People see AI as aware but not conscious: Study

People remain reluctant to regard artificial intelligence (AI) as conscious even when it behaves exactly like a human, according to a new study by researchers at Ludwig Maximilian University of Munich (LMU).

Published in the journal Cognition, the study found that people distinguish between an AI system’s ability to respond intelligently to its surroundings and the possession of consciousness, even when humans and machines display identical behaviour.

The research was led by Dr Louis Longin of LMU’s Chair of Philosophy of Mind, alongside Dr Bahador Bahrami, Professor Ophelia Deroy and other researchers.

The team conducted experiments involving nearly 1,100 participants to examine how people attribute consciousness and awareness to humans and artificial agents.

One group read descriptions of AI systems displaying different levels of responsiveness to their environments, such as reacting to sounds or recognising other people’s emotional states. Another group read identical scenarios in which the characters were described as humans.

Participants then assessed how conscious or aware each character appeared to be.

The results showed that greater responsiveness generally led participants to perceive both humans and AI as more aware and conscious. However, a substantial gap emerged when they specifically evaluated consciousness.

Even the most responsive AI systems were rated as less conscious than the least responsive humans, despite their behaviour being identical.

Judgments of awareness, by contrast, followed broadly similar patterns for humans and AI, suggesting that people understand the two concepts differently.

“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’,” Longin said.

The researchers suggested that awareness is generally understood as the ability to detect and respond to information in the surrounding environment, while consciousness carries deeper implications about subjective experience and an inner mental life.

This distinction may explain why people readily describe AI systems as aware without necessarily believing they possess human-like consciousness.

Senior author Ophelia Deroy said people often use mental terms to describe AI because the systems are trained to act and communicate like humans, making such descriptions seem natural.

The findings also have implications for how AI is discussed in public, particularly as systems become increasingly capable of producing convincing human-like behaviour.

Bahador Bahrami, a senior author of the study and an expert in social neuroscience, said companies and journalists often describe AI models in terms that suggest intentions, plans or moral judgment.

The study indicates that people’s perceptions are more nuanced and depend partly on the language used to describe AI, he said.

For relatively neutral terms such as “awareness”, the distinction between humans and AI can largely disappear, whereas the term “consciousness” leads people to draw a clearer boundary.

The researchers said the findings highlight the importance of precise language when describing increasingly sophisticated AI systems and interpreting their behaviour.