The Global Consciousness Project

The project extends the Princeton laboratory's reasoning to a global scale. If a single operator can nudge a random generator, the argument runs, then perhaps the attention of very large numbers of people focused on one event at one moment might nudge many generators at once. To test it, Roger Nelson established a network of hardware random number generators, referred to within the project as eggs, hosted at sites in dozens of countries, each producing a continuous stream of bits and reporting them to a central archive. The network has run without interruption since 1998 and the archive is public. The analysis proceeds by designating an event, specifying a window of time around it, and asking whether the network's combined output during that window departs from expectation more than it should.

Events in the registry have included New Year celebrations, major disasters, funerals watched worldwide and the attacks of September 2001. The project reports that the accumulated departures across the whole event series are large enough to be very unlikely by chance. The hardware itself is unremarkable and that is deliberate: each node is a small commercial noise-based generator, chosen so that the network's behaviour cannot be attributed to any exotic property of one device. The interest is entirely in what happens when many ordinary devices are read together.

Nobody disputes the network or the archive. The dispute concerns the step between them and the conclusion, and it is a clean example of a problem that recurs across this whole field. Three decisions have to be made for each entry in the analysis: whether an event counts as globally attended, when the window of interest starts and stops, and which statistic to compute over it. All three are made by people who have access to the data. With a long enough archive, a large enough space of candidate events and freedom over window boundaries, a body of data with no structure at all will yield a set of apparently significant windows, and no amount of subsequent arithmetic can distinguish that from a real effect. The project maintains that its event registry is fixed before the data are examined in each case.

Critics reply that this cannot be verified from outside and that the historical event list has grown by criteria that were not published in advance. The test that would settle it is straightforward in principle: an independent party defines a list of events and windows prospectively, hands it over, and the analysis runs once. That has not been done, and until it is, the central objection stands unanswered rather than refuted.