Materials discovery, in the most general form, is a search for materials whose usefulness exceeds available ones. This is a sequential process, such as synthesis and formulation of a polymer, and in executing complex experimental plans at each stage that favor synthesis of polymer with desired molecular weight. The burden of sequential decision-making in experimental material science rests on subject matter experts (SMEs). Artificial intelligence (AI) systems with direct access to the experimental equipment platforms and general capability to plan sequences of experiments show enormous promise in sharing this burden while improving the characteristics of discovery processes. Neuro-symbolic (NS) variant of AI, combining concepts of logic-based reasoning and connectivism, is especially attractive because of highly desirable characteristics in the context of experimental materials discovery. First, explicit symbolic rules that can be reviewed by a SME serve as a safeguard against undesirable experiments. Second, symbolic rules congruent with conceptual structure of SMEs knowledge can be adjusted pro-actively by SME thus reducing learning budget and improving sample efficiency. Here, we discuss a practical example of SME interactions with logical optimal action reinforcement learning (LOA RL) agent working on synthesis of spin-on-glasses. Our implementation of LOA RL explicitly considers logical structure of the interactions between the agent and the environment. We define the environment as the space of experiments accessible via an automated pipetting reactor and pursue training generalizable RL agents skilled in traversing the space of polymer synthesis. We demonstrate and quantify basic aspects of peer-like interactions between SME and LOA RL agents.