We developed a technique showing that non stationarities in EEG signal carry information about cognition. This technique was successfully tested in two different databases: a working memory database, and an Alzheimer disease database. We also provide evidence suggesting that EEG might not be even piecewise stationary. Therefore, as changes between different stationary regimes are linked to transitions between metastable states in the brain, transitions between those states might not occur in a discrete manner, after a short period of metastability, but rather in a continuous way. Transitions between neighbouring states would occur more often, whereas large transitions occur as well. Large transitions suggesting discreteness had been detected by other techniques, but small fluctuations are not noise, as they can be successfully used to infer aspects of cognition.