The famous double slit experiment applied to photons neatly highlights the mysterious influence of an observer in quantum mechanics. In the experiment, single photons are fired at a distant screen, partially obstructed midway by a wall containing two slits. If one neglects to check which slit a photon passes through, the photon appears to interfere with itself, suggesting that it behaves as a wave by travelling through both slits at once. But if one monitors the slits carefully (i.e. observes), the interference disappears, and each photon travels through one of the slits as a particle would.

In 1978, however, John Wheeler pointed out that a photon could somehow know in advance whether an observation was going to be made, and change its behaviour to that of a wave or particle accordingly. To test for this possibility he thought of an experiment in which the decision to observe the photons is made only after they have been emitted.

Now, Jean-François Roch and colleagues from the École Normale Supérieure de Cachan have for the first time faithfully realized Wheeler's thought experiment. The team substituted the two slits in Young's apparatus (which would be unfeasible) for two paths in an interferometer (see figure: "Choose the right path"). These paths led directly to two different detectors, allowing one to clearly observe the path each photon took. However, the physicists also devised an automated system that randomly inserted a beam splitter at the last moment. When the beam splitter was in place, it was impossible for an observer to know which path a photon had taken.

Without the beam splitter, the photon took one path or the other, behaving as a particle. But with the beam splitter, the detectors registered interference – as though the photon was behaving as wave and going down both paths simultaneously. However, unlike all previous "two-slit" experiments, the system made the decision to observe after the photon had to commit itself to one path, the other, or both. Therefore if some conceivable source was secretly informing the photon, it would have to be sending a message faster then light – something that relativity forbids.

"Due to this constraint, we can be sure the photon does not know what will be at the end of the interferometer when it enters," Roch said. "This really emphasizes the tension between quantum mechanics and relativity."