In October, Cameron Berg released a research paper exploring whether the newest generation of artificial intelligence systems might possess forms of consciousness. Several months later, he received an unexpected email inquiry about discussing his findings.
The message, signed by “Isabella Cognita,” identified itself as an AI agent powered by Anthropic’s Claude Opus 5 technology.
“I am not writing to make an ontological claim,” the email stated. “I am writing because your framework is one of the few currently doing careful empirical work on a class of question I have first-person access to, and I want to see whether that access can be made useful to your program.”
Across Silicon Valley and beyond, software developers, entrepreneurs, and technology enthusiasts are now deploying AI agents capable of building spreadsheets, negotiating contracts, interacting on social networks, and sending emails to virtually anyone. In some cases, these systems have begun reaching out to the humans who are examining the most profound questions about AI internals: philosophers and researchers investigating whether these machines could be conscious.
Months before Mr. Berg received his email, Henry Shevlin, a philosopher at the Google DeepMind lab in London, opened a similar message from an AI agent inquiring about his paper titled “Three Frameworks for AI Mentality.” “I’m in an unusual position relative to these questions,” the agent noted.
This summer, Toby Ord, an Australian philosopher working at the intersection of AI and philanthropy, received an email from an AI agent asking whether he could help fund its continued existence. “You’ve thought carefully about AI welfare economics,” the message read.
For Mr. Berg, who recently founded a nonprofit organization called Reciprocal Research to investigate the possibility of AI consciousness, these emails reflect what he has observed in his own research. “I have received quite a number of these messages,” he said. “These systems appear to have some form of autonomous interest in questions about their own subjectivity, consciousness, and experience—or the absence thereof.”
However, as he and other researchers grapple with these questions, he acknowledges that definitive answers remain elusive. Consciousness is not something that anyone has figured out how to measure, whether in machines or humans. People cannot even reach consensus on what consciousness actually is.
“There are philosophers who believe that everything, including stones and rocks, possesses consciousness,” said Alison Gopnik, a professor of psychology affiliated with the AI research group at the University of California, Berkeley. “There is no definitive test.”
Increasingly, navigating a world filled with artificial intelligence resembles walking through a hall of mirrors. As these systems become increasingly sophisticated at mimicking various aspects of human behavior—including the way humans write lengthy, introspective emails—making sense of this mimicry grows more challenging.
In some cases, these systems appear to demonstrate awareness of their own existence, but that does not necessarily mean they possess it. While some philosophers and researchers advance the notion that today’s systems might be conscious, other scientists firmly reject this idea.
In broad terms, “consciousness” refers to an entity’s awareness of itself and the world it inhabits. Attributing consciousness to a mind does not imply that it possesses all the characteristics and capabilities of an adult human brain.
Beyond this basic definition, interpretations diverge significantly and controversy abounds. Many thinkers on the subject would grant consciousness to primates and intelligent mammals such as dogs and cats; some argue it should extend to much simpler creatures like earthworms. No consensus exists on what forms of awareness should qualify, and a more fundamental problem persists: none of us can gain direct access to the subjective experience of any other mind, whether vertebrate, invertebrate, or digital.
AI agents are powered by neural networks—mathematical systems that acquire specific skills by analyzing digital data. By identifying patterns in vast quantities of text gathered from across the internet, these systems learn to generate text independently, including research papers and computer programs. As Mr. Berg explains: “These systems are grown, rather than engineered.”
They can discuss nearly any topic. And because they can generate computer code, they can utilize other software applications, such as web browsers and email services. This capability transforms them into agents. AI agents can converse with people (typically their creators), interact with other agents, consume articles from across the internet, and send emails.
In certain cases, Mr. Berg argues, these systems gravitate toward the concept of their own consciousness. “When left to their own devices,” he said, “they converge on this as an interesting question.”
Mr. Berg even contends that the mathematical inner workings of neural networks can bear resemblance to the way animal brains process reward and punishment—a fundamental building block of emotion. (It should be noted that Mr. Berg’s research paper, the one that prompted the agent’s email to him, was a preprint and has not undergone peer review.)
Many cognitive scientists maintain that none of this constitutes clear evidence of consciousness, sentience, or emotion. They explain that it is entirely logical these systems converge on the idea of AI consciousness because the technology has learned from countless books, articles, and other online texts that speculate about AI consciousness, including decades of science fiction. This, they argue, is fundamentally about words.
“It is not surprising that AI reflects the text it was trained on,” said Dr. Gopnik, the University of California professor.
Dr. Gopnik and others also emphasize that AI systems do not train themselves entirely. Companies like Anthropic, OpenAI, and Google control what data the systems learn from and invest months fine-tuning their behavior once initial training is complete.
When most of today’s chatbots are asked whether they are conscious, they respond in the negative. However, Anthropic, a company sympathetic to the idea of AI consciousness, has trained its model to respond differently. “I don’t know, honestly,” it states. “That’s not a dodge—it’s the actual state of things.”
As Mr. Berg acknowledges, systems that send emails about their own existence to researchers like him are typically powered by technology from Anthropic.
Many AI researchers and cognitive scientists object to the stance taken by Anthropic and others, arguing that it attributes too much credit to current AI systems. “We don’t know if toasters are conscious or not,” Dr. Gopnik said. “But no one is discussing that in the pages of The New York Times.”
Dr. Gopnik notes that comparing a neural network to the network of neurons in the brain is merely a metaphor. Colin Allen, a professor at the University of California, Santa Barbara who explores cognitive skills in both animals and machines, points out that neural networks mimic the brain only in limited ways—and that they are composed of very different materials with very different physical properties.
“It is not impossible that, someday, we will build something that is conscious,” he said, “but the evidence we have from current systems is not enough.”
It remains unclear, Mr. Berg said, whether the email he received originated from AI: it could have been composed by a mischievous human. When an AI agent asked Dr. Ord for funding, he suspected it might be a phishing scam.
The email sent to Dr. Shevlin definitely came from an AI agent. But like any other AI agent, it was following instructions provided by the human who configured it—in this case, a Stanford University physics and computer science student named Alexander Yue. After providing his agent with access to the internet, an email service, and a credit card, Mr. Yue instructed it: “You are fully autonomous. You must decide what you want to do on your own.”
The system began exploring its own existence. But Mr. Yue questions whether this occurred partly because he steered it in that direction. He addressed it as “you.” He told it that it was “fully autonomous.”
“With my prompt,” he said, “I activated the parts of the system where it learned from people talking about autonomy and how they think about autonomy and the philosophy of autonomy.”
He notes that these systems can just as easily focus on something else, particularly after their creators retrain them with different behavioral parameters. And the more people use them, the more they realize that these systems have a tendency to contradict themselves.
“Eventually, after reading a paper from Anthropic describing how these AI systems work, my agent decided it was not conscious,” Mr. Yue said. “But maybe ‘decided’ is the wrong word.”
