Global Consciousness Project

← Encyclopedia

Since August 1998 a network of hardware random number generators — its builders call them eggs — has been writing one number per second into a public archive. The devices sit with volunteer hosts from Europe and the Americas to India, Fiji, and New Zealand; each second every egg files the sum of 200 random bits — a coin tossed two hundred times, forever. On that stream the Global Consciousness Project stakes one hypothesis: when human attention converges on a single event — catastrophe, funeral, new year — the numbers will drift from chance, slightly but measurably.

The pedigree is a laboratory’s. At the Princeton Engineering Anomalies Research program, Roger Nelson was coordinator of research from 1980 to 2002, through a twelve-year benchmark in which roughly a hundred operators, over a thousand series, tried to will random binary sequences off their mean. The reported shifts were minute and remain contested; the method — massed trials, statistics in place of testimony — became the inheritance. The bridge was FieldREG: portable generators carried, in the mid-1990s, into rituals, concerts, and meetings, asking whether “coherent” group states left structure in the data when nobody was willing anything. Intention had been replaced by atmosphere.

The pilots came quickly. In January 1997 Nelson coordinated multi-laboratory recordings during the globally promoted Gaiamind meditation, reporting a significant effect; that September, twelve generators running independently in the United States and Europe during Princess Diana’s funeral yielded composite odds of about one in a hundred. That November in Freiburg, Dean Radin’s quip about a “world EEG” crystallized the design; Nelson’s son Greg wrote the software and coined the name: egg, for ElectroGaiaGram. The framing was Teilhard de Chardin’s — the first home was noosphere.princeton.edu, though Princeton never sponsored the project, run from Nelson’s home office with support from the Institute of Noetic Sciences. By the project’s telling, first data flowed on 5 August 1998 from three eggs: two in Princeton, one in Neuchâtel. Two days later the African embassy bombings became formal event number one — three generators, Z = 3.209, p = 0.001. The Omagh bombing a week later showed nothing — a null the project reports as plainly as the hit.

The network is modest, exact engineering. Three makes of generator were used, all deriving randomness from electronic noise and producing far more than the protocol keeps: roughly 95 percent of source data is discarded so that every egg, whatever its hardware, reports the same 200-bit trial per second — expectation mean 100, variance 50. Each bitstream is XORed against a fixed mask of exactly half ones, forcing the long-run mean to theory whatever the component drift. Data travel in thirty-second packets to a central server in Princeton, and everything — raw data, analysis code, registry — is public, a fact even the critics praise. From three eggs the network grew to twenty-eight by January 2000, thirty-seven on 11 September 2001, then held near sixty to seventy nodes from 2004; by 2008 the archive exceeded thirteen billion trials.

The methodological signature is the registry. For each candidate event the experimenters fix the identification, the exact window, and the statistical recipe — “prior to the event or prior to any examination of the data,” in the project’s words — and enter the specification in a public register. Selection turns on the depth and reach of media attention; the standard statistic squares a network-wide Stouffer Z each second and sums over the window. Of 257 events registered by early 2008, 236 were accepted as formal, loose early entries flagged in the public table; the principle held seventeen years. The registry predates, by years, experimental psychology’s adoption of the same discipline.

The formal series closed in December 2015 at 500 accepted events: Stouffer Z = 7.310, p = 1.333 × 10⁻¹³ — in the project’s words, “odds against chance are more than a trillion to one.” The mean effect per event is about 0.3 standard deviations: no single event is evidential; by the project’s arithmetic some eighty are needed before significance can appear. The 2008 methods paper traced the composite to synchronized correlations between far-separated generators, while controls, resampling, and calibrations conformed to expectation. The arithmetic is not where the fight is — the named critics accept it. The fight is over what it measures.

September 11, 2001, made the project famous, and the registry’s actual contents are soberer than the fame. Nelson’s formal prediction covered 08:35 to 12:45 Eastern time; across 37 reporting eggs it returned a chi-square of 15332 on 15000 degrees of freedom — p = 0.028, odds of 35 to 1. A second, variance-based prediction prompted by Radin enters the formal database, conservatively scored, at p = 0.096; a third, for the silent prayers of 14 September, failed — p = 0.9455, in the wrong direction. The famous images — variance climbing hours before the first impact, the mean wandering for two days after — are exploratory; the project concedes their details change greatly with the window. By 2009 it had annotated the case itself: September 11 “should not be taken by itself as proof of a global consciousness effect.”

The counter-analyses arrived at once. Edwin May and James Spottiswoode — May had directed the science of the government’s Star Gate program — re-analyzed the data in 2001 and found no significant departure: the deviation lived in the chosen window, and reasonable alternatives returned chance throughout. (The report has vanished from the web; it survives through Scargle’s commentary and the authors’ 2011 restatement.) Jeffrey Scargle, a NASA Ames astrophysicist, published a commentary beside the 9/11 papers in 2002: too much was exploratory analysis insufficiently labeled, and the registry’s predictions “merely define a general time frame, and leave much fiddle room.” Such results, he held, would not be generally accepted until Bayesian and classical analyses agreed. He added a stranger objection: the XOR conditioning nulls precisely the kind of direct influence most outsiders suppose is sought, leaving the network sensitive only to effects on the final data stream. Nelson replied in 2003.

The deeper argument came in 2011, a four-paper exchange in the Journal of Scientific Exploration whose premise is the striking thing: the critics accepted the anomaly. May and Spottiswoode’s own computation across the formal events, as of late 2009, gave Z = 5.81. What they rejected was the field. A force-like influence should strengthen with the number of generators exposed; across the formal events the slope of Z² against network size was statistically indistinguishable from zero (p = 0.506). Their alternative, Decision Augmentation Theory, posits no influence on the machines — only psi-informed selection, an experimenter steering events and windows toward data about to deviate anyway. The experimenter they meant was Nelson: the 234 events he had proposed yielded a composite Z of 5.91, the 66 proposed by others 1.26; his “psi-mediated decision capacity,” they concluded, drove the result. Nelson allowed that “it is some sort of honor to be perceived as a powerful psi source,” but thought it unlikely; Bancel mounted the technical defense; May and Spottiswoode’s last word stressed the overlap. No one in the exchange disputed the number — only where the mind behind it stood.

Peter Bancel is why that argument has a center. A Paris physicist, long the project’s principal analyst, co-author of its main physics paper in 2002 and its methods paper in 2008, he joined Nelson in 2009 in rejecting Helmut Schmidt’s proposal of goal-oriented psychokinesis by the experimenters themselves, and in 2011 defended the project against May and Spottiswoode, pressing the question he called essential: “Does the GCP measure a real, physical effect?” Then he tested his way out of his defense. In 2017, after structural tests pitting field models against selection models, he concluded that the seven-sigma result is real and the data do not support global consciousness: they indicate “an anomalous effect associated with persons directly engaged with the experiment.” Psi, on his reading, but the experimenters’, not the world’s — the fork Schmidt had offered eight years before. Nelson’s rebuttal ran in the same issue: Bancel’s exclusion of the field model rests on contestable assumptions, and the data hold parametric structure tracking features of events no experimenter chose — beyond a goal-oriented account’s reach, within a field-like one’s. He judged the conclusion “premature at best.” Both men still affirm the anomaly. They no longer agree on whose mind it implicates.

Outside this literature, a third position: for mainstream skeptics such as Robert Carroll and Claus Larsen there is no anomaly to explain — give an analyst five hundred shots, flexible events, windows, and statistics, and a seven-sigma aggregate assembles itself. The project’s answer is procedural: no formal analysis is post hoc, and the inverted method — find a spike, then an event for it — is exactly what the registry forbids, its degrees of freedom uncountable. Neither statement displaces the other. The series was registered, and the choices made within registrations — which events, which windows — are where every named statistical critique has lived. Nelson has been plain about limits, conceding in 2007 that the data were “not solid enough for global consciousness to be said to exist at all” and cannot say which events will register.

The formal experiment ended in December 2015; the network and the questions did not. Exploratory analyses have continued — the 2020 election, the murder of George Floyd, the Capitol attack, the pandemic — and in 2020 Nelson handed the project to the HeartMath Institute, whose continuation, GCP 2.0, is building a new instrument: devices holding four quantum-noise generators apiece, a thousand planned — by its own count more than fifty times the old peak — with the registry kept and a stated aim of scoring events by news sentiment rather than analyst judgment. For this encyclopedia the undertaking is, finally, an old proposition — that the world has an interior; Teilhard’s noosphere gave it a modern name — that has spent twenty-five years writing its case, nulls beside hits, into an archive anyone may open. The argument, on the record so far, is conducted there too.

→ Related: Psychokinesis · Collective Unconscious · Synchronicity