{"title": "Theories of Access Consciousness", "book": "Advances in Neural Information Processing Systems", "page_first": 289, "page_last": 296, "abstract": null, "full_text": "              Theories Of Access Consciousness\n\n\n\n             Michael D. Colagrosso                              Michael C. Mozer\n        Department of Computer Science                    Institute of Cognitive Science\n            Colorado School of Mines                          University of Colorado\n             Golden, CO 80401 USA                            Boulder, CO 80309 USA\n            mcolagro@mines.edu                              mozer@colorado.edu\n\n\n\n                                            Abstract\n\n         Theories of access consciousness address how it is that some mental states but\n         not others are available for evaluation, choice behavior, and verbal report. Farah,\n         O'Reilly, and Vecera (1994) argue that quality of representation is critical; De-\n         haene, Sergent, and Changeux (2003) argue that the ability to communicate rep-\n         resentations is critical. We present a probabilistic information transmission or\n         PIT model that suggests both of these conditions are essential for access con-\n         sciousness. Having successfully modeled data from the repetition priming litera-\n         ture in the past, we use the PIT model to account for data from two experiments\n         on subliminal priming, showing that the model produces priming even in the ab-\n         sence of accessibility and reportability of internal states. The model provides a\n         mechanistic basis for understanding the dissociation of priming and awareness.\n\n\nPhilosophy has made many attempts to identify distinct aspects of consciousness. Perhaps\nthe most famous effort is Block's (1995) delineation of phenomenal and access conscious-\nness. Phenomenal consciousness has to do with \"what it is like\" to experience chocolate\nor a pin prick. Access consciousness refers to internal states whose content is \"(1) inferen-\ntially promiscuous, i.e., poised to be used as a premise in reasoning, (2) poised for control\nof action, and (3) poised for rational control of speech.\" (p. 230) The scientific study of con-\nsciousness has exploded in the past six years, and an important catalyst for this explosion\nhas been the decision to focus on the problem of access consciousness: how is it that some\nmental states but not others become available for evaluation, choice behavior, verbal report,\nand storage in working memory. Another reason for the recent explosion of consciousness\nresearch is the availability of functional imaging techniques to explore differences in brain\nactivation between conscious and unconscious states, as well as the development of clever\npsychological experiments that show that a stimulus that is not consciously perceived can\nnonetheless influence cognition, which we describe shortly.\n\n\n1    Subliminal Priming\n\nThe phenomena we address utilize an experimental paradigm known as repetition priming.\nPriming refers to an improvement in efficiency in processing a stimulus item as a result\nof previous exposure to the item. Efficiency is defined in terms of shorter response times,\nlower error rates, or both. A typical long-term perceptual priming experiment consists of\na study phase during which participants are asked to read aloud a list of words, and a test\nphase during which participants must name or categorize a series of words, presented one\nat a time. Reaction time is lower and/or accuracy is higher for test words that were also on\nthe study list. Repetition priming occurs without strategic effort on the part of participants,\nand therefore appears to be a low level mechanism of learning, which likely serves as the\nmechanism underlying the refinement of cognitive skills with practice.\n\n\f\nIn traditional studies, priming is supraliminal--the prime is consciously perceived. In the\nstudies we model here, primes are subliminal. Subliminal priming addresses fundamental\nissues concerning conscious access: How is it that a word or image that cannot be identified,\ndetected, or even discriminated in forced choice can nonetheless influence the processing\nof a subsequent stimulus word? Answering this question in a computational framework\nwould be a significant advance toward understanding the nature of access consciousness.\n\n\n2    Models of Conscious and Unconscious Processing\n\nIn contrast to the wealth of experimental data, and the large number of speculative and\nphilosophical papers on consciousness, concrete computational models are rare. The do-\nmain of consciousness is particularly ripe for theoretical perspectives, because it is a sig-\nnificant contribution to simply provide an existence proof of a mechanism that can explain\nspecific experimental data. Ordinarily, a theorist faces skepticism when presenting a model;\nit often seems that hundreds of alternative, equally plausible accounts must exist. However,\nwhen addressing data deemed central to issues of consciousness, simply providing a con-\ncrete handle on the phenomena serves to demystify consciousness and bring it into the\nrealm of scientific understanding.\n\nWe are familiar with only three computational models that address specific experimental\ndata in the domain of consciousness. We summarize these models, and then present a\nnovel model and describe its relationship to the previous efforts. Farah, O'Reilly, and Ve-\ncera (1994) were the first to model specific phenomena pertaining to consciousness in a\ncomputational framework. The phenomena involve prosopagnosia, a deficit of overt face\nrecognition following brain damage. Nonetheless, prosopagnosia patients exhibit residual\ncovert recognition by a variety of tests. For example, when patients are asked to categorize\nnames as famous or nonfamous, their response times are faster to a famous name when the\nname is primed by a picture of a semantically related face (e.g., the name \"Bill Clinton\"\nwhen preceded by a photograph of Hillary), despite the fact that they could not identify the\nrelated face. Farah et al. model face recognition in a neural network, and show that when\nthe network is damaged, it loses the ability to perform tasks requiring high fidelity represen-\ntations (e.g., identification) but not tasks requiring only coarse information (e.g., semantic\npriming). They argue that conscious perception is associated with a certain minimal quality\nof representation.\n\nDehaene and Naccache (2001) outline a framework based on Baars' (1989) notion of con-\nscious states as residing in a global workspace. They describe the workspace as a \"dis-\ntributed neural system...with long-distance connectivity that can potentially interconnect\nmultiple specialized brain areas in a coordinated, though variable manner.\" (p. 13) De-\nhaene, Sergent, and Changeaux (2003) implement this framework in a complicated archi-\ntecture of integrate-and-fire neurons and show that the model can qualitatively account for\nthe attentional blink phenomenon. The attentional blink is observed in experiments where\nparticipants are shown a rapid series of stimuli, which includes two targets (T1 and T2).\nIf T2 appears shortly after T1, the ability to report T2 drops, as if attention is distracted.\nDehane et al. explain this phenomenon as follows. When T1 is presented, its activation\npropagates to frontal cortical areas (the global workspace). Feedback connections lead to\na resonance between frontal and posterior areas, which strengthen T1 but block T2 from\nentering the workspace. If the T1-T2 lag is sufficiently great, habituation of T1 sufficiently\nweakens the representation such that T2 can enter the workspace and suppress T1. In this\naccount, conscious access is achieved via resonance between posterior and frontal areas.\n\nAlthough the Farah et al. and Dehaene et al. models might not seem to have much in\ncommon, they both make claims concerning what is required to achieve functional con-\nnectivity between perceptual and response systems. Farah et al. focus on aspects of the\nrepresentation; Dehaene et al. focus on a pathway through which representations can be\n\n\f\ncommunicated. These two aspects are not incompatible, and in fact, a third model incorpo-\nrates both. Mathis and Mozer (1996) describe an architecture with processing modules for\nperceptual and response processes, implemented as attractor neural nets. They argue that in\norder for a representation in some perceptual module to be assured of influencing a response\nmodule, (a) it must have certain characteristicstemporal persistence and well-formedness\nwhich is quite similar to Farah et al.'s notion of quality, and (b) the two modules must be\ninterconnected--which is the purpose of Dehaene et al.'s global workspace. The model has\ntwo limitations that restrict its value as a contemporary account of conscious access. First,\nit addressed classical subliminal priming data, but more reliable data has recently been\nreported. Second, like the other two models, Mathis and Mozer used a complex neural\nnetwork architecture with arbitrary assumptions built in, and the sensitivity of the model's\nbehavior to these assumptions is far from clear. In this paper, we present a model that em-\nbodies the same assumptions as Mathis and Mozer, but overcomes its two limitations, and\nexplains subliminal-priming data that has yet to be interpreted via a computational model.\n\n\n3    The Probabilistic Information Transmission (PIT) Framework\n\nOur model is based on the probabilistic information transmission or PIT framework of\nMozer, Colagrosso, and Huber (2002, 2003). The framework characterizes the transmission\nof information from perceptual to response systems, and how the time course of informa-\ntion transmission changes with experience (i.e., priming). Mozer et al. used this framework\nto account for a variety of facilitation effects from supraliminal repetition priming.\n\nThe framework describes cognition in terms of a collection of information-processing path-\nways, and supposes that any act of cognition involves coordination among multiple path-\nways. For example, to model a letter-naming task where a letter printed in upper or lower\ncase is presented visually and the letter must be named, the framework would assume a\nperceptual pathway that maps the visual input to an identity representation, and a response\npathway that maps a identity representation to a naming response. The framework is for-\nmalized as a probabilistic model: the pathway input and output are random variables and\nmicroinference in a pathway is carried out by Bayesian belief revision.\n\nThe framework captures the time course of information processing for a single experi-\nmental trial. To elaborate, consider a pathway whose input at time t is a discrete random\nvariable, denoted X(t), which can assume values x1, x2, x3, . . . , xn corresponding to al-\n                                                                       x\nternative input states. Similarly, the output of the pathway at time t is a discrete random\nvariable, denoted Y (t), which can assume values y1, y2, y3, . . . , yn . For example, in the\n                                                                      y\nletter-naming task, the input to the perceptual pathway would be one of nx = 52 visual\npatterns corresponding to the upper- and lower-case letters of the alphabet, and the output\nis one of ny = 26 letter identities. To present a particular input alternative, say xi, to the\nmodel for T time steps, we specify X(t) = xi for t = 1 . . . T , and allow the model to\ncompute P(Y (t) | X(1) . . . X(t)).\n\nA pathway is modeled as a dynamic Bayes network; the minimal version of the model\nused in the present simulations is simply a hidden Markov model, where the X(t) are\nobservations and the Y (t) are inferred state (see Figure 1a). In typical usage, an HMM\nis presented with a sequence of distinct inputs, whereas we maintain the same input for\nmany successive time steps; and an HMM transitions through a sequence of distinct hidden\nstates, whereas we attempt to converge with increasing confidence on a single state.\n\nFigure 1b illustrates the time course of inference in a single pathway with 52 input and\n26 output alternatives and two-to-one associations. The solid line in the Figure shows, as\na function of time t, P(Y (t) = yi | X(1) = x2i . . . X(t) = x2i), i.e., the probability\nthat input i (say, the visual pattern of an upper case O) will produce its target output (the\nletter identity). Evidence for the target output accumulates gradually over time, yielding a\nspeed-accuracy curve that relates the number of iterations to the accuracy of identification.\n\n\f\n                                                                        1\n\n                                                                  0.8\n       Y          Y\n            0          1     Y2          Y3\n                                                                  0.6                     O\n\n                                                                  0.4\n                                                               P(Output) 0.2              Q\n                  X          X\n                       1          2      X3                             0\n(a)                                                     (b)                                    Time\n\nFigure 1: (a) basic pathway architecture--a hidden Markov model; (b) time course of inference in a\npathway when the letter O is presented, causing activation of both O and the visually similar Q.\n\n\nThe exact shape of the speed-accuracy curve--the pathway dynamics--are determined by\nthree probability distributions, which embody the knowledge and past experience of the\nmodel. First, P(Y (0)) is the prior distribution over outputs in the absence of any informa-\ntion about the input. Second, P(Y (t) | Y (t - 1)) characterizes how the pathway output\nevolves over time. We assume the transition probability matrix serves as a memory with\ndiffusion, i.e., P(Y (t) = yi|Y (t - 1) = yj) = (1 - )ij + P(Y (0) = yi), where  is the\ndiffusion constant and ij is the Kronecker delta. Third, P(X(t) | Y (t)) characterizes the\nstrength of association between inputs and outputs. The greater the association strength,\nthe more rapidly that information about X will be communicated to Y . We parameterize\nthis distribution as P(X(t) = xi|Y (t) = yj)  1 +                      \n                                                                k               ik kj , where ij indicates the\nfrequency of experience with the association between states xi and yj, and ik specifies the\nsimilarity between states xi and xk. (Although the representation of states is localist, the \nterms allow us to design in the similarity structure inherent in a distributed representation.)\nThese association strengths are highly constrained by the task structure and the similarity\nstructure and familiarity of the inputs.\n\nFundamental to the framework is the assumption that with each experience, a pathway be-\ncomes more efficient at processing an input. Efficiency is reflected by a shift in the speed-\naccuracy curve to the left. In Mozer, Colagrosso, and Huber (2002, 2003), we propose two\ndistinct mechanisms to model phenomena of supraliminal priming. First, the association\nfrequencies, ij, are increased following a trial in which xi leads to activation of yj, re-\nsulting in more efficient transmission of information, corresponding to an increased slope\nof the solid line in Figure 1b. The increase is Hebbian, based on the maximum activation\nachieved by xi and yj: ij =  maxt P(X(t) = xi)P(Y (t) = yj), where  is a step size.\nSecond, the priors, which serve as a model of the environment, are increased to indicate a\ngreater likelihood of the same output occurring again in the future. In modeling data from\nsupraliminal priming, we found that the increases to association frequencies are long last-\ning, but the increases to the priors decay over the course of a few minutes or a few trials.\nAs a result, the prior updating does not play into the simulation we report here; we refer\nthe reader to Mozer, Colagrosso, and Huber (2003) for details.\n\n\n4      Access Consciousness and PIT\n\nWe have described the operation of a single pathway, but to model any cognitive task,\nwe require a series of pathways in cascade. For a simple choice task, we use a percpet-\nual pathway cascaded to a response pathway. The interconnection between the pathways\nis achieved by copying the output of the perceptual pathway, Y p(t), to the input of the\nresponse pathway, Xr(t), at each time t.\n\nThis multiple-pathway architecture allows us to characterize the notion of access con-\nsciousness. Considering the output of the perceptual pathway, access is achieved when:\n(1) the output representation is sufficient to trigger the correct behavior in the response\npathway, and (2) the perceptual and response pathways are functionally interconnected. In\nmore general terms, access for a perceptual pathway output requires that these two condi-\n\n\f\ntions be met not just for a specific response pathway, but for arbitrary response pathways\n(e.g., pathways for naming, choice, evaluation, working memory, etc.). In Mozer and Co-\nlagrosso (in preparation) we characterize the sufficiency requirements of condition 1; they\ninvolve a representation of low entropy that stays active for long enough that the represen-\ntation can propagate to the next pathway.\n\nAs we will show, a briefly presented stimulus fails to achieve a representation that supports\nchoice and naming responses. Nonetheless, the stimulus evokes activity in the perceptual\npathway. Because perceptual priming depends on the magnitude of the activation in the\nperceptual pathway, not on the activation being communicated to response pathways, the\nframework is consistent with the notion of priming occurring in the absence of awareness.\n\n\n4.1    Simulation of Bar and Biederman (1998)\n\nBar and Biederman (1998) presented a sequence of masked line drawings of objects and\nasked participants to name the objects, even if they had to guess. If the guess was incorrect,\nparticipants were required to choose the object name from a set of four alternatives. Unbe-\nknownst to the participant, some of the drawings in the series were repeated, and Bar and\nBiederman were interested in whether participants would benefit from the first presenta-\ntion even if it could not be identified. The repeated objects could be the same or a different\nexemplar of the object, and it could appear in either the same or a different display position.\n\nParticipants were able to name 13.5% of drawings on presentation 1, but accuracy jumped\nto 34.5% on presentation 2. Accuracy did improve, though not as much, if the same shape\nwas presented in a different position, but not if a different drawing of the same object\nwas presented, suggesting a locus of priming early in the visual stream. The improvement\nin accuracy is not due to practice in general, because accuracy rose only 4.0% for novel\ncontrol objects over the course of the experiment. The priming is firmly subliminal, because\nparticipants were not only unable to name objects on the first presentation, but their four-\nalternative forced choice (4AFC) performance was not much above chance (28.5%).\n\nTo model these phenomena, we created a response pathway with fifty states representing\nnames of objects that are used in the experiment, e.g., chair and lamp. We also created a\nperceptual pathway with states representing visual patterns that correspond to the names\nin the response pathway. Following the experimental design, every object identity was\ninstantiated in two distinct shapes, and every shape could be in one of nine different visual-\nfield positions, leading to 900 distinct states in the perceptual pathway to model the possible\nvisual stimuli. The following parameters were fit to the data. If two perceptual states, xi\nand xk are the same shape in different positions, they are assigned a similarity coefficient\nik = 0.95; all other similarity coefficients are zero. The association frequency, , for\nvalid associations in the perceptual pathway was 22, and the response pathway 18. Other\nparameters were p = .05, r = .01, and  = 1.0.\n\nThe PIT model achieves a good fit to the human experimental data (Figure 2). Specifi-\ncally, priming is greatest for the same shape in the same position, some priming occurs for\nthe same shape in a different position, and no substantial priming occurs for the different\nshape. Figure 3a shows the time course of activation of a stimulus representation in the\nperceptual pathway when the stimulus is presented for 50 iterations, on both the first and\nthird presentations. The third presentation was chosen instead of the second to make the\neffect of priming clearer.\n\nEven though a shape cannot be named on the first presentation, partial information about\nthe shape may nonetheless be available for report. The 4AFC test of Bar and Bieder-\nman provides a more sensitive measure of residual stimulus information. In past work,\nwe modeled forced-choice tasks using a response pathway with only the alternatives under\nconsideration. However, in this experiment, forced-choice performance must be estimated\nconditional on incorrect naming. In PIT framework, we achieve this using naming and\n\n\f\n                            40                                                                                                                   40\n\n                                       First Block                                                                                                         First Block\n                            35                                                         Second Block                                              35                                                        Second Block\n\n\n                            30                                                                                                                   30\n\n\n                            25                                                                                                                   25\n\n\n                            20                                                                                                                   20\n\n\n                            15                                                                                                                   15\n\n                            10\n  Percent Correct Naming                                                                                                                         10\n                                                                                                                       Percent Correct Naming\n\n                             5                                                                                                                    5\n\n\n                             0                                                                                                                    0\n                                   Control    Prime    SHAPE:    Same     Same        Different Different Second                                       Control    Prime    SHAPE:    Same     Same        Different Different Second\n                                  Objects     Objects POSITION: Same     Different     Same    Different    Control                                    Objects    Objects POSITION: Same     Different     Same    Different    Control\n\n\nFigure 2: (left panel) Data from Bar and Biederman (1998) (right panel) Simulation of PIT. White\nbar: accuracy on first presentation of a prime object. Black bars: the accuracy when the object is\nrepeated, either with the same or different shape, and in the same or different position. Grey bars:\naccuracy for control objects at the beginning and the end of the experiment.\n\nforced-choice output pathways having output distributions N (t) and F (t), which are linked\nvia the perceptual state, Y p(t). F (t) must be reestimated with the evidence that N (t) is\nnot the target state. This inference problem is intractable. We therefore used a shortcut in\nwhich a single response pathway is used, augmented with a simple three-node belief net\n(Figure 3b) to capture the dependence between naming and forced choice. The belief net\nhas a response pathway node Y r(t) connected to F (t) and N (t), with conditional distribu-\ntion P (N (t) = ni|Y r(t) = yj) = ij + (1 - )/|Y r|, and an analogous distribution for\nP (F (t) = fi|Y r(t) = yj). The free parameter  determines how veridically naming and\nforced-choice actions reflect response-pathway output. Over a range of ,  < 1, the model\nobtains forced-choice performance near chance on the first presentation when the naming\nresponse is incorrect. For example, with  = 0.72, the model produces a forced-choice\naccuracy on presentation 1 of 26.1%. (Interestingly, the model also produces below chance\nperformance on presentation 2 if the object is not named correctly--23.5%--which is also\nfound in the human data--20.0%.) Thus, by the stringent criterion of 4AFC, the model\nshows no access consciousness, and therefore illustrates a dissociation between priming\nand access consciousness. In our simulation, we followed the procedure of Bar and Bieder-\nman by including distractor alternatives with visual and semantic similarity to the target.\nThese distractors are critical: with unrelated distractors, the model's 4AFC performance is\nsignificantly above chance, illustrating that a perceptual representation can be adequate to\nsupport some responses but not others, as Farah et al. (1994) also argued.\n\n\n4.2                               Simulation of Abrams and Greenwald (2000)\n\nDuring an initial phase of the experiment, participants categorized 24 clearly visible target\nwords as pleasant (e.g., HUMOR) or unpleasant (e.g., SMUT). They became quite familiar\nwith the task by categorizing each word a total of eight times. In a second phase, partici-\npants were asked to classify the same targets and were given a response deadline to induce\nerrors. The targets were preceded by masked primes that could not be identified. Of interest\nis the effective valence (or EV) of the target for different prime types, defined as the error\nrate difference between unpleasant and pleasant targets. A positive (negative) EV indicates\nthat responses are biased toward a pleasant (unpleasant) interpretation by the prime. As one\nwould expect, pleasant primes resulted in a positive EV, unpleasant primes in a negative EV.\nOf critical interest is the finding that a nonword prime formed by recombining two pleas-\nant targets (e.g., HULIP from HUMOR and TULIP) or unpleasant targets (e.g., BIUT from\nBILE and SMUT) also served to bias the targets. More surprising, a positive EV resulted\nfrom unpleasant prime words formed by recombining two pleasant targets (TUMOR from\nTULIP and HUMOR), indicating that subliminal priming arises from word fragments, not\nwords as unitary entities, and providing further evidence for an early locus of subliminal\npriming. Note that the results depend critically on the first phase of the experiment, which\ngave participants extensive practice on a relatively small set of words that were then used\nas and recombined to form primes. Words not studied in the first phase (orphans) provided\n\n\f\n                         0.6\n                                                                                object, first presentation\n                         0.5                                                   object, third presentation\n                         0.4                                                                                                      N(t)             F(t)\n                                                                                          different object\n                         0.3\n                         0.2\n         Probability     0.1\n                          0 1                                                                                                             Yr(t)\n  (a)                            50                                                                            1000      (b)\n                                                                        Time (msec)\nFigure 3: (a) Activation of the perceptual representation in PIT as a function of processing iterations\non the first (thin solid line) and third (thick solid line) presentations of target. (b) Bayes net for\nperforming 4AFC conditional on incorrect naming response.\n\n                                                0.4                                                                      Experiment\n                                                0.3                                                                      Model\n                                  alence\n                                                0.2\n\n                                                0.1\n                                            fective VEf 0\n                                                             targets        hulip-type           tumor-type            orphans\n\nFigure 4: Effective valence of primes in the Abrams and Greenwald (2000) experiment for human\nsubjects (black bars) and PIT model (grey bars). HULIP-type primes are almost as strong as target\nrepetitions, and TUMOR-type primes have a positive valence, contrary to the meaning of the word.\n\n\nno significant EV effect when used as primes.\n\nIn this simulation, we used a three pathway model: a perceptual pathway that maps vi-\nsual patterns to orthography with 200 input states corresponding both to words, nonwords,\nand nonword recombinations of words; a semantic pathway that maps to 100 distinct lex-\nical/semantic states; and a judgement pathway that maps to two responses, pleasant and\nunpleasant. In the perceptual pathway, similarity structure was based on letter overlap,\nso that HULIP was similar to both TULIP and HUMOR, with  = 0.837. No similarity\nwas assumed in the semantic state representation; consistent with the previous simulation,\np = .05, s = .01, j = .01, and  = .01. At the outset of the simulation,  frequencies\nfor correct associations were 15, 19, and 25 in the perceptual, semantic, and judgement\npathways. The initial phase of the experiment was simulated by repeated supraliminal\npresentation of words, which increased the association frequencies in all three pathways\nthrough the ij learning rule.\n\nLong-term supraliminal priming is essential in establishing the association strengths, as\nwe'll explain. Short-term subliminal priming also plays a key role in the experiment. Dur-\ning the second phase of the experiment, residual activity from the prime--primarily in the\njudgement pathway--biases the response to the target. Residual activation of the prime\nis present even if the representation of the prime does not reach sufficient strength that it\ncould be named or otherwise reported.\n\nThe outcome of the simulation is consistent with the human data (Figure 4). When a\nHULIP-type prime is presented, HUMOR and TULIP become active in the semantic path-\nway because of their visual similarity to HULIP. Partial activation of these two practiced\nwords pushes the judgement pathway toward a pleasant response, resulting in a positive\nEV. When a TUMOR-type prime is presented, three different words become active in the\nsemantic pathway: HUMOR, TULIP, and TUMOR itself. Although TUMOR is more active, it\nwas not one of the words studied during the initial phase of the experiment, and as a result,\nit has a relatively weak association to the unpleasant judgement, in contrast to the other\ntwo words which have strong associations to the pleasant judgement. Orphan primes have\nlittle effect because they were not studied during the initial phase of the experiment, and\nconsequently their association to pleasant and unpleasant judgements is also weak. In sum-\nmary, activation of the prime along a critical, well-practiced pathway may not be sufficient\nto support an overt naming response, yet it may be sufficient to bias the processing of the\n\n\f\nimmediately following target.\n\n\n5    Discussion\n\nAn important contribution of this work has been to demonstrate that specific experimental\nresults relating to access consciousness and subliminal priming can be interpreted in a con-\ncrete computational framework. By necessity, the PIT framework, which we previously\nused to model supraliminal priming data, predicts the existence of subliminal priming,\nbecause the mechanisms giving rise to priming depend on degree of activation of a rep-\nresentation, whereas the processes giving rise to access consciousness also depend on the\ntemporal persistence of a representation.\n\nAnother contribution of this work has been to argue that two previous computational mod-\nels each tell only part of the story. Farah et al. argue that quality of representation is\ncritical; Dehaene et al. argue that pathways to communicate representations is critical. The\nPIT framework argues that both of these features are necessary for access consciousness.\n\nAlthough the PIT framework is not completely developed, it nonetheless makes a clear\nprediction: that subliminal priming is can never be stronger than supraliminal priming,\nbecause the maximal activation of subliminal primes is never greater than that of supral-\niminal primes. One might argue that many theoretical frameworks might predict the same,\nbut no other computational model is sufficiently well developed--in terms of addressing\nboth priming and access consciousness--to make this prediction.\n\nIn its current stage of development, a weakness of the PIT framework is that it is silent\nas to how perceptual and response pathways become flexibly interconnected based on task\ndemands. However, the PIT framework is not alone in failing to address this critical issue:\nThe Dehaene et al. model suggests that once a representation enters the global workspace,\nall response modules can access it, but the model does not specify how the appropriate per-\nceptual module wins the competition to enter the global workspace, or how the appropriate\nresponse module is activated. Clearly, flexible cognitive control structures that perform\nthese functions are intricately related to mechanisms of consciousness.\n\n\nAcknowledgments\n\nThis research was supported by NIH/IFOPAL R01 MH6154901A1.\n\nReferences\n\nAbrams, R. L., & Greenwald, A. G. (2000). Parts outweigh the whole (word) in unconscious analysis\n      of meaning. Psychological Science, 11(2), 118124.\nBaars, B. (1989). A cognitive theory of consciousness. Cambridge: Cambridge University Press.\nBar, M., & Biederman, I. (1998). Subliminal visual priming. Psychological Science, 9(6), 464468.\nBlock, N. (1995). On a confusion about a function of consciousness. Brain and Behavioral Sciences,\n      18(2), 227247.\nDehaene, S., & Naccache, L. (2001). Towards a cognitive neuroscience of consciousness: basic\n      evidence and a workspace framework. Cognition, 79, 137.\nDehaene, S., Sergent, C., & Changeux, J.-P. (2003). A neuronal network model linking subjec-\n      tive reports and objective physiological data during conscious perception. Proceedings of the\n      National Academy of Sciences, 100, 85208525.\nFarah, M. J., O'Reilly, R. C., & Vecera, S. P. (1994). Dissociated overt and covert recognition as an\n      emergent property of a lesioned neural network. Psychological Review, 100, 571588.\nMathis, D. W., & Mozer, M. C. (1996). Conscious and unconscious perception: a computational\n      theory. In G. Cottrell (Ed.), Proceedings of the Eighteenth Annual Conference of the Cognitive\n      Science Society (pp. 324328). Hillsdale, NJ: Erlbaum & Associates.\nMozer, M. C., Colagrosso, M. D., & Huber, D. E. (2002). A rational analysis of cognitive control\n      in a speeded discrimination task. In T. G. Dietterich, S. Becker, & Z. Ghahramani (Eds.),\n      Advances in Neural Information Processing Systems 14. Cambridge, MA: MIT Press.\nMozer, M. C., Colagrosso, M. D., & Huber, D. E. (2003). Mechanisms of long-term repetition\n      priming and skill refinement: A probabilistic pathway model. In Proceedings of the Twenty-\n      Fifth Annual Conference of the Cognitive Science Society. Hillsdale, NJ: Erlbaum Associates.\n\n\f\n", "award": [], "sourceid": 2715, "authors": [{"given_name": "Michael", "family_name": "Colagrosso", "institution": null}, {"given_name": "Michael", "family_name": "Mozer", "institution": null}]}