Belief in Bayesian optimality is an example of recurring efforts to escape from the study of basic mechanisms into a world of ideals. Reality takes its revenge as more and more departures from ideal are found and attempts to explain them are refuted or forced into extra detail, as Rahnev and Denison (R&D) show.
Departures from optimality do not need explaining nor can they illuminate mechanisms of perceptual performance. The Bayesian programme fails to reckon with Claude Shannon's insistence that the quantity of information tells us nothing about what the information contains (Shannon & Weaver Reference Shannon and Weaver1949). Success or failure at meeting a criterion of optimal use of the amount of information in experimenter's stimuli is irrelevant to what is actually going on in making a perceptual decision. First, any experiment is rich in information of which optimality calculations take no account. Crucial contextual information is explicit as physical arrangements and social signals such as verbal instructions, and implicit in the cultural and material memory that the individual participant brings to each response. Second, the contextual information contents can make some of the content of the experimenter's stimuli dispensable for the perceptual decision. Hence, calculations that do not take context into account can yield an illusory suboptimality. In fact, a substantial number of participants use the whole of the information that each processes from the present and past (e.g., Booth et al. Reference Booth, Freeman, Konle, Wainwright and Sharpe2011a; Reference Booth, Sharpe, Freeman and Conner2011b).
R&D identify individuality as one source of suboptimality. Far more than that, disregard of individuality prevents mechanistic understanding. Every perceptual decision is determined by an individual's use of information contained in the cultural and material environment of the test. This causal mediation is the transient structure in discrimination-scaled distances between the individual's present and past output/input values (Booth & Freeman Reference Booth and Freeman1993; Booth et al. Reference Booth, Sharpe, Freeman and Conner2011b). The processed information content varies across individuals and circumstances, and even between particular occasions of the same situation in the same person. Therefore, raw data from individuals should never be averaged before testing a mechanistic hypothesis (Booth Reference Booth2017; Booth & Freeman Reference Booth and Freeman1993; Conner et al. Reference Conner, Haddon, Pickering and Booth1988), as is now becoming more widely acknowledged (e.g., Luce Reference Luce2013). The standard observer models sought by R&D also neglect the idiosyncrasies of information content in the actual mechanisms of perceptual decision making.
Many of results cited by R&D indicate that physical stimuli and context provided by the investigators interact with social context brought by the participant. One of the paradigms reviewed by R&D is psychophysical judgement. In the usual design, the experimenter uses one of a pair of stimuli as a standard of comparison with the other stimulus, which is varied. In fact, each stimulus presentation, whether test or standard, is compared with memory of previous stimuli. The comparative decision is determined by the difference in distances of test and standard from memory of previous exposures in a similar context (e.g., Booth & Freeman Reference Booth and Freeman1993; Stewart et al. Reference Stewart, Brown and Chater2005). The standard stimulus is at best redundant and may even be a source of range-frequency bias (Poulton Reference Poulton1989; cp. Conner et al. Reference Conner, Land and Booth1987 and Booth Reference Booth2015).
In other words, the experimenter's standard is part of the physical context for the test stimulus on which the perceptual decision is supposedly made. Far from the memory of the first stimulus decaying, as R&D cite, long-term memory is updated at each presentation. That is how pretreatment with a high incidence of positive stimuli reduces a bias to making negative responses, also cited by R&D. Accurate diagnosis of the causal structure of a session of perceptual decisions depends on personal tailoring of stimulus levels to be balanced around the familiar level within the range of Weber fraction constancy (Booth et al. Reference Booth, Higgs, Schneider and Klinkenberg2010; Reference Booth, Freeman, Konle, Wainwright and Sharpe2011a; Conner et al. Reference Conner, Haddon, Pickering and Booth1988).
R&D review a number of the paradigms showing effects of social context, disguised as personality score. Personality inventories are designed to obscure differences in behaviour between situations to create a stable trait, but they vary with state to unknown extents. To permit mechanistic analysis, each relevant social signal has to be presented at two or more levels, unconfounded with other signals within a session. For example, anxiety about being a lying witness in a detection task might be manipulated by the experimenter indicating that some stimuli test for absence of the signal.
Confidence ratings merely express optimistic behaviour or other habits, rather than giving introspective access to mechanisms of perception. Whether the causation is conscious or unconscious, the only access is through output-input relationships placed on a universal scale of discrimination between present and past.
R&D discuss the variations in the tradeoff between speed and accuracy in reaction times induced by direct instructions or time limits on massed tests, without considering these designs as social pressures. If stimuli provided more scope for using past experience, then the mechanisms of interaction with social context could be investigated. For example, conventional demands for a fast decision or a right answer could be presented at different levels.
Similarly, deficiencies in signal detection cited by R&D could be reduced by more ecological validity of the random background provided for the test signal. If a familiar enough context were provided throughout, the variance of the response distribution would less likely be higher in the presence of the signal. Furthermore, the line of investigation could be relevant to theoretical and practical issues in such contexts (Booth Reference Booth2015).
R&D review evidence that improper perceptual criteria and supposed misweightings in cue combinations account for suboptimality. If the experiments were designed to be analogs of familiar real life scenarios, personally relevant content of the cues could be tested as mechanisms to explain performance.
Finally, it should be noted that effect of unidentified contextual factors can be measured from the individual's causal structure of discrimination-scaled content in a session of tests. First, the most successful combination of known output/input relationships may account for substantially less than the total variance in the perceptual response. Second, the discrimination distances between present and past of observed features of the situation may not interpolate through the zero from the past. The eccentricity measures the contextual defect in discrimination units or response quantity (Booth et al. Reference Booth, Sharpe, Freeman and Conner2011b).
You have
Access
Belief in Bayesian optimality is an example of recurring efforts to escape from the study of basic mechanisms into a world of ideals. Reality takes its revenge as more and more departures from ideal are found and attempts to explain them are refuted or forced into extra detail, as Rahnev and Denison (R&D) show.
Departures from optimality do not need explaining nor can they illuminate mechanisms of perceptual performance. The Bayesian programme fails to reckon with Claude Shannon's insistence that the quantity of information tells us nothing about what the information contains (Shannon & Weaver Reference Shannon and Weaver1949). Success or failure at meeting a criterion of optimal use of the amount of information in experimenter's stimuli is irrelevant to what is actually going on in making a perceptual decision. First, any experiment is rich in information of which optimality calculations take no account. Crucial contextual information is explicit as physical arrangements and social signals such as verbal instructions, and implicit in the cultural and material memory that the individual participant brings to each response. Second, the contextual information contents can make some of the content of the experimenter's stimuli dispensable for the perceptual decision. Hence, calculations that do not take context into account can yield an illusory suboptimality. In fact, a substantial number of participants use the whole of the information that each processes from the present and past (e.g., Booth et al. Reference Booth, Freeman, Konle, Wainwright and Sharpe2011a; Reference Booth, Sharpe, Freeman and Conner2011b).
R&D identify individuality as one source of suboptimality. Far more than that, disregard of individuality prevents mechanistic understanding. Every perceptual decision is determined by an individual's use of information contained in the cultural and material environment of the test. This causal mediation is the transient structure in discrimination-scaled distances between the individual's present and past output/input values (Booth & Freeman Reference Booth and Freeman1993; Booth et al. Reference Booth, Sharpe, Freeman and Conner2011b). The processed information content varies across individuals and circumstances, and even between particular occasions of the same situation in the same person. Therefore, raw data from individuals should never be averaged before testing a mechanistic hypothesis (Booth Reference Booth2017; Booth & Freeman Reference Booth and Freeman1993; Conner et al. Reference Conner, Haddon, Pickering and Booth1988), as is now becoming more widely acknowledged (e.g., Luce Reference Luce2013). The standard observer models sought by R&D also neglect the idiosyncrasies of information content in the actual mechanisms of perceptual decision making.
Many of results cited by R&D indicate that physical stimuli and context provided by the investigators interact with social context brought by the participant. One of the paradigms reviewed by R&D is psychophysical judgement. In the usual design, the experimenter uses one of a pair of stimuli as a standard of comparison with the other stimulus, which is varied. In fact, each stimulus presentation, whether test or standard, is compared with memory of previous stimuli. The comparative decision is determined by the difference in distances of test and standard from memory of previous exposures in a similar context (e.g., Booth & Freeman Reference Booth and Freeman1993; Stewart et al. Reference Stewart, Brown and Chater2005). The standard stimulus is at best redundant and may even be a source of range-frequency bias (Poulton Reference Poulton1989; cp. Conner et al. Reference Conner, Land and Booth1987 and Booth Reference Booth2015).
In other words, the experimenter's standard is part of the physical context for the test stimulus on which the perceptual decision is supposedly made. Far from the memory of the first stimulus decaying, as R&D cite, long-term memory is updated at each presentation. That is how pretreatment with a high incidence of positive stimuli reduces a bias to making negative responses, also cited by R&D. Accurate diagnosis of the causal structure of a session of perceptual decisions depends on personal tailoring of stimulus levels to be balanced around the familiar level within the range of Weber fraction constancy (Booth et al. Reference Booth, Higgs, Schneider and Klinkenberg2010; Reference Booth, Freeman, Konle, Wainwright and Sharpe2011a; Conner et al. Reference Conner, Haddon, Pickering and Booth1988).
R&D review a number of the paradigms showing effects of social context, disguised as personality score. Personality inventories are designed to obscure differences in behaviour between situations to create a stable trait, but they vary with state to unknown extents. To permit mechanistic analysis, each relevant social signal has to be presented at two or more levels, unconfounded with other signals within a session. For example, anxiety about being a lying witness in a detection task might be manipulated by the experimenter indicating that some stimuli test for absence of the signal.
Confidence ratings merely express optimistic behaviour or other habits, rather than giving introspective access to mechanisms of perception. Whether the causation is conscious or unconscious, the only access is through output-input relationships placed on a universal scale of discrimination between present and past.
R&D discuss the variations in the tradeoff between speed and accuracy in reaction times induced by direct instructions or time limits on massed tests, without considering these designs as social pressures. If stimuli provided more scope for using past experience, then the mechanisms of interaction with social context could be investigated. For example, conventional demands for a fast decision or a right answer could be presented at different levels.
Similarly, deficiencies in signal detection cited by R&D could be reduced by more ecological validity of the random background provided for the test signal. If a familiar enough context were provided throughout, the variance of the response distribution would less likely be higher in the presence of the signal. Furthermore, the line of investigation could be relevant to theoretical and practical issues in such contexts (Booth Reference Booth2015).
R&D review evidence that improper perceptual criteria and supposed misweightings in cue combinations account for suboptimality. If the experiments were designed to be analogs of familiar real life scenarios, personally relevant content of the cues could be tested as mechanisms to explain performance.
Finally, it should be noted that effect of unidentified contextual factors can be measured from the individual's causal structure of discrimination-scaled content in a session of tests. First, the most successful combination of known output/input relationships may account for substantially less than the total variance in the perceptual response. Second, the discrimination distances between present and past of observed features of the situation may not interpolate through the zero from the past. The eccentricity measures the contextual defect in discrimination units or response quantity (Booth et al. Reference Booth, Sharpe, Freeman and Conner2011b).