As AI systems evolve could they really become conscious?

When debates about animal minds, conscious machines, and even fetal awareness spill into public life, the science behind those claims matters as much as the claims themselves. A new analysis argues that the field may still lack a reliable way to tell consciousness apart from ordinary information processing.

That distinction sits at the center of the paper, published in Neuron by a team led by Hakwan Lau of the Institute for Basic Science’s Center for Neuroscience Imaging Research, working with collaborators from the Université de Montréal and New York University. The authors are not arguing that animals, AI systems, fetuses, or organoids are definitely not conscious. Their point is narrower, and more unsettling: today’s methods may not actually measure subjective experience itself.

“Many current theories of consciousness appear to be supported by a range of experimental findings,” Lau said. “But those findings may actually reflect general information processing rather than consciousness itself — so it remains difficult to conclude that these theories truly explain consciousness.”

That challenge matters because strong claims are now being made in several directions at once. Recent years have brought growing discussion of consciousness in mammals and birds, possible sentience in some invertebrates, and speculation about AI agents, embryos, and organoids. The new analysis asks whether the evidence used in those debates is as solid as it appears.

The authors focus on a problem they say has become deeply embedded in consciousness research. Many common experiments compare a condition in which a subject consciously perceives something with a condition in which that same stimulus is not consciously seen. But those same manipulations often also weaken the brain’s overall ability to process the stimulus.

That makes it hard to know what the experiment is really capturing.

The paper points to three widely used approaches: binocular rivalry, visual masking, and perceptual threshold detection. In binocular rivalry, different images are shown to each eye, and awareness appears to flip back and forth between them. In visual masking, a brief image is followed quickly by another image that disrupts awareness of the first. Finally, in threshold tasks, subjects try to detect very faint stimuli near the edge of perception.

In all three cases, the team argues, researchers may be changing more than consciousness. They may also be changing perception, categorization, and the ability to respond meaningfully. If so, brain signals linked to these tasks could reflect general perceptual and cognitive strength rather than the feeling of subjective experience.

The analysis goes further. Some findings often treated as support for consciousness theories may instead reflect what it takes for the brain to function normally. If that is true, the apparent evidence for major theories becomes much less decisive.

The paper places today’s debate in a longer scientific cycle. In the late 19th and early 20th centuries, psychology began as a study of consciousness, often relying on introspection and analogy. Animal behavior resembling human behavior was sometimes treated as evidence of conscious experience.

That approach produced confidence, but not firm grounding. The backlash helped fuel behaviorism, which pushed consciousness to the margins of psychology for decades. John Watson later dismissed the earlier tradition in scathing terms, arguing that psychologists had simply replaced the word “soul” with “consciousness.”

The authors see a cautionary parallel. They suggest that some modern declarations about consciousness may reflect a shift in cultural mood as much as decisive scientific progress. If the field overstates what its methods can establish, it risks repeating an old pattern: bold claims first, scientific backlash later.

That concern extends beyond academic debate. Lau said the stakes are no longer purely theoretical.

“Questions about consciousness increasingly carry ethical and societal implications,” he said. “If scientific claims about consciousness are going to influence discussions about animal welfare, AI ethics, or bioethics, then the scientific foundations supporting those claims must be especially rigorous.”

The paper argues that better clues may come from neuropsychological conditions in which awareness and behavior come apart.

One example is blindsight. These patients, after damage to the primary visual cortex, report no conscious visual experience in part of their field of view. Yet they can sometimes correctly guess whether something is there, and in some cases even avoid obstacles. That suggests meaningful visual processing can survive without reported awareness.

The authors also point to hemispatial neglect, in which patients fail to notice objects on one side yet sometimes behave as if some information from that side still shaped their decisions. Split-brain cases raise similar questions, because perceptual processing can occur in ways that do not line up neatly with verbal report. Anterograde amnesia offers another divide, with patients showing implicit learning despite failing to consciously recall new events.

These cases matter because they show that subjective experience and information processing can sometimes be pulled apart more cleanly than in standard laboratory paradigms.

The paper also notes similar dissociations in the general population. In peripheral vision, people may feel they see more detail than their actual visual processing supports. In aphantasia, some people report little or no vivid mental imagery while still performing tasks that rely on internal visual representations.

The authors are especially concerned about inconsistent standards. A simple photodiode can detect a signal near threshold. Neural networks can be trained to behave flexibly. Yet many people who are open to attributing consciousness to animals are hesitant to extend the same logic to machines.

That does not prove machines are conscious. Instead, it exposes how shaky the criteria may be.

The same problem appears with biological entities. If organoids or very young fetuses show some of the signatures now taken as relevant, would current standards force the conclusion that they might also be conscious? The paper argues that these judgments are hard to defend consistently when the underlying measures are so confounded.

Rather than treating current theories as settled facts, the team says the field should acknowledge how much remains untested. More rigorous work, they argue, may come from studying lesion cases, animal models, electrophysiology, and computational approaches that target mechanisms tied more specifically to subjective experience, not just perception and cognition in general.

The paper does not close the door on consciousness in animals, AI, fetuses, or organoids. It raises the bar for claiming evidence.

That could affect how scientists design experiments, how journals and reviewers judge bold conclusions, and how the public interprets headlines about sentience.

If future policy debates about animal welfare, AI ethics, or bioethics are going to lean on neuroscience, the authors argue, the field first needs methods that can isolate subjective experience with much greater precision.

Research findings are available online in the journal Neuron.

The original story "As AI systems evolve could they really become conscious?" is published in The Brighter Side of News.

Like these kind of feel good stories? Get The Brighter Side of News' newsletter.