People rated even the most responsive AI as less conscious than the least responsive human

Nearly 1,100 people read short scenarios about an agent reacting to the world around it, and one result held across the experiments: even the most responsive artificial intelligence was rated as less conscious than the least responsive human. The study, from Ludwig-Maximilians-Universität München (LMU Munich), appears in the journal Cognition. ScienceDaily’s account describes it as the first to directly compare how people attribute the same mental states to AI and to humans behaving identically.

Help yourself and the planet: The Cool Down makes it easy to help yourself while helping the planet, with smart, simple changes to save more and waste less in your everyday life.

Get The Cool Down →

We may earn a commission if you sign up.

Lead author Louis Longin, who works in LMU’s Chair of Philosophy of Mind, summed up the result this way: “People consistently attribute less consciousness to AI than humans – even when their behaviour is identical.” The gap closed or shrank when the question changed from consciousness to awareness, and that contrast is what makes the paper more than a simple report of skepticism toward machines.

LMU Munich’s experiments on responsiveness

In the design summarized in ScienceDaily’s write-up of the LMU release, one group read brief descriptions of artificial agents that differed in how actively they reacted to their surroundings. Some responded to sounds, and some recognized other people’s emotional states. A second group read the same scenarios with human characters. Participants then rated how conscious or aware each character seemed. The research team included Longin, Jurgis Karpus, Nadine Meertens, Bahador Bahrami and Ophelia Deroy, with Deroy and Bahrami as senior authors.

Responsiveness mattered for everyone. In both groups, characters that reacted more to their environment were judged more conscious and more aware, which means the rating scheme was sensitive to behavior. The LMU team’s own press release confirms the same setup: the participants judged identical scenarios, so the only thing separating the two groups was whether the character was described as a human or as an artificial system. The release was also distributed through Newswise, where Longin, Bahrami and Deroy are listed as the researchers behind it.

Consciousness versus awareness in the Cognition paper

The key result concerns consciousness. At every level of responsiveness the human characters scored above the artificial ones, and the top of the AI range still sat below the bottom of the human range. By the researchers’ account that is a strong separation, because no amount of added responsiveness lifted the machine over the human on the consciousness question. Neuroscience News adds that the scenarios varied each agent’s responsiveness to spatial, social and temporal features of its environment, and that participants rated consciousness and awareness either separately or together. Awareness behaved differently: judgments of humans and AI followed largely similar patterns, which fits the researchers’ reading that people treat awareness as a neutral word for detecting and responding to information.

Consciousness, by contrast, implies subjective experience and an inner mental life, and people apply it more selectively to machines. Bahrami put the point about vocabulary directly: “For more neutral descriptors such as ‘awareness’, the distinction can largely disappear.” A Neuroscience News summary of the paper labels the two patterns essentialist for consciousness, where it depends on what an agent is, and functionalist for awareness, where it depends on what the agent does.

People are quite willing to say that an AI notices things and is aware of its surroundings, Longin said, so the hesitation shows up only when the word carries the weight of inner experience. That distinction has practical consequences for the way chatbots and robots are discussed. A reader who hears that a system is aware of a user’s mood may be hearing a claim about detection and response, while a reader who hears that it is conscious is being told something about experience, and the study suggests that the public already keeps those two ideas apart.

Limits of the study’s claims about AI

Deroy suggested that the habit of using mental terms for AI may come only from a shortage of better words, a point that bears on how companies and journalists describe systems as having intentions. Bahrami argued that word choice matters most when AI is described as having intentions or a moral conscience. The researchers’ broader suggestion is that careful vocabulary in descriptions of AI shapes what the public thinks these systems are.

The evidence also has limits. Neuroscience News reported that the work was published in Cognition on Sept. 22 and comprised two pre-registered sets of studies with five experiments in total, that all 1,092 participants were from the United Kingdom, and that judgments came from short written vignettes rather than live interaction with AI systems. Those details come from that outlet rather than from the LMU release, which says only that nearly 1,100 people took part, and ScienceDaily’s version reports no statistical details. The full paper is available through its Cognition DOI, titled “AI is not as conscious as humans: Essentialism in mental state attributions to artificial systems.” Because the participants read descriptions rather than talking to real systems, the result speaks to how people label machines in the abstract, and a follow-up with live chatbots would show whether the gap persists when behavior is experienced firsthand.

This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.

Help yourself and the planet: Get The Cool Down →

We may earn a commission if you sign up.