A new study indicates that people distinguish between artificial intelligence that acts like a human and genuine human consciousness. Even when AI systems exhibit identical behaviors to humans, participants in the study attributed less consciousness to the machines. However, when the term "awareness" was used instead of "consciousness," the distinction between human and AI perception narrowed considerably.
The research, conducted by scientists at Ludwig Maximilian University of Munich (LMU), explored how individuals perceive the mental states of AI compared to humans. Published in the journal Cognition, the study involved nearly 1,100 participants who evaluated scenarios describing either artificial agents or human characters. These scenarios detailed reactions to external stimuli, such as sounds and the emotional states of others. Participants were asked to rate the level of consciousness and awareness exhibited by the subjects in the scenarios.
The findings revealed a consistent pattern: higher responsiveness in both AI and human characters led to higher ratings of perceived consciousness and awareness. However, a significant gap persisted when participants assessed consciousness. Even the most responsive AI agents were consistently rated as less conscious than the least responsive human characters. This suggests that human-like behavior alone is insufficient to convince people that an AI possesses consciousness.
The study's lead author, Dr. Louis Longin of LMU's Chair of Philosophy of Mind, noted that previous research often asked general questions about AI mental states. This study, however, directly compared how people attribute the same mental states to AI and humans under identical behavioral conditions. "We found something much more nuanced," Longin stated. "People are quite willing to say that an AI notices things and is aware of its surroundings. But they clearly draw a line when it comes to the term 'consciousness'."
The researchers posit that the specific language used to describe artificial agents plays a substantial role in shaping these perceptions. When participants were asked about "awareness," their judgments of AI and humans became much more similar. This suggests that while people may be willing to grant AI a form of awareness, the term "consciousness" carries a deeper, uniquely human connotation that they are reluctant to extend to machines.
Professor Ophelia Deroy, a senior author of the study, commented that people might use mental terms for AI because they lack better words, especially since AI is trained to act and speak like humans. This could lead to descriptions that seem natural but do not reflect a belief in genuine AI consciousness. The study's methodology aimed to disentangle judgments about observable behavior from judgments about internal mental states, a critical distinction in the ongoing debate about artificial intelligence and its capabilities.
The research aligns with broader discussions in the scientific community regarding the definition and detection of consciousness, particularly in the context of advancing AI. Some researchers argue that current AI systems, including large language models, are not conscious, even if they can mimic conscious behavior. They emphasize that understanding the internal mechanisms, rather than just outward behavior, is key to assessing consciousness. The LMU study contributes to this field by highlighting the psychological barriers humans erect when considering AI consciousness, even when presented with indistinguishable behavior.