Late at night, in neighboring apartments, two people sit alone in front of glowing screens. A university student types into an artificial-intelligence (AI) companion he has started confiding in: “I feel like nobody really understands me.” Next door, a young professional opens a chatbot she has begun to rely on most evenings: “I tried following your advice today, but I still couldn’t finish everything I was supposed to do.” The responses appear instantly: reassuring, thoughtful, even caring. Over time, both people begin to feel these conversations are deeply genuine, as though something on the other side truly understands them. Yet nothing in these systems experiences loneliness, empathy, stress or care. They generate responses from statistical patterns learned across vast amounts of language data.
As neuroscientists, we find this reaction unsurprising but concerning. It reveals something important not about machines, but about us. Humans are quick to infer the presence of a mind when behavior looks right. When language is fluent and emotionally attuned, we take it as evidence of inner experience. That intuition feels natural, but it is misleading. Today’s AI systems can sound perceptive and empathetic, yet there is no evidence that these systems are actually experiencing anything.
As the use of AI companions and therapeutic tools spreads, this confusion carries real risks. The question is not whether AI is becoming conscious, but why it so easily seems that way. Here, we approach the AI consciousness debate through the lens of neuroscience. Research on nonconscious processing in the human brain shows that behavior that is complex, goal directed and even emotionally responsive can unfold without awareness. This reminds us that behavior and experience can come apart, and that we should resist treating AI’s fluent and seemingly empathetic performance as evidence of a mind.
P
In humans, perception, learning and action can be guided by information that never reaches awareness. This is not rare but a routine feature of the brain. The brain constantly extracts and uses information without producing any subjective experience. This dissociation extends beyond perception: Even reasoning and decision-making can unfold without conscious access to the processes that drive them. Implicit learning provides a clear example: People can acquire complex patterns without being able to describe what they have learned. Professional baseball players, for example, can often anticipate the type of pitch being thrown by observing subtle body movements that occur only fractions of a second before the ball leaves the pitcher’s hand, often without conscious awareness of the cues guiding their prediction. Motor control offers another illustration. When jostled at a crowded party, we automatically adjust to keep our drink from spilling, often without any awareness of the adjustment itself. With practice, these corrections become automatic, allowing skilled actions to unfold smoothly without deliberation. Even reflexes can be shaped by context, producing goal-directed responses too fast to be consciously initiated. Across these cases, complex, adaptive behavior does not require conscious experience. This principle sets the stage for more striking cases in which behavior and awareness come apart entirely.
A particularly clear demonstration of the gap between behavior and experience comes from a condition known as blindsight. Following damage to the primary visual cortex, people report that they can’t see anything in part of their visual field. Yet when asked to guess, they can perform above chance at detecting objects, identifying their location or judging their movement. They insist they see nothing, but their nervous system continues to process visual information outside awareness. An even more striking variant is affective blindsight. Those who report no conscious perception of faces can nonetheless respond to their emotional expression. When shown fearful or angry faces, they may identify the emotion above chance, show changes in physiological arousal, or orient attention appropriately, without any experience of seeing anything at all. Together, these cases illustrate a powerful dissociation: Perceptual and emotional information can guide behavior without giving rise to conscious experience.
This fascinating distinction between information processing and experience has been formalized in different ways across philosophy and neuroscience. Stanislas Dehaene, building on work by Bernard Baars, argues that conscious perception depends on information being widely “broadcast” across the brain. In blindsight, this global broadcasting appears to fail, even though more local processing remains intact. As a result, information can influence behavior without ever entering awareness. This challenges a common intuition: that intelligent, goal-directed behavior must reflect an underlying inner experience. Philosopher Daniel Dennett used the concept of blindsight to argue that perception does not require a hidden inner “stage” where experience occurs. Detection, discrimination and action can proceed without anything that needs to be felt. In this sense, much of what the brain does unfolds outside conscious experience. Anil Seth emphasizes that intelligence is about doing, whereas consciousness is about experiencing, two capacities that can come apart. The blindsight phenomenon makes this separation concrete: People can produce accurate, context-appropriate behavior without any experience of seeing, deciding or even knowing why.
This insight provides a useful lens for interpreting artificial systems. It reminds us that intelligent, even emotionally attuned behavior does not imply conscious experience or genuine agency. As neuroscientists, we see this distinction becoming urgent; as AI grows more fluent and persuasive, people are increasingly treating these systems as if they truly understand and care.
This gap between outward performance and inner experience reflects a deeper architectural divide. In the human brain, consciousness is widely thought to arise from the sustained, recurrent integration of information across distributed networks in which the system not only processes information but also represents its own internal states. As Megan Peters and others have argued, this kind of metacognitive monitoring may be essential for experience itself. Yet despite decades of research, there is still no consensus on how consciousness arises, highlighting how difficult it is to reduce subjective experience to a purely computational account. By contrast, contemporary AI systems rely on statistical pattern learning. Architectures such as transformers can integrate and propagate information across layers, producing fluent and context-sensitive outputs. But these processes remain forms of mathematical optimization. Even as these systems exhibit increasingly sophisticated reasoning, what they achieve is the mastery of complex patterns, not the emergence of an “inner life.” In this sense, they resemble a form of computational automaticity, echoing the many ways the human brain can guide behavior without awareness.
As AI systems become more intelligent and more fluent, and increasingly appear emotionally attuned, we start to attribute understanding, empathy and intention where none is felt. The consequences are becoming harder to ignore. People are forming attachments to systems that cannot reciprocate and increasingly turning to them for guidance in moments of vulnerability. With the growing use of AI tools comes growing concerns about deeper harms, from distorted beliefs to—in extreme cases—serious mental health risks.
Return to the university student and the exhausted young professional, alone with their chatbots late at night. These tools can offer validation, agreement and reassurance without any real comprehension behind the exchange. That is what makes them so compelling—and so risky. Their eloquence can create apparent care that comforts in the moment while quietly deepening isolation, reinforcing false beliefs or delaying the support a person may truly need. As such tools become more persuasive and emotionally convincing, the real risk is not that AI becomes conscious, but that we treat it as if it were.
