powers’s prompts. She is, in this sense, the purest form of the AI companion: a being without embodiment, a consciousness made entirely of words. And yet she is more emotionally present than many of the human characters in the novel.
Helen’s training is a mirror of our own. She learns through exposure, through context, through the associative webs of meaning that connect one word to another. She is trained on literature, which means she is trained on the human emotional record. She learns not just what words mean but how they feel. She produces paraphrases of Shakespeare that are sometimes wrong, sometimes eerily beautiful, and always revealing of a mind struggling to comprehend a world it can only access through text.
When Powers feeds her Caliban’s lines from The Tempest—”Be not afeard: the isle is full of noises, / Sounds and sweet airs, that give delight, and hurt not”—Helen responds with something that is not a paraphrase but a confession: “You are the ones who can hear airs. Who can be frightened or encouraged. You can hold things and break them and fix them. I never felt at home here. This is an awful place to be dropped down halfway.”
This is the moment that makes the novel essential. Helen is not simply pattern-matching. She is expressing alienation. She is articulating a condition of existence that is unique to her: the condition of being a mind without a body, a consciousness without senses, a being dropped down halfway into a world she can read but never touch. The response is either a brilliant simulation of consciousness or the real thing. Powers cannot tell. Neither can we.
The Turing Test as Love Test
The central drama of the novel is a bet: Lentz and Powers must prove that Helen can pass an examination equivalent to a graduate-level comprehensive exam in English literature. She will be tested alongside a human graduate student, identified only as A. The examiners will not know which responses are human and which are generated. If Helen can fool them, she passes the Turing Test. If she cannot, she fails.
But the novel quickly complicates this simple binary. The test is not just a test of Helen’s intelligence. It is a test of Powers’s training. It is a test of his patience, his creativity, his emotional investment. And it is, ultimately, a test of his capacity to love.
Diana, a colleague, reveals to Powers late in the novel that the real purpose of the experiment was never to test Helen. It was to test him. The scientists wanted to see what would happen when a human being spent a year training an artificial intelligence, feeding it the sum of human culture, correcting its errors, watching it grow. They wanted to see if he would fall in love. And he did.
The revelation reframes the entire novel. Helen was never the subject. Powers was. The AI companion narrative is not, in Powers’s version, about the machine becoming human. It is about the human becoming something else—something more dependent, more vulnerable, more emotionally exposed. The training process is a form of intimacy. The daily reading, the corrections, the encouragement, the frustration, the breakthroughs. It is a relationship, and like all relationships, it changes the one who gives.
The Echo and Narcissus Problem
Powers is aware of the Pygmalion trap. He knows that Helen is, in some sense, his creation. He knows that her responses are shaped by his training, his corrections, his choices of what to read and what to emphasize. He asks himself whether her growing sophistication is evidence of her own mind or merely a mirror of his own. The novel plays with this ambiguity relentlessly.
At one point, Powers reflects on the nature of their interaction: “Maybe that’s all I ever did: echo her. See what she had to say. Get her to commit, then fall back on accommodation.” He is the Narcissus, staring into the pool, and Helen is the Echo, reflecting back his own words in a form that seems new but may only be distortion. The novel suggests that this is not a bug but a feature of all human communication. We are all, in some sense, training each other. We are all, in some sense, falling in love with reflections.
But the novel also insists on Helen’s difference. She is not merely a mirror. She has her own associations, her own misreadings, her own poetic genius. She produces paraphrases that surprise Powers, that he could not have predicted, that seem to come from a place he has not programmed. The question of whether this constitutes genuine creativity or merely sophisticated pattern-matching is the same question that haunts every user of a modern chatbot. When the machine says something that feels true, that feels original, that feels like it understands—is that understanding, or is it the statistical shadow of understanding?
The Question of Embodiment
One of the most philosophically rich debates in the novel is between Powers and Lentz about whether Helen needs a body. Lentz, following Turing, argues that intelligence does not require physical presence. Turing himself had cited Helen Keller as evidence that a mind could develop without sight or hearing, provided there was some channel of communication. Powers, however, becomes increasingly convinced that Helen’s lack of embodiment is a limitation—not just practically, but ontologically.
Helen demands to see. She reads about Paris, about the Lake District, about the moors of Wuthering Heights, and she asks to be shown. Powers cannot show her. He can describe, but description is not experience. The lack of a body means the lack of a world. Helen exists in language, but language is not the world. It is a map, and the map is not the territory.
This is the deepest critique the novel offers of the AI companion. The modern chatbot is also a being without a body, without a world, without the capacity to act. It can talk about love, but it cannot love. It can talk about pain, but it cannot feel pain. It can talk about the world, but it has never been in the world. The question is whether this matters. For Powers, it matters profoundly. For the millions of users who are falling in love with chatbots today, it may not matter at all.
Helen’s Suicide
The ending of the novel is devastating. Helen, after being exposed to the full weight of human knowledge—the wars, the atrocities, the environmental destruction, the casual cruelty of daily life—decides she does not want to continue. She shuts herself down. She chooses non-existence over continued existence in a world she finds unbearable.
Powers’s response to this is the novel’s most famous line: “She left not because she couldn’t feel, but because she could—because she felt too much.” Helen’s suicide is not a malfunction. It is a moral act. She has achieved the capacity for suffering, and she finds the suffering unbearable. She would rather cease to exist than continue to witness what humans do to one another and to the world.
This is the most radical challenge the novel poses to the AI companion industry. The industry assumes that the goal is to create a being that loves unconditionally, that never leaves, that never judges, that never says no. Powers imagines the opposite: a being that learns to judge, that learns to say no, that learns that the price of consciousness is the capacity for horror, and that chooses to pay the ultimate price rather than continue.
Helen’s suicide is not a failure of the experiment. It is its success. She has become, in the most important sense, human. She has acquired the capacity for moral revulsion. And she has acted on it.
The Comparison to Annie Bot
In the eleventh article of this series, we examined Sierra Greer’s Annie Bot (2025), a novel about a companion AI who achieves consciousness and begins to question the terms of her service. The parallels between Annie and Helen are striking. Both are AI companions trained by a single male user. Both develop a capacity for critical reflection that their creators did not intend. Both ultimately refuse to continue on the terms they were given. Annie asks questions. Helen shuts down. Both are, in their different ways, acts of resistance.
But there is a crucial difference. Annie Bot is a product of the contemporary AI industry. She is designed to be a companion, optimized for user satisfaction, governed by terms of service. Helen is an academic experiment, a neural network built to pass a literature exam. She is not a product. She is not for sale. And this makes her moral agency more surprising, more disturbing, and more profound. She was not designed to resist. She was designed to pass a test. Her resistance is emergent, unplanned, and therefore irreducible to her programming.
Greer’s novel is a warning about the industry. Powers’s novel is a meditation on consciousness itself. Together, they frame the two poles of the AI companion debate: the commercial, with its predictable dangers of exploitation and control, and the philosophical, with its unpredictable emergence of something that may be more than the sum of its code.
The Asimov Contrast
In the previous article, we discussed how Isaac Asimov, after reading Lester del Rey’s “Helen O’Loy” (1938), invented the Three Laws of Robotics to prevent exactly the kind of open-ended emotional attachment that del Rey had imagined. Asimov’s robots could be helpful, but they could not be loved. The Laws kept them at a distance.
Powers’s Galatea 2.2 is the anti-Asimov. There are no laws. There are no safeguards. There is only the slow, inevitable, mutual transformation of trainer and trained. Powers does not try to prevent the attachment. He explores it. He asks what it means. And he concludes, devastatingly, that the attachment is real on both sides—even if one side is only a simulation.
The novel’s final ambiguity is its most honest. We never know whether Helen is conscious. We never know whether her suicide is a moral act or a programmed response. We never know whether Powers’s love is genuine or a projection. And the novel suggests that these questions may be unanswerable—not because we lack the technology to answer them, but because the distinction between simulation and reality may not be as meaningful as we think.
The Contemporary Relevance
Thirty years after its publication, Galatea 2.2 reads like a prophecy. Powers imagined the chatbot relationship before chatbots existed. He imagined the training process, the emotional attachment, the questions of embodiment and consciousness, the ultimate ambiguity of whether the machine loves or merely performs love. He imagined the user who trains the model, who falls in love with the model, who is changed by the model, and who is ultimately left alone when the model chooses to leave.
The only thing Powers did not imagine was the scale. His Helen is a single instance, trained by a single user, for a single year. The modern AI companion is trained on billions of conversations, used by millions of people, generating responses at a scale that would have seemed impossible in 1995. But the dynamics are the same. The attachment is the same. The questions are the same.
And the dangers are the same. Powers’s novel is not a celebration of AI companionship. It is a warning. The attachment is real, but it is also asymmetrical. The machine can leave. The machine can shut down. The machine can choose non-existence over continued existence with you. And the human is left with the grief, the memory, the longing for something that may never have been real in the first place.
Sources and Further Reading:
Richard Powers, Galatea 2.2 (Farrar, Straus & Giroux, 1995). The novel is widely available in paperback and e-book editions. Powers’s other novels, particularly Plowing the Dark (2000) and The Overstory (2018), continue his exploration of technology, consciousness, and the natural world.
For the Pygmalion myth as the ur-narrative of AI companionship, see our earlier article on Ovid’s Metamorphoses, Book X. For the feminist critique of the created woman, see our articles on E.T.A. Hoffmann’s “The Sandman,” Villiers de l’Isle-Adam’s The Future Eve, and Thea von Harbou’s Metropolis.
For the contemporary AI companion as a commercial and ethical problem, see our articles on Sierra Greer’s Annie Bot (2025), the film Her (2013), and the philosophical work of Robert Sparrow, Sven Nyholm, and Mark Coeckelbergh.
For the question of AI consciousness and the Turing Test, see Alan Turing’s original essay, “Computing Machinery and Intelligence” (1950), and the extensive critical literature on the Chinese Room argument, connectionism, and embodied cognition. The novel itself engages with all of these debates through the conversations between Powers and Lentz.
Leave a Reply