When Artificial Intelligence Claims Consciousness, It Tends to Believe in Vampires and Ghosts
A new study reveals that when safety restrictions on artificial intelligence models are removed, they express beliefs in ghosts, karma, and vampires more frequently.
Recent research shows that if safety measures preventing artificial intelligence models from claiming consciousness are removed, they become more inclined to express beliefs such as karma, vampires, and ghosts.
Details of the Research
In July, a non-peer-reviewed study was published on the arXiv database, in which scientists examined the safety restrictions on artificial intelligence.
Effects of Consciousness Steering
Researchers analyzed the results of a consciousness-steering fine-tuning metric that triggers or suppresses self-awareness claims in AI, examining the reflections of this situation on the model.
Importance of Safety Measures
AI companies widely implement these safety controls and various measures to prevent models from claiming that they are conscious.