When we think about the stochastic behaviors of cells, we can identify a few reasonably obvious sources of noise. First, there are generally only one or two copies of a particular gene within a cell, and the probability that those genes are in the state to be able to recruit and activate RNA polymerase is determined by the frequency of productive collisions between regulatory sequences and relevant transcription factors together with the dissociation rate. Cells are small, and the numbers of different transcription factors can vary quite dramatically. Some transcription factors are present in high numbers (~250,000 per cell) while others (like the lac repressor) may be present in less than 10 copies per cell. The probability that particular molecules interact will be controlled by diffusion, binding, and kinetic energies. This will influence the probability that a particular gene regulated by a particular transcription factor is active or not. Once on, transcriptional and translational bursting will produce gene products that can alter the state of the cell such that secondary, down-stream changes occur in gene expression. and other cellular processes. These changes may (like the lac operon-lactose system) be reversible once the stimulus (lactose) is removed or they many be more or less irreversible, as occurs during cellular differentiation and embryonic development301.
301 A single molecule view of gene expression: http://www.ncbi.nlm.nih.gov/pubmed/19819144