We present a molecular-level investigation of nitrogen dissociation at high temperature. The computational technique, called direct molecular simulation (DMS), solely relies on an ab initio potential energy surface and both N2+N2 and N+N2 processes are simulated as they concurrently take place in an evolving nonequilibrium gas system. Quasiclassical trajectory calculations (QCT) reveal that dissociation rate coefficients calculated at thermal equilibrium, i.e., assuming Boltzmann energy distributions, are approximately equal (within less than 15%) for both N2+N2 and N+N2 collisions for the range of temperatures considered. The DMS (nonequilibrium) results indicate, however, that the presence of atomic nitrogen significantly affects the dissociation rate of molecular nitrogen, but indirectly. In fact, the presence of atomic nitrogen causes an important reduction of the vibrational relaxation time of N2, by almost one order of magnitude. This, in turn, speeds up the replenishment of high-v states that are otherwise significantly depleted if only N2+N2 collisions are considered. Because of the strong favoring of dissociation from high-v states, this results in dissociation rates that are 2-3 times higher when significant atomic nitrogen is present compared to systems composed of mainly diatomic nitrogen, such as during the initial onset of dissociation. Specifically, we find that exchange events occur frequently during N+N2 collisions and that such exchange collisions constitute an effective mechanism of scrambling the internal energy states, resulting in multiquantum jumps in vibrational energy levels that effectively promote energy transfers. The resulting vibrational relaxation time constant we calculate for N+N2 collisions is significantly lower than the widely used Millikan-White model. Significant discrepancies are found between predictions of the Park two-temperature model (using the Millikan-White vibrational relaxation model) and the DMS results for dissociating nitrogen systems involving both atomic and molecular nitrogen. Such direct comparisons also illustrate how the DMS method is able to reveal all relevant nonequilibrium physics without the need to compute large numbers of state-transition probabilities. In this manner, DMS presents an accurate and tractable approach to construct models for direct-simulation Monte Carlo and computational fluid dynamics simulations from first principles.