Abstract
Theory construction is a process, i.e., a set of state changes by an autonomous agent, or by an organism composed of several autonomous agents. In the first case, we may recall the approach of Pierce (1931) that considers three logical operations (inference rules) on a knowledge base, i.e., a set of propositions asserted to be valid: abduction, deduction, and induction. Abduction generates new hypotheses from which deduction derives predictions to be confirmed by experience. The confirmed hypotheses are structured by induction into laws of general validity. A similar way to describe the life cycle of theory construction within a single agent is to say that the real world asks the agent for a concrete solution in a single instance case, then the solution is abstracted in order to identify laws that are more general ; finally the abstracted solution is applied to other classes of instances of the abstract problem, i.e., it is generalized. The interplay of these operations in one single autonomous (artificial) agent is widely modeled in the work on machine learning. The second social scenario - communicating agents learning by exchanging messages - is less easily formalized but probably more realistic when describing human learning. In this article we give support to the conjecture that the process of construction of knowledge in science (theory construction or scientific discovery) and human learning is an interactive human process of a social nature that presents profound similarities and relations with each other so that we may profit from advances in one domain to infer properties of the other one and the reverse. In this approach we are strongly influenced by constructivism (Piaget 1970) and social constructivism.