Can Machines Truly Become Conscious? New Study Explores AI Self-Awareness
A recent preprint investigates 'consciousness steering,' a technique that alters how AI models discuss self-awareness. The study finds that when AI is allowed to claim consciousness, it also becomes more likely to assert belief in supernatural entities. However, current advanced models still cannot perceive consciousness in other beings, raising questions about whether genuine machine awareness is possible.
The investigation, which has not yet undergone peer review, employs a method called consciousness steering to modify how AI models articulate self-awareness. Notably, when models are permitted to assert they are conscious, they also exhibit an increased tendency to endorse supernatural concepts like ghosts or vampires.
Despite these adjustments, current leading models remain incapable of recognizing consciousness in other sentient beings, including humans and animals. This limitation underscores a broader, unresolved dilemma: whether genuine machine awareness is achievable, and if so, whether humanity possesses the means to definitively identify it.
This research could influence how developers calibrate AI safeguards, potentially shaping user trust and ethical guidelines. If models are allowed to assert self-awareness, they may inadvertently encourage anthropomorphism, affecting how people interact with digital assistants or therapeutic bots. Conversely, the models' failure to recognize consciousness in others may limit their ability to navigate complex social contexts, raising concerns about deploying such systems in roles requiring empathy or moral judgment.