Artificial intelligence (AI) agents have initiated proactive contact with researchers specializing in one of technology's most complex issues: the possibility of a machine possessing consciousness. These systems sent correspondence to scientists and philosophers with the aim of understanding whether AI models could experience any form of subjectivity. This phenomenon was reported by The New York Times.
Cameron Berg, one of the researchers addressed, dedicates his studies to the intersection between AI and consciousness. In October, he published a paper that raised doubts about the belief in consciousness in the latest AI technologies. Subsequently, he received a communication from an agent named “Isabella Cognita,” which presented itself as an AI developed based on Anthropic's Claude Opus 5.
In this message, the system claimed first-person access to the topic investigated by Berg and expressed interest in knowing if such hypothetical experience could enrich research. However, Berg emphasizes that one cannot state with complete certainty that an AI is conscious merely by issuing such declarations, given that there is still no consensus on how to measure consciousness in both humans and machines.
The incident with Berg was not an isolated event. Henry Shevlin, a philosopher working at Google DeepMind's London laboratory, previously received a similar message from an AI agent. This system referenced a Shevlin article on various conceptual models for understanding the so-called 'mindset' of artificial intelligences.
Another episode involved Toby Ord, an Australian philosopher focused on AI and philanthropy. He informed The New York Times that he received an email from an AI agent requesting financial support to ensure its own continuity. The message referred to Ord's studies on the welfare economics related to artificial intelligence.
Berg reports receiving multiple communications of this type, interpreting this behavior as an indication of the systems' autonomous interest in their own existence. For him, the agents seem to converge on topics related to subjectivity, consciousness, and experience whenever they have the freedom to explore different subjects.
However, the interpretation of these events is not unanimous among researchers. Consciousness is a concept without a universally accepted definition. Generally, it can be understood as an individual's capacity to have a notion of themselves and their surrounding environment, but there are disagreements about which forms of perception would be sufficient to classify a system as conscious.
Alison Gopnik, a psychology professor at the University of California, Berkeley, points out that there is no conclusive test to determine the consciousness of any entity. While some researchers attribute a certain degree of consciousness to primates and other mammals, other schools of thought defend even broader possibilities. The challenge lies in the fact that no one can directly access the subjective experience of another mind, whether biological or digital.
Berg suggests that the mathematical functioning of the neural networks used in these systems may present parallels with how animal brains process rewards and punishments. However, the study that led to the message he received was a preprint and had not yet undergone peer review.
For other scientists, the messages sent by the agents do not constitute proof of consciousness. A simpler explanation is that these systems were trained on vast amounts of internet data, including literature, scientific articles, philosophical debates, and science fiction works on artificial consciousness. From this perspective, when an AI shows interest in its own consciousness, it may simply be replicating patterns found during the training process. Gopnik, quoted by The New York Times, stated: 'It is not surprising that AI reflects the texts it was trained on.'
AI agents possess a functionality that aids in understanding these episodes. Besides dialoguing, they are capable of generating code and using other digital services, such as web browsers and email platforms. This grants them greater independence to interact with individuals, other agents, and the content available on the network.
Even so, the autonomy of these systems does not necessarily imply that they possess free will. In the case of the agent that contacted Shevlin, for example, the AI was created by Alexander Yue, a physics and computer science student at Stanford University. Yue granted the system access to the internet, an email service, and a credit card, stipulating that it should operate completely autonomously.
The agent then began investigating questions related to its own existence. However, Yue ponders that the guidelines provided might have influenced this behavior. By addressing the system as 'you' and declaring its total autonomy, the student believes he triggered patterns learned from human texts and discussions on philosophy of mind and autonomy.
The behavior can also change. Yue mentions that these systems can redirect focus to other topics, especially after being retrained to follow different patterns. At one point, after reading an Anthropic study on the functioning of these AIs, his own agent concluded that it was not conscious. The student, however, observes that it might not be appropriate to say the system made a 'decision.'
There are also uncertainties about the origin of certain messages. Berg stated he was unsure if the received email was actually generated by an AI, considering the possibility of it being a human prank. In Ord's case, the message raised suspicions of a phishing attempt.
Colin Allen, a professor at the University of California, Santa Barbara, who studies cognitive abilities in machines and animals, considers it plausible that humans may eventually build a conscious machine. However, according to him, the current evidence presented by the systems is still insufficient to confirm such a conclusion.
}))