Interactive identification of systems is the subject of this ambiguously titled work. The author treats the following “system identification” problem: given a physical system X and a set M of candidate models, find the model m ∈ M that is the best model of X, or a satisfactory model of X, given some criteria for “best” or “satisfactory for some purpose.” System identification is distinguished from modeling as follows:

Identification means finding a mathematical model of a physical object, given a class of tentative models and given a set of response data from experiments. Modeling means constructing a class of tentative models, based on any source of information (possibly including experimental data).

The author notes that for some classes of models, interactive computer programs exist that identify the best model in the class, given a set of stimuli and responses. No such programs are identified, described, or evaluated; rather, general principles for guidance in the selection and use of such tools are developed.

After an introduction defining the identification problem, the book reviews probability, and especially Bayesian  theory,  as applied to modeling problems. Then experiments are discussed, and the concepts of sufficient stimulation, reproducibility, separation of experiment from experimenter, and representability of systems are covered. Next, the author returns to the identification problem and treats model validation and falsification. A pitfall is defined as a case in which an identification activity produces an incorrect model without identifying it as incorrect. Pitfalls are attributable to incorrect assumptions or insufficiently tested hypotheses. Bohlin presents the relations among these concepts in set-theoretic language and illustrates them using numerous Venn diagrams. Later chapters discuss various modeling and identification issues in more detail, leading to a proposed structure for an interactive identification program supporting a wide variety of model types, hypotheses, and constraints.

The treatment is highly theoretical throughout, making extensive use of probabilistic and statistical methods. The reader should have significant experience in stochastic modeling of dynamical systems. The numerous examples are general and abstract. Some concrete numerical examples would have been helpful.