“An Information Integration Theory of Consciousness,” published by the neuroscientist Giulio Tononi in BMC Neuroscience in 2004, is the founding paper of what became known as Integrated Information Theory, or IIT. Tononi proposed that consciousness corresponds to a system’s capacity to integrate information - to bind many distinctions together into a single, unified state that cannot be broken into independent parts without losing something. He introduced a quantity, phi, intended to measure how much integrated information a system generates.
The theory tries to answer why some physical systems are conscious and others are not. Tononi’s recurring contrast is a digital camera sensor with a million photodiodes: it can register an enormous number of distinct images, but each diode acts independently, so the system integrates no information and, on this account, has no experience. A brain, by contrast, integrates information across densely interconnected regions, producing a unified experience that is simultaneously highly differentiated. Consciousness, in IIT, is graded rather than all-or-nothing, and is a property of how a system is causally organized, not of what it is made from.
IIT is one of the leading scientific theories of consciousness, and it carries striking and controversial implications for AI. Because it ties consciousness to a specific kind of integrated causal structure rather than to intelligent behavior, IIT implies that a system could be highly intelligent yet have very low phi - and therefore little or no consciousness - if its architecture is feed-forward and modular, as many current AI systems largely are. That makes IIT a key reference point in expert reports on whether machines could be conscious.
Why a general reader should care: IIT is one of the few frameworks that offers a concrete, in-principle measurable criterion for machine consciousness, which is becoming a practical question as AI systems grow more sophisticated and more humanlike.