Inductive learning system learns classification from training examples and uses induced rules for classifying new instances. If a decision cannot be inferred from system rule base, a default rule is usually applied. No approaches with human interaction exist that would provide model of interactivity appropriate for dealing with non-classifiable instances. In the paper a new interactive approach is proposed where in uncertain conditions interactive inductive learning system can ask for human decision and improve its knowledge base with the rule derived from this decision. Problems and solutions of incorporation of human-made decision into rule base and aspects of choosing between static and incremental learning algorithms are analyzed in the context of proposed approach.