The Perturbational Complexity Index Measures Consciousness From the Brain's Response
How do you detect consciousness in someone who cannot tell you about it? Marcello Massimini’s laboratory at the University of Milan built the instrument that defines the field’s answer. The Perturbational Complexity Index (PCI) perturbs the cortex directly with transcranial magnetic stimulation, records the brain’s spatiotemporal EEG response, and compresses that response with an algorithm whose score separates conscious from unconscious states. The index was introduced by Silvia Casali, Angela Casali, and colleagues in Science Translational Medicine in 2013 (DOI 10.1126/scitranslmed.3006294, PMID 23946194), validated independently across the disorders of consciousness in 2016 (DOI 10.1002/ana.24779), and by 2024 it had entered a randomized controlled trial as a biomarker for treatment response (DOI 10.1186/s12984-024-01455-1). For a site that spends its measurement posts on AI indicators, PCI is the biological gold standard those proposals are measured against.
How the Perturbational Complexity Index Works
PCI is a two-part construction. The perturbation is a pulse of TMS that pushes a cortical region out of its current state, which turns the brain into a system responding to a known input without any sensory channel or behavioral report involved. The response is the EEG-recorded pattern of cortical activation as it propagates. The index then applies source modeling to the scalp signal and computes a Lempel-Ziv complexity score on the binarized response, the same compression logic used to measure the structure of a sequence. A complex, differentiated response compresses poorly and scores high. A stereotyped response, whether locally persisting or globally bursting, compresses well and scores low.
The theoretical grounding matters as much as the algorithm. The index operationalizes a specific claim, that conscious states support integrated, differentiated cortical activity while unconscious states collapse into either local isolation or global stereotypy. That claim is a direct descendant of the complexity literature this site tracks in the global workspace ignition threshold analysis, and PCI converts it into a number that requires no cooperation from the subject.
The Validation Record
The 2013 paper reported a clean separation. Across physiological sleep, various anesthetic states, and patients with severe brain injuries, a single threshold on PCI divided conscious from unconscious conditions, with intermediate values in the transition zones. The 2016 Annals of Neurology study by Silvia Casarotto and colleagues then did the work an index must do to be trusted. In a large behavioral and neuroimaging cohort of behaviorally unresponsive patients, PCI stratified unresponsive patients into groups matching covert command-following evidence, showing that a bedside-unresponsive patient with a high PCI had a brain that behaved, under perturbation, like a conscious one. Independent laboratories have since extended the index to rodents (DOI 10.1016/j.isci.2023.106186), and mechanism work showed why unresponsive patients score low, with sleep-like cortical OFF-periods disrupting causal integration (DOI 10.1038/s41467-018-06871-1). The anesthesia side has its own signature literature, including the propofol EEG signatures of Purdon and colleagues (DOI 10.1073/pnas.1221180110), and this site’s DOSE-I sedation dataset analysis covers the transition-annotation effort those signatures depend on.
From Laboratory Index to Treatment Biomarker
The 2024 randomized controlled trial by Xu and colleagues moved PCI out of diagnosis. The trial used PCI to assess responsiveness to repetitive TMS treatment in patients with disorders of consciousness, with the index as a cross-over outcome measure. The bioelectric measurement of consciousness is now a clinical instrument with a therapeutic feedback loop, which is the trajectory this site tracks for AI consciousness measurement in miniature, from proposal to validation to deployment. The broader measurement context is tracked in AI Consciousness in 2026, the state of the field.
Comparison to The Consciousness AI
The Consciousness AI project’s evaluation stance requires that any consciousness-relevant indicator be validated against known cases before it is applied to novel systems. PCI is the strongest existing example of that discipline done right, a theoretically motivated index, validated across states and etiologies, with mechanism studies explaining its failures and a clinical trial using it prospectively. The project’s own research program, which measures consciousness-relevant dynamics in silico, inherits a specific burden from the PCI lineage. Every proposed machine indicator should state its equivalent of the TMS pulse, the EEG response, and the validation cohort. The Bekinschtein clinical biomarker analysis makes the same demand from the clinical side.
What It Changes and What It Leaves Open
PCI changed the ethics and the neurology of unresponsive patients, giving covert consciousness a measurable signal. What it leaves open is theory dependence and scope. The index quantifies complexity of a cortical response, and the Farisco and Changeux analysis of its compatibility with global workspace theory (DOI 10.1093/nc/niad016) shows the theoretical reading is still under construction. It measures states on a human cortical substrate, and nothing about the score itself settles which theory of consciousness is true. For the AI debate, that is the point. PCI is what a mature consciousness indicator looks like, and the field’s machine-side proposals are still far from it.
Sources. Casali et al., Science Translational Medicine 5(198):198ra105, 2013, DOI 10.1126/scitranslmed.3006294. Casarotto et al., Annals of Neurology 80(5):718-729, 2016, DOI 10.1002/ana.24779. Cavelli et al., iScience 26(3):106186, 2023, DOI 10.1016/j.isci.2023.106186. Xu et al., Journal of Neuroengineering and Rehabilitation 21:167, 2024, DOI 10.1186/s12984-024-01455-1.