The evaluation of the preferences based utility function is a goal of the human cantered control (management) design.The achievement of this goal depends on the determination and on the presentation of the requirements, characteristics and preferences of the human behaviour in the appropriate environment (management, control or administration of complex processes). The decision making theory, the utility and the probability theory are a possible approach under consideration. This paper presents an approach to evaluation of human’s preferences and their utilization in complex problems.The stochastic approximation is a possible resolution to the problem under consideration. The stochastic evaluation bases on mathematically formulated axiomatic principles and stochastic procedures. The uncertainty of the human preferences is eliminated as typically for the stochastic programming. The evaluation is preferences-oriented machine learning with restriction of the “certainty effect and probability distortion” of the utility assessment. The mathematical formulations presented here serve as basis of tools development. The utility and value evaluation leads to the development of preferences-based decision support in machine learning environments and iterative control design in complex problems.