A constitutive model that represents soil behavior under wide ranging loading conditions is needed in the simulation of complex boundary value problems. However, most laboratory tests provide information on stresses and strains under very limited loading conditions which are interpreted assuming uniform stresses and strains within the tested soil specimen. Among these tests, the Direct Simple Shear (DSS) test experiences non-uniform stress-strain distribution within the test specimen during simple shear. Nevertheless, there have been few studies on the non-uniform states of stress and strain within the specimen, and a relatively simplified interpretation of stresses and strains is used. This paper utilizes an evolutionary inverse analysis approach, referred to as self-learning simulations (SelfSim), to extract and interpret the non-uniform stresses and strains within an undrained DSS test specimen. SelfSim uses boundary measurements of forces and displacements, dual numerical force and displacement controlled simulations and an evolutionary material model, a neural network based material model, to extract the non-uniform states of stresses and strains throughout the specimen. The proposed approach is first demonstrated using simulated DSS tests with a known material constitutive model, the Modified Cam-clay model. The result shows that SelfSim successfully extracts the non-linear stress-strain behavior experienced throughout the specimen. The approach is then applied to a DSS test on K0 normally consolidated Boston Blue Clay (BBC) specimen. Preliminary results of extracted stress-strain soil behavior are presented and discussed. The paper demonstrates the potential of the proposed approach for developing a new way to interpret soil behavior under general loading conditions from laboratory tests, required to solve geotechnical boundary value problems.