This chapter presents an analysis of the distinction between operant (instrumental) and respondent (Pavlovian, classical) conditioning, in terms of a computational, neural-network model. The importance of this distinction cannot be overemphasized, judging by the extensive literature on it (e.g., Asratyan, 1974; Bindra, 1972; Davis & Hurwitz, 1977; Dykman, 1976; Gormezano & Tait, 1976; Gray, 1975; Guthrie, 1935; Hearst, 1975; Henton & Iversen, 1978; Hineline, 1986; Hull, 1943; Kimmel, 1976; Konorski & Miller, 1937a, 1937b; Logan, 1960; Mackintosh & Dickinson, 1979; Miller & Konorski, 1928; Mowrer, 1960; Pear & Eldrige, 1984; Ray & Brown, 1976; Rehfeldt & Hayes, 1998; Rescorla & Solomon, 1967; Schlosberg, 1937; Schoenfeld, 1966; Sheffield, 1965; Skinner, 1935, 1937; Spence, 1956; Trapold & Overmier, 1972). Obviously, this abundance of analyses is too long to summarize in a way that does them justice. Instead, I will begin with how the distinction has been treated in the field of computational modeling of conditioning.
A glance at the field reveals that most models are of respondent conditioning (e.g., Gibbon & Balsam, 1981; Klopf, 1988; Mackintosh, 1975; Pearce & Hall, 1980; Rescorla & Wagner, 1972; Schmajuk & Moore, 1986; Stout & Miller, 2007; Sutton & Barto, 1981; Wagner & Brandon 1989), or operant conditioning (e.g., Dragoi & Staddon, 1999; Killeen, 1994; Machado, 1997), with little if any communication between the two types of models. The field is thus deeply divided. If there is any validity to the motto “United we stand, divided we fall,” the field has fallen long ago.