Human learning optimization (HLO) is a novel inborn binary meta-heuristic algorithm inspired by a simple human learning model. Although HLO, as a binary algorithm, can directly solve discrete (integer) problems, the efficiency may be significantly spoiled for the large-scale discrete problems due to the curse of dimension. Therefore, an adaptive discrete human learning optimization for TSP-like problems (ADHLOT) is proposed in which the random learning operator, individual learning operator, social learning operator and the re-learning operation of HLO are extended or re-designed to improve the efficiency of the algorithm for TSP-like problems. Furthermore, an adaptive strategy is used in ADHLOT to dynamically adjust the social learning rate for achieving a better balance between exploration and exploitation. Finally, ADHLOT is evaluated by solving TSP benchmark problems and compared with recent meta-heuristic algorithms. The experimental results show that the ADHLOT has obvious advantages and shows promising potential for applications.