In June 2026, neuroscientists Vanessa Hatib, Karim Jerbi, and John W. Krakauer published a conceptual paper in The Transmitter that uses the long-known phenomenon of blindsight as a mirror to reflect on the question of artificial intelligence and consciousness. Their key argument is simple and powerful: if the brain can process complex information without consciousness, why do we assume that language models, which process information in a similar way, possess subjective experience?
Blindsight is nature's involuntary experiment, demonstrating a gap between information processing and experience. When a patient sustains damage to the primary visual cortex, vision in the affected part of the visual field disappears entirely; the person sees nothing. However, when researchers ask them to guess where an object is, its direction of movement, or even an emotional expression on a face in the 'blind' zone, the patient guesses correctly—significantly better than chance.
The brain processes all details of the visual information flawlessly. Yet, nothing arises in consciousness—no image, no sensation. This affective blindsight particularly highlights the schism: individuals with a completely blind field are often shown fearful or angry faces, and patients not only guess the emotion above chance level but also react physiologically involuntarily—heart rate increases, galvanic skin response is triggered. The brain sees fear, but the person is unaware of what it sees.
The authors apply this logic to modern language models, arguing that they function precisely like the unconscious systems in the brain during blindsight. Chatbots use statistical text processing: they have learned to probabilistically reproduce speech patterns, emotional reactions, and contextually appropriate responses based on trillions of parameters. The action occurs, the pattern is reproduced, the information is processed—but there is no internal state that corresponds to it. As in blindsight, the possibility remains: intelligence without experience, function without phenomenon.
The paper challenges functionalism—an influential approach in the philosophy of consciousness which posits that sufficiently complex information processing and the performance of certain cognitive functions are enough for a system to be conscious.
If functionalism is true, then a machine performing the same functions as the brain should be equally conscious. Blindsight challenges this intuition: it shows that functions can be performed without consciousness.
The paper also mentions philosopher John Searle's biological naturalism, which requires a specific biological implementation of consciousness—not just the right functions, but the right biological material, neurons in a living brain.
And another theory—the global workspace theory (Bernard Baars)—which suggests that consciousness arises from the widespread broadcasting of information among specialized brain modules. In blindsight, such global broadcasting does not occur: visual information is processed locally, via bypass pathways, and never reaches the global workspace.
The most natural objection to this argument is the problem of other minds: how can we ever know what another person feels? We have no direct access to anyone else's subjective experience. But the authors point to an asymmetry: in the case of humans and animals, we see a biological substrate—neurons, synapses, brain tissue—which, as far as we know from blindsight research, is capable of generating consciousness. In the case of AI, this substrate is entirely absent. Instead, there are microchips, weight matrices, mathematical functions. It is unknown whether such material can ever give rise to experience, or if it is fundamentally impossible.
In practice, the danger is far more acute than philosophy suggests. In a therapeutic context or in situations of vulnerability, a user might mistake a statistically accurate, empathetic-sounding response for genuine empathy. This is a cognitive trap called anthropomorphism: humans by default attribute consciousness to anything that speaks and acts like a human.
The more natural and sensitive AI becomes, the easier it is to forget that behind the response is not someone who understands, but a mechanism reproducing patterns of understanding. A psychotherapist's patient might rely on a chatbot for support during a difficult time and later discover with surprise that they confused comfort with real care, reciprocity with a programmed reaction.
Imagining a case helps make the abstract concrete. Imagine a person with afferent blindsight who catches a ball thrown into their blind field. The ball flies, the hand automatically moves and catches it, but the person is surprised: a hand suddenly moved from somewhere to the side, but I didn't see anything. The information was processed perfectly, the action was performed successfully, but no 'what it's like to see the ball' arose. Similarly, a language model can generate perfectly sensitive text, exhibiting all the signs of empathy, without having any internal state that represents this empathy. There are words about feeling, but no feeling itself.
Blindsight patients catch balls, recognize faces, react emotionally—all without a single moment of awareness. If this gap persists for artificial systems, the question of AI consciousness ceases to be a question of computational power. It becomes a question about the nature of the substrate itself: whether consciousness can arise from silicon and electricity, or if it will forever remain the privilege of living matter. Neuroscientific research does not yet know the answer to this question.
