N ico O rlandi Interestingly, and despite the dissimilarities between indirect realism and constructivism, some of the considerations that are sometimes brought up in support of the inferential idea are parallel to some of the considerations introduced to suppose that our perceptual access to ordinary objects is mediated The ability to explain illusion and misperception, for example, is mentioned as one advantage of constructivism in a way that is reminiscent of the argument from illusion (Fodor and Pylyshyn 1981; McClamrock 1995) Because of its use of the notion of representation, constructivism has a Kantian flavor and, although introduced by Helmholtz in the 19th century, it gained popularity only in the second half of the 20th century (Gregory 1970; Hochberg 1964; Rock 1983) This was partly due to two important, and related, developments One was the introduction of computers as models and simulators of mental processes (Newell and Simon 1963; Turing 1950) The other was the idea that the brain is an information processing system (Broadbent 1958; Neisser 1967; von Neumann 1951) Both developments shaped the progress of perceptual science and have been influential in philosophy of mind Constructivism weds well with a computational and information-processing model of mental activity because computations are traditionally understood as operations on symbols in virtue of rules or algorithms (Fodor 1975; Ullman 1980) If we recognize symbols as information-carrying structures, and algorithms as containing assumptions and rules to process the symbols, then viewing the perceptual process as computational pretty much amounts to viewing it the way the constructivist does – as an inference from some informational states to others Partly because of its fit with computer science, constructivism has been widely accepted in philosophy since the 1950s (Marr 1982; Rock 1983; 1997; Ullman 1980; Palmer 1999; Pylyshyn 1984; Fodor 1984; Churchland 1988) Philosophers who not agree on much else – for example, Jerry Fodor and Paul Churchland – tend to accept it Derivatively, the notions of representation, computation and information have received extensive scrutiny in philosophy.15 The most current version of constructivism is the Bayesian model of perception (Brainard 2009; Clark 2013; Maloney and Mamassian 2009; Mamassian et al 2002; Rescorla forthcoming) According to such model, the visual system performs inductive inferences on some hypotheses about the environment These hypotheses are initially selected based on prior experience The hypotheses are tested in real time given sensory stimulation and either confirmed or changed Like Gestalt psychology, constructivism prompted reflection on the neutrality of perceptual appearances According to constructivism, what we perceive depends on unconscious inferences that employ assumptions about the world If the assumptions change – through, for example, the acquisition of new knowledge – then it seems that what we perceive should correspondingly change Following these developments in psychology, a number of philosophers of science from the middle of the 20th century started questioning the status of observation in scientific theories (Hanson 1958; Kuhn 1962) The idea was that, contrary to common belief, observation is theory-laden and scientists working in different 116