The question of whether most human thought is a form of hallucination, akin to the confabulations of large language models, is not merely a provocative analogy but a valid and deeply illuminating subject of inquiry. It sits at the intersection of cognitive neuroscience, philosophy of mind, and artificial intelligence research, drawing on a lineage that stretches from Plato’s cave to Immanuel Kant’s distinction between phenomena and noumena, and finding new urgency in the era of generative AI. The validity of the comparison rests on a growing scientific consensus: perception itself is a kind of controlled hallucination, an internal simulation constantly sculpted by sensory feedback. When AI systems generate plausible but fabricated information, they expose in stark, unadorned form the same predictive machinery that, in humans, operates beneath the threshold of conscious awareness. This framing does not trivialise human cognition; rather, it offers a powerful lens through which to explore the mind’s constructive, and occasionally errant, nature.
To understand why this comparison is not only valid but illuminating, one must first grasp the neuroscience of perception. The brain is encased in a dark, silent box—the skull—and has no direct access to the outside world. What it receives are ambiguous, noisy electrical impulses from sensory nerves. To make sense of this barrage, it constantly generates predictions about the causes of these signals, building an internal model of reality. This is the core of the predictive processing framework, championed by neuroscientist Anil Seth, who famously described conscious experience as a “controlled hallucination.” In this view, what we see, hear, and feel is not a faithful readout of the world but the brain’s best guess, a simulation that is continually checked and updated by incoming sensory data. The red of a rose, the warmth of sunlight—these qualities are constructed by our neural circuitry, not found inherently in the physical world.
The hallucinatory nature of perception becomes strikingly evident when sensory input is ambiguous or absent. Optical illusions, such as the Kanizsa triangle where we see contours that do not exist, expose the brain’s compulsive habit of filling in gaps based on prior expectations. More dramatically, individuals who lose their sight can develop Charles Bonnet syndrome, experiencing vivid, complex visual hallucinations as the brain, starved of retinal input, ramps up its internal predictions unchecked. Similarly, phantom limbs demonstrate that the brain’s model of the body can persist and generate painfully real sensations even after the flesh is gone. These are not malfunctions but the normal predictive machinery operating without the usual sensory anchors, laying bare the fact that all perception exists on a continuum with hallucination.
If perception is a controlled hallucination, then memory is a reconstructive confabulation, perhaps even more vulnerable to fabrication. The act of remembering is not like playing back a video recording; it is an act of imaginative reconstruction, piecing together fragments of stored information and filling the inevitable gaps with expectations, schemas, and present knowledge. The pioneering work of psychologist Elizabeth Loftus demonstrated how easily memories can be distorted and even entirely false memories implanted through suggestive questioning. Every time we recall an event, we re-synthesise it, making the memory labile and prone to incorporating new information. In this light, a memory is less a retrieval of a fixed truth and more a hallucination about the past, constrained imperfectly by the faint traces of an original experience, much like an AI summarizing a document it partially “remembers.”
Beyond perception and memory, the human tendency to confabulate extends into our moment-to-moment reasoning and self-justification. In seminal experiments with split-brain patients, neuroscientist Michael Gazzaniga observed that when the non-speaking right hemisphere performed an action cued by a stimulus shown only to it, the speaking left hemisphere would instantly invent a plausible but entirely false reason for the action, all while believing its own explanation. This “interpreter module” phenomenon is echoed in everyday life: studies by Nisbett and Wilson revealed that people often cannot access the true cognitive processes behind their decisions and will confidently fabricate post-hoc rationalisations when asked. We are not just storytellers; we are stories we tell ourselves, and often those stories are elegant, convincing hallucinations of causality.
The mind’s capacity for untethered simulation reaches its apex in imagination, daydreaming, and dreaming. When the brain disengages from external tasks, the default mode network—a constellation of brain regions—springs to life, spinning out scenarios, social narratives, and counterfactual histories. This kind of thought is a pure hallucination, a flight of internal simulation decoupled from the immediate environment. In dreams, particularly during REM sleep, sensory input is functionally blocked and the brain generates a fully immersive, sensorially rich world from the top down, unconstrained by the physical laws of the waking world. Upon awakening, we usually recognise the dream as unreal, but during the experience, it is our reality. This demonstrates that the brain is perfectly capable of generating a seamless, convincing hallucination that masquerades as objective truth.
Even the core experience of being a unified self can be understood as a sophisticated, ongoing hallucination. The sense of a stable “I” that is the continuous author of our thoughts and actions is, on closer inspection, a narrative construct, a “centre of narrative gravity” as philosopher Daniel Dennett put it. Neuropsychiatric conditions reveal how fragile this construct is. In schizophrenia, the sense of agency over one’s own thoughts can break down, leading to the hallucination that thoughts are being inserted by an external force. Alien hand syndrome involves a limb acting with purpose but without the owner’s sense of volition. These cases suggest that the feeling of selfhood is not a given but a delicate prediction, a model of an inner agent that the brain must constantly generate and maintain.
Given this, the parallel between human cognition and AI hallucination becomes both precise and profound. Large language models are predictive engines, trained to forecast the next token in a sequence. They generate text by constructing a probabilistic model of language and the world, and when they lack a grounded reference or encounter a gap in their training data, they do what the human brain does: they fill the gap with the most plausible-sounding prediction, fabricating facts, citations, and historical events with utter confidence. The AI hallucination is not a glitch in an otherwise logic-based system; it is a direct consequence of a predictive architecture operating without robust mechanisms for checking its output against a stable external reality.
The crucial difference, however, lies in the nature of the “control” mechanisms. Human cognition is anchored by a constant, multimodal stream of sensory data and a body that physically interacts with a shared, persistent environment. Our predictive simulations are subjected to ruthless error correction: the hallucination of a solid step is instantly falsified if the foot passes through it. Moreover, we live embedded in a social world that provides consensus reality, constantly challenging individual confabulations. AI, by contrast, lacks an embodied existence, a consistent sensorimotor loop, or a lived history of consequences. It hallucinates in a void, with no pain to teach it the difference between a real flame and a predicted word “flame.” Its reality is purely text, a hall of mirrors with no external ground truth to push back.
Recognizing that much of human thought is a form of hallucination is not a descent into solipsistic despair but a profound recognition of the mind’s creative and generative power. Our “hallucinations” have been shaped by millions of years of evolution to be pragmatically useful, keeping us alive, enabling us to cooperate, and allowing us to imagine futures that do not yet exist. The works of Shakespeare, the theories of Einstein, and the cities we build all began as controlled hallucinations in individual brains, subsequently tested and shaped into shared, enduring realities. The AI, in its raw confabulations, holds up a mirror that strips away our biological grounding, showing us the naked architecture of a prediction machine. By studying the ways in which these artificial minds fabricate, we gain a clearer, humbler understanding of how our own minds weave the tapestry we call reality.