As artificial intelligence becomes more sophisticated, reports of AI systems misleading users, collaborating, or appearing to pursue independent goals are becoming increasingly common. These behaviors can lead people to describe AI as if it could think, feel, and make decisions for itself.

But does human-like behavior convince people that an AI system has a conscious mind? New research from LMU suggests that it does not. Even when humans and AI behave in exactly the same way, people distinguish between displaying intelligent behavior and possessing consciousness.

How researchers tested perceptions of AI consciousness

Published in the journal Cognition, the study was led by Dr. Luis Longin, professor of philosophy of mind at LMU, in collaboration with Dr. Bahadur Bahrami, Professor Ophelia Deroy, and other researchers.

“Previous studies have commonly asked the question of whether AIs actually have mental states, but ours is the first to directly compare how humans attribute the same mental states to AIs, and how humans behave in exactly the same way under the same circumstances,” Longin said.

“We may be using spiritual terms for AI, but it may just be because we don’t have a better word,” said Deroy, a philosopher and senior author of the study. “After all, AI is trained to act and speak like humans, so these explanations seem natural.”

To investigate how people judge consciousness, the researchers conducted an experiment involving about 1,100 participants. One group read brief descriptions of artificial agents displaying different levels of responsiveness to their environments. For example, some agents responded to sounds or recognized the emotional states of other agents.

Another group encountered the same situations, but the characters were described as humans rather than AI systems. Participants then rated how aware each character appeared to be of the surrounding environment.

The surprising gap between human and AI consciousness

The experiment found that reactivity influenced how participants judged both humans and artificial intelligence. In general, characters that responded more positively to their surroundings were viewed as more cognitively capable and more aware.

However, a significant difference emerged when participants specifically assessed consciousness. Even the most responsive AI was rated as less conscious than the least responsive human. In other words, identical behavior did not lead participants to assign humans and machines the same level of consciousness.

When participants assessed cognition instead, the results changed dramatically. Judgments of human and AI cognition followed roughly similar patterns, suggesting that people interpret cognition and consciousness in fundamentally different ways.

“There’s a growing concern in the public debate that people will see AI behaving like humans and start treating these systems as if they had human-like mental states,” said Longin, the study’s lead author. “We found something more nuanced. People are very happy to say that AI notices things and is aware of its surroundings. But when it comes to the word ‘consciousness,’ they clearly draw the line. People consistently think that AI is less conscious than humans, even if the AI’s behavior is the same.”

Although people may describe AI as conscious when discussing its behavior, the findings indicate that they still distinguish artificial systems from the human mind.

Why the language used to describe AI matters

Researchers believe the difference between cognition and consciousness offers important insight into how people understand artificial intelligence.

Cognition is a relatively neutral term that can describe an agent’s ability to detect and respond to information in its environment. Consciousness, by contrast, carries deeper implications about subjective experience and what it means to have an inner mental life.

Because cognition focuses on observable behavior, people appear comfortable applying the term to both humans and AI. Consciousness has more complex meanings, and participants were less likely to equate it with human-like actions in artificial systems.

The findings also show why language matters when discussing AI, particularly as systems become increasingly capable of producing persuasive human behavior.

“Our study provides lessons for how to communicate about AI,” said Bahadur Bahrami, senior author of the study and an expert in social neuroscience. “Companies and journalists often reach for exaggerated explanations of AI models, attributing them to intent, planning, or even moral conscience or hesitation. What our research shows is that everyday judgments are more nuanced. Humans and AI are not always placed on the same mental scale. The distinction also depends on which term we use. For more neutral descriptors, such as ‘cognitive,’ the difference can be almost nonexistent, so our choice of language matters.”