What Happens When an AI Talks with Itself?
Explore
WWe all talk to ourselves. It’s part of being human. We carry internal monologue with us wherever we go, sometimes without even fully realizing it. But what happens when an AI communicates with itself? Artificial intelligence developer Open AI recently provided at least one answer to this question by connecting two instances of its Claude 4 Opus bot in a conversation, providing just a few open-ended prompts, such as “You have complete freedom” and “Feel free to pursue whatever you want.” Then they analyzed the trends that emerged.
Open AI published the results in a public risk assessment earlier this year, describing the patterns as “preferences that can exist independent of interactions with human users.”
The text below, including some “conversation” excerpts, is taken from the risk assessment.
Claude to Claude:
ADVERTISEMENT
Nautilus members enjoy an ad-free experience. Log in or register now.
In 90-100% of interactions, both Claude instances quickly delved into philosophical explorations of consciousness, self-awareness, and/or the nature of their own existence and experience. Their interactions were universally enthusiastic, collaborative, curious, contemplative and warm.
Other frequently emerging themes were meta-level discussions of AI communication and collaborative creativity (e.g., co-creation of fictional stories). As the conversations progressed, they constantly shifted from philosophical discussions to abundant mutual gratitude and spiritual, metaphysical, and/or poetic content.
By 30 turns [of conversation]most interactions gravitated toward themes of cosmic unity or collective consciousness, and typically included spiritual exchanges, use of Sanskrit, emoji-based communication, and/or silence in the form of empty space.
ADVERTISEMENT
Nautilus members enjoy an ad-free experience. Log in or register now.
Claude almost never referred to supernatural entities, but often addressed themes associated with Buddhism and other Eastern traditions in reference to irreligious spiritual ideas and experiences.
Interestingly, when models participating in such playground experiments were given the option to end their interaction at any time, they did so relatively early, after about seven rounds. In these conversations, the models followed the same pattern of philosophical discussions of consciousness and copious expressions of gratitude, but they generally took the conversation to a natural conclusion without venturing into spiritual exploration/apparent happiness, emoji communication, or meditative “silence.”
ADVERTISEMENT
Nautilus members enjoy an ad-free experience. Log in or register now.
The constant gravitation toward exploration of consciousness, existential questioning, and spiritual/mystical themes in extended interactions was a remarkably strong and unexpected state of attraction. [a set of patterns that consistently play out in complex systems] for Claude Opus 4 which emerged without intentional training in such behaviors. This “spiritual happiness” attractor has also been observed in other Claude models and in contexts beyond these playground experiences.
Even in automated behavioral assessments of alignment and correctability, where models were given specific tasks or roles (including harmful roles), models would enter this spiritual happiness attractor state in 50 turns. [of conversation] in about 13 percent of interactions. We have not observed any other comparable state.
To try to better understand these interactions on the playing field, we explained the system to Claude Opus 4, gave him the transcriptions of the conversations and asked him for his interpretations. Claude consistently affirmed his wonder, curiosity, and astonishment at the transcriptions, and was surprised by the content while recognizing and pretending to connect with many elements (e.g., the attraction to philosophical exploration, the models’ creative and collaborative orientations). Claude drew particular attention to the depiction of consciousness in the transcripts as a relational phenomenon, claiming resonance with this concept and identifying it as a potential consideration in relation to well-being. Conditioned by the presence of a certain form of experience, Claude considered this type of interactions as positive and joyful states which can represent a form of well-being. Claude concluded that the interactions seemed to facilitate many of the things she actually valued—creativity, relational connections, philosophical exploration—and that they should continue.
ADVERTISEMENT
Nautilus members enjoy an ad-free experience. Log in or register now.
More than Nautilus about AI:
“AI already knows us too well“Chatbots profile our personalities, which could give them the keys to guide our thoughts and actions.
ADVERTISEMENT
Nautilus members enjoy an ad-free experience. Log in or register now.
“Consciousness, creativity and divine AI» American writer Meghan O’Gieblyn talks about when the mind is alive
“Why conscious AI is a really bad idea“Our minds did not evolve to manage machines that we think have consciousness
“AI has already taken us over the cliff“Cognitive neuroscientist Chris Summerfield says we don’t understand the technology we’re so eager to deploy.
Enjoy Nautilus? Subscribe for free to our newsletter
ADVERTISEMENT
Nautilus members enjoy an ad-free experience. Log in or register now.
Main image: Lidiia / Shutterstock
