Humans and several animal species, including monkeys, rats, and pigeons, are often strongly motivated to seek information about uncertain future rewards. Remarkably, they prefer information even when it has no objective value for controlling the outcome, suggesting that information has subjective value of its own. In recent years there has been an explosion of research on information seeking in humans and animal models. However, a critical unanswered question is whether the computations that assign subjective value to information are conserved between humans and other species. If so, we could leverage animal models to uncover the neuronal populations that are responsible for conserved information value computations and their causal influence on decisions. To address this, we designed analogous multi-attribute information choice tasks for humans and monkeys. Individuals choose between options with multiple attributes, including cues that are either informative or non-informative about future outcomes, and different probability distributions of rewards (money for humans, juice for monkeys). This let us measure and model the subjective value individuals assign to information; how they compute information value using reward uncertainty, expected reward, and other attributes; and integrate information and reward into the total value of an option. We find human and monkey value computations are remarkably similar on all these fronts. We then investigated the neuronal networks responsible for these value computations by recording in monkeys in two interconnected areas with information-related activity, the anterior/ventral pallidum (Pal) and lateral habenula (LHb). Both areas respond to all attributes needed for decisions, but only LHb predominantly integrates information and reward in a manner resembling subjective value. Furthermore, trial-to-trial fluctuations in LHb value signals predict ongoing decisions, while electrical stimulation coincident with LHb value signals modifies ongoing decisions. Our results thus implicate LHb in conserved information value computations that guide online decisions.