Felipe De Brigard

Attention and consciousness

For the past three decades there has been a substantial amount of scientific evidence supporting the view that attention is necessary and sufficient for perceptual representations to become conscious (i.e., for there to be something that it is like to experience a representational perceptual state). This view, however, has been recently questioned on the basis of some alleged counterevidence. In this paper we survey some of the most important recent findings. In doing so, we have two primary goals. The first is descriptive: we provide a literature review for those seeking an understanding of the present debate. The second is editorial: we suggest that the evidence alleging dissociations between consciousness and attention is not decisive. Thus, this is an opinionated overview of the debate. By presenting our assessment, we hope to bring out both sides in the debate and to underscore that the issues here remain matters of intense controversy and ongoing investigation. Copyright

Episodic future thinking and episodic counterfactual thinking: Intersections between memory and decisions

This article considers two recent lines of research concerned with the construction of imagined or simulated events that can provide insight into the relationship between memory and decision making. One line of research concerns episodic future thinking, which involves simulating episodes that might occur in ones personal future, and the other concerns episodic counterfactual thinking, which involves simulating episodes that could have happened in ones personal past. We first review neuroimaging studies that have examined the neural underpinnings of episodic future thinking and episodic counterfactual thinking. We argue that these studies have revealed that the two forms of episodic simulation engage a common core network including medial parietal, prefrontal, and temporal regions that also supports episodic memory. We also note that neuroimaging studies have documented neural differences between episodic future thinking and episodic counterfactual thinking, including differences in hippocampal responses. We next consider behavioral studies that have delineated both similarities and differences between the two kinds of episodic simulation. The evidence indicates that episodic future and counterfactual thinking are characterized by similarly reduced levels of specific detail compared with episodic memory, but that the effects of repeatedly imagining a possible experience have sharply contrasting effects on the perceived plausibility of those events during episodic future thinking versus episodic counterfactual thinking. Finally, we conclude by discussing the functional consequences of future and counterfactual simulations for decisions.

Is memory for remembering? Recollection as a form of episodic hypothetical thinking

Misremembering is a systematic and ordinary occurrence in our daily lives. Since it is commonly assumed that the function of memory is to remember the past, misremembering is typically thought to happen because our memory system malfunctions. In this paper I argue that not all cases of misremembering are due to failures in our memory system. In particular, I argue that many ordinary cases of misremembering should not be seen as instances of memory’s malfunction, but rather as the normal result of a larger cognitive system that performs a different function, and for which remembering is just one operation. Building upon extant psychological and neuroscientific evidence, I offer a picture of memory as an integral part of a larger system that supports not only thinking of what was the case and what potentially could be the case, but also what could have been the case. More precisely, I claim that remembering is a particular operation of a cognitive system that permits the flexible recombination of different components of encoded traces into representations of possible past events that might or might not have occurred, in the service of constructing mental simulations of possible future events. So that imagination and memory are but one thing, which for diverse considerations hath diverse names. Thomas Hobbes, Leviathan 1.2. So that imagination and memory are but one thing, which for diverse considerations hath diverse names. Thomas Hobbes, Leviathan 1.2.AbstractMisremembering is a systematic and ordinary occurrence in our daily lives. Since it is commonly assumed that the function of memory is to remember the past, misremembering is typically thought to happen because our memory system malfunctions. In this paper I argue that not all cases of misremembering are due to failures in our memory system. In particular, I argue that many ordinary cases of misremembering should not be seen as instances of memory’s malfunction, but rather as the normal result of a larger cognitive system that performs a different function, and for which remembering is just one operation. Building upon extant psychological and neuroscientific evidence, I offer a picture of memory as an integral part of a larger system that supports not only thinking of what was the case and what potentially could be the case, but also what could have been the case. More precisely, I claim that remembering is a particular operation of a cognitive system that permits the flexible recombination of different components of encoded traces into representations of possible past events that might or might not have occurred, in the service of constructing mental simulations of possible future events. So that imagination and memory are but one thing, which for diverse considerations hath diverse names.Thomas Hobbes, Leviathan 1.2.

Coming to Grips With the Past Effect of Repeated Simulation on the Perceived Plausibility of Episodic Counterfactual Thoughts

When people revisit previous experiences, they often engage in episodic counterfactual thinking: mental simulations of alternative ways in which personal past events could have occurred. The present study employed a novel experimental paradigm to examine the influence of repeated simulation on the perceived plausibility of upward, downward, and neutral episodic counterfactual thoughts. Participants were asked to remember negative, positive, and neutral autobiographical memories. One week later, they self-generated upward, downward, and neutral counterfactual alternatives to those memories. The following day, they resimulated each of those counterfactuals either once or four times. The results indicate that repeated simulation of upward, downward, and neutral episodic counterfactual events decreases their perceived plausibility while increasing ratings of the ease, detail, and valence of the simulations. This finding suggests a difference between episodic counterfactual thoughts and other kinds of self-referential simulations. Possible implications of this finding for pathological and nonpathological anxiety are discussed.

If you like it, does it matter if it's real?

Most peoples intuitive reaction after considering Nozicks experience machine thought-experiment seems to be just like his: we feel very little inclination to plug in to a virtual reality machine capable of providing us with pleasurable experiences. Many philosophers take this empirical fact as sufficient reason to believe that, more than pleasurable experiences, people care about “living in contact with reality.” Such claim, however, assumes that peoples reaction to the experience machine thought-experiment is due to the fact that they value reality over virtual experiences—an assumption that has seldom (if ever) been questioned. This paper challenges that very assumption. I report some experimental evidence suggesting that the intuition elicited by the thought-experiment may be explainable by the fact that people are averse to abandon the life they have been experiencing so far, regardless of whether such life is virtual or real. I use then an explanatory model, derived from what behavioral economists and psychologists call the status quo bias, to make sense of these results. Finally, I argue that since this explanation also accounts for peoples reaction toward Nozicks thought-experiment, it would be wrong to take such intuition as evidence that people value being in touch with reality.Most peoples intuitive reaction after considering Nozicks experience machine thought-experiment seems to be just like his: we feel very little inclination to plug in to a virtual reality machine capable of providing us with pleasurable experiences. Many philosophers take this empirical fact as sufficient reason to believe that, more than pleasurable experiences, people care about “living in contact with reality.” Such claim, however, assumes that peoples reaction to the experience machine thought-experiment is due to the fact that they value reality over virtual experiences—an assumption that has seldom (if ever) been questioned. This paper challenges that very assumption. I report some experimental evidence suggesting that the intuition elicited by the thought-experiment may be explainable by the fact that people are averse to abandon the life they have been experiencing so far, regardless of whether such life is virtual or real. I use then an explanatory model, derived from what behavioral economists...

Neural activity associated with self, other, and object-based counterfactual thinking

Previous research has shown that autobiographical episodic counterfactual thinking-i.e., mental simulations about alternative ways in which ones life experiences could have occurred-engages the brains default network (DN). However, it remains unknown whether or not the DN is also engaged during impersonal counterfactual thoughts, specifically those involving other people or objects. The current study compares brain activity during counterfactual simulations involving the self, others and objects. In addition, counterfactual thoughts involving others were manipulated in terms of similarity and familiarity with the simulated characters. The results indicate greater involvement of DN during person-based (i.e., self and other) as opposed to object-based counterfactual simulations. However, the involvement of different regions of the DN during other-based counterfactual simulations was modulated by how close and/or similar the simulated character was perceived to be by the participant. Simulations involving unfamiliar characters preferentially recruited dorsomedial prefrontal cortex. Simulations involving unfamiliar similar characters, characters with whom participants identified personality traits, recruited lateral temporal gyrus. Finally, our results also revealed differential coupling of right hippocampus with lateral prefrontal and temporal cortex during counterfactual simulations involving familiar similar others, but with left transverse temporal gyrus and medial frontal and inferior temporal gyri during counterfactual simulations involving either oneself or unfamiliar dissimilar others. These results suggest that different brain mechanisms are involved in the simulation of personal and impersonal counterfactual thoughts, and that the extent to which regions associated with autobiographical memory are recruited during the simulation of counterfactuals involving others depends on the perceived similarity and familiarity with the simulated individuals.

The Role of Attention in Conscious Recollection

Most research on the relationship between attention and consciousness has been limited to perception. However, perceptions are not the only kinds of mental contents of which we can be conscious. An important set of conscious states that has not received proper treatment within this discussion is that of memories. This paper reviews compelling evidence indicating that attention may be necessary, but probably not sufficient, for conscious recollection. However, it is argued that unlike the case of conscious perception, the kind of attention required during recollection is internal, as opposed to external, attention. As such, the surveyed empirical evidence is interpreted as suggesting that internal attention is necessary, but probably not sufficient, for conscious recollection. The paper begins by justifying the need for clear distinctions among different kinds of attention, and then emphasizes the difference between internal and external attention. Next, evidence from behavioral, neuropsychological, and neuroimaging studies suggesting that internal attention is required for the successful retrieval of memorial contents is reviewed. In turn, it is argued that internal attention during recollection is what makes us conscious of the contents of retrieved memories; further evidence in support of this claim is also provided. Finally, it is suggested that internal attention is probably not sufficient for conscious recollection. Open questions and possible avenues for future research are also mentioned.Most research on the relationship between attention and consciousness has been limited to perception. However, perceptions are not the only kinds of mental contents of which we can be conscious. An important set of conscious states that has not received proper treatment within this discussion is that of memories. This paper reviews compelling evidence indicating that attention may be necessary, but probably not sufficient, for conscious recollection. However, it is argued that unlike the case of conscious perception, the kind of attention required during recollection is internal, as opposed to external, attention. As such, the surveyed empirical evidence is interpreted as suggesting that internal attention is necessary, but probably not sufficient, for conscious recollection. The paper begins by justifying the need for clear distinctions among different kinds of attention, and then emphasizes the difference between internal and external attention. Next, evidence from behavioral, neuropsychological and neuroimaging studies suggesting that internal attention is required for the successful retrieval of memorial contents is reviewed. In turn, it is argued that internal attention during recollection is what makes us conscious of the contents of retrieved memories; further evidence in support of this claim is also provided. Finally, it is suggested that internal attention is probably not sufficient for conscious recollection. Open questions and possible avenues for future research are also mentioned.

Consistent Belief in a Good True Self in Misanthropes and Three Interdependent Cultures

People sometimes explain behavior by appealing to an essentialist concept of the self, often referred to as the true self. Existing studies suggest that people tend to believe that the true self is morally virtuous; that is deep inside, every person is motivated to behave in morally good ways. Is this belief particular to individuals with optimistic beliefs or people from Western cultures, or does it reflect a widely held cognitive bias in how people understand the self? To address this question, we tested the good true self theory against two potential boundary conditions that are known to elicit different beliefs about the self as a whole. Study 1 tested whether individual differences in misanthropy-the tendency to view humans negatively-predict beliefs about the good true self in an American sample. The results indicate a consistent belief in a good true self, even among individuals who have an explicitly pessimistic view of others. Study 2 compared true self-attributions across cultural groups, by comparing samples from an independent country (USA) and a diverse set of interdependent countries (Russia, Singapore, and Colombia). Results indicated that the direction and magnitude of the effect are comparable across all groups we tested. The belief in a good true self appears robust across groups varying in cultural orientation or misanthropy, suggesting a consistent psychological tendency to view the true self as morally good.

Event-related functional magnetic resonance imaging changes during relational retrieval in normal aging and amnestic mild cognitive impairment.

The earliest cognitive deficits observed in amnestic mild cognitive impairment (aMCI) appear to center on memory tasks that require relational memory (RM), the ability to link or integrate unrelated pieces of information. RM impairments in aMCI likely reflect neural changes in the medial temporal lobe (MTL) and posterior parietal cortex (PPC). We tested the hypothesis that individuals with aMCI, as compared to cognitively normal (CN) controls, would recruit neural regions outside of the MTL and PPC to support relational memory. To this end, we directly compared the neural underpinnings of successful relational retrieval in aMCI and CN groups, using event-related functional magnetic resonance imaging (fMRI), holding constant the stimuli and encoding task. The fMRI data showed that the CN, compared to the aMCI, group activated left precuneus, left angular gyrus, right posterior cingulate, and right parahippocampal cortex during relational retrieval, while the aMCI group, relative to the CN group, activated superior temporal gyrus and supramarginal gyrus for this comparison. Such findings indicate an early shift in the functional neural architecture of relational retrieval in aMCI, and may prove useful in future studies aimed at capitalizing on functionally intact neural regions as targets for treatment and slowing of the disease course. (JINS, 2012, 18, 1-12).

Predictive memory and the surprising gap.

Clark (in press) has offered a forceful defense of the “hierarchical prediction machine” (HPM) approach to the brain. Roughly, HPM suggests that brains are in the business of making sense of incoming information by generating top-down models aimed at providing the optimal fit for the input data. A better fit between the model and the data minimizes prediction error, which Clark – following Friston (e.g., Friston, 2010) – construes as tantamount to reducing surprisal, i.e., “the sub-personally computed implausibility of some sensory state given the model of the world” (p. 17). Notwithstanding the breadth of his defense, Clark’s case is entirely built upon research on perception, attention, and action, all of which are on-line cognitive processes. With practically no mention of offline cognition, the theoretical pretensions of the HPM approach, which Clark so vigorously defends as a “single unifying explanatory framework” (p. 61) in cognitive science, are questionable. I suggest that this conspicuous absence might be partially remedied, at least for the case of remembering, by looking at recent Bayesian accounts of memory retrieval developed after Anderson’s Adaptive Control of Thought-Rational (ACT-R) model (Anderson and Milson, 1989; Anderson, 1990; Anderson and Schooler, 1991, 2000). Specifically, I suggest that the ACT-R model can be read as describing how memory retrieval attempts to minimize prediction error when finding the optimal memory given the costs of its retrieval and the organism’s current needs. Originally, the ACT-R model stated that remembering is a cognitive operation whose costs are offset by the gains attained when retrieval is successful. As such, our adaptive memory system would search for a particular memory as long as the probability of recovering it given our current needs is greater than the costs of its retrieval. The ACT-R model captures this insight in Bayesian terms thus: let Hi be the hypothesis that a particular memory is needed during a particular context, and let E be the evidence for an element of said context. Then, P(Hi|E)αP(E|Hi)P(Hi) (1) where P(E|Hi) determines the likelihood ratio that E is the case given Hi (i.e., the context factor), and P(Hi) gives the prior probability that a particular memory will be needed (i.e., the history factor). For present purposes, two consequences that follow from this formulation are relevant. First, as Anderson and Milson (1989) remarked, given the multiplicity of elements present in a retrieval context, the likelihood ratio representing the context factor is best understood as the multiplicative product of all the likelihood ratios for every element of the context given Hi1. As a result, certain contextual elements are going to be better cues than others (i.e., representing a larger positive contribution to the overall product), as it is the case with elements that were present in the context of encoding (Craik and Tulving, 1975). The second thing to notice is that the prior probability, according to the ACT-R model, depends on the history of previous retrievals. Originally, Anderson and Milson (1989) noted that determining the history factor could be daunting, if not impossible, as one “would have to follow people about their daily lives, keeping a complete record of when they use various facts [and] such an objective study of human information is close to impossible” (p. 705). To get around this problem, Anderson and Schooler (1991) suggested extracting prior probabilities from the statistical distribution of existent databases that, according to them, would capture “coherent slices of the environment.” One such environmental database, for instance, contained 2 years worth of word usage in the New York Times headlines. They found that the odds that a particular word was used in a certain headline was inversely correlated to its having occurred in a previous headline, with the probability diminishing the more time had passed since its last usage. Importantly, Anderson and Schooler (1991) showed that this model could fit extant data on recency and frequency effects on memory retrieval remarkably well2. Taken together, the context and the history factors suggest that the probability that a certain memory will be needed in a particular context can be predicted from the probability that it has been needed in the recent past in relevantly similar contexts. From the point of view of Clark’s HPM approach then, context and history factors combine in a hierarchical model that tries to find the most predictable memory – i.e., that which minimizes prediction error – for a needed memory given a cue3. Notwithstanding Anderson and Schooler’s impressive results, priors based on statistical distributions of limited environments do not seem to capture the full complexity of human memory retrieval. Recently, however, Hemmer and Steyvers (2009b, see also Hemmer and Steyvers, 2009a) tried a different tack. They obtained the prior probability of remembering the size of a certain object from the statistical distribution of participant’s responses on a norming phase, in which relative size judgments on a number of objects had to be performed. Thus, instead of determining the prior by collecting data from the participant’s size-judgment behavior before the study, they did it via generating a probability distribution from the participants’ judgments themselves. This model allowed them to predict with remarkable accuracy hits and false alarms in a recognition test, as participant’s responses approximated the means of the prior distribution for each item. So far I’ve tried to draw parallels between ACT-R inspired Bayesian models on memory retrieval and Clark’s HPM approach as a way to show that his explanatory framework can be extended to an offline cognitive process such as remembering. But in so doing I intentionally drew a stark contrast between the way in which Anderson and Schooler obtained prior probabilities and likelihood rates, and the way in which Hemmer and Steyvers did, because I think this difference illustrates a difficult challenge for the HPM framework. According to Clark, although the HPM approach is primarily thought of as describing the way in which the brain aims at reducing surprisal at anon-agential level, these very same processes may help to understand the agent-level experience of surprise reduction – the experience of sensing a stimulus as the least surprising (“surprisal-ing”!, p. 47). But this agent/non-agent gap may be difficult to bridge. Indeed, Clark himself acknowledges this difficulty when he says: “[T]here seems to be a large disconnect between ‘surprisal’ (the implausibility of some sensory state given a model of the world) and agent-level surprise. This is evident from the simple fact that the percept that, overall, best minimizes surprisal (hence minimizes prediction errors) ‘for’ the brain may well be, for me the agent, some highly surprising and unexpected state of affairs” (p. 46). Nonetheless, Clark (in press) believes that the two levels “are easily reconciled” when one recognizes that what appears to the agent as a surprising event may just be, in reality, only improbable. The agent might not have been expecting to experience some mental content or another, but from the point of view of the brain, such a content may actually be perfectly predictable. I find Clark’s response unsatisfying, for this surprise-surprisal gap – this “surprising gap” – between the agent and the non-agent levels is likely to occur more often than Clark assumes, and the frequency of this occurrence puts pressure on Clark to come up with a clearer explanation as to how HPM can in fact illuminate cognition at the agent-level. Consider the two approaches to generating prior probabilities and likelihood ratios mentioned above. In the case of Anderson and Schooler, the approach is agent-independent, as it involves collecting probability distributions of frequency responses that are independent of the subject’s own frequency-judgments. Likewise, priors generated from data at the neural level, such as those referenced by Clark in his essay, are also agent-independent. Conversely, Hemmer and Steyver’s approach is paradigmatically agent-dependent, as it involves generating a probability distribution from the participant’s own frequency-judgments. However, we have plenty of evidence showing that what we think is most frequent does not always correspond to what it is actually most frequent (Tversky and Kahneman, 1973, but see Manis et al., 1993). Moreover, the agent/non-agent mismatch that gives rise to this “surprising gap” may actually occur even when there is no experience of surprise at the agent-level. It may occur, for instance, when there is a prediction mismatch due to independent processes of prior updating at the agential and non-agential levels. As a result, although models with agent-independent priors may be equally good at fitting data as models with agent-dependent priors, they need not be, and it is an open empirical question whether or not they do – a question that cannot be simply dismissed on a priori grounds, as Clark does. So it seems to me that studying this surprising gap is itself an exciting avenue for future research. Why are there percepts that may appear surprising to the agent? What are the conditions under which surprise reduction meets surprisal reduction? Are false alarms in perception or in recognition memory better predicted with agent-dependent or agent-independent priors? These, I think, are all interesting questions worthy of being examined, and for which the HPM needs to find an answer if it really attempts to be a “unifying explanatory framework” for both agent and non-agent level cognitive phenomena4.where P(E|H i ) determines the likelihood ratio that E is the case given H i (i.e., the context factor), and P(H i ) gives the prior probability that a particular memory will be needed (i.e., the history factor). For present purposes, two consequences that follow from this formulation are relevant. First, as Anderson and Milson (1989) remarked, given the multiplicity of elements present in a retrieval context, the likelihood ratio representing the context factor is best understood as the multiplicative product of all the likelihood ratios for every element of the context given H i . As a result, certain contextual elements are going to be better cues than others (i.e., representing a larger positive contribution to the overall product), as it is the case with elements that were present in the context of encoding (Craik and Tulving, 1975). The second thing to notice is that the prior probability, according to the ACT-R model, depends on the history of previous retrievals. Originally, Anderson and Milson (1989) noted that determining the history factor could be daunting, if not impossible, as one “would have to follow people about their daily lives, keeping a complete record of when they use various facts [and] such an objective study of human information is close to impossible” (p. 705). To get around this problem, Anderson and Schooler (1991) suggested extracting prior probabilities from the statistical distribution of existent databases that, according to them, would capture “coherent slices of the environment.” One such environmental database, for instance, contained 2 years worth of word usage in the New York Times headlines. They found that the odds that a particular word was used in a certain headline was inversely correlated to its having occurred in a previous headline, with the probability diminishing the more time had passed since its last usage. Importantly, Anderson and Schooler (1991) showed that this model could fit extant data on recency and frequency effects on memory retrieval remarkably well. Taken together, the context and the history factors suggest that the probability that a certain memory will be needed in a particular context can be predicted from the probability that it has been needed in the recent past in relevantly similar contexts. From the point of view of Clark’s HPM approach then, context and history factors combine in a hierarchical model that tries to find the most predictable memory – i.e., that which minimizes prediction error – for a needed memory given a cue. Clark (in press) has offered a forceful defense of the “hierarchical prediction machine” (HPM) approach to the brain. Roughly, HPM suggests that brains are in the business of making sense of incoming information by generating top-down models aimed at providing the optimal fit for the input data. A better fit between the model and the data minimizes prediction error, which Clark – following Friston (e.g., Friston, 2010) – construes as tantamount to reducing surprisal, i.e., “the subpersonally computed implausibility of some sensory state given the model of the world” (p. 17). Notwithstanding the breadth of his defense, Clark’s case is entirely built upon research on perception, attention, and action, all of which are on-line cognitive processes. With practically no mention of offline cognition, the theoretical pretensions of the HPM approach, which Clark so vigorously defends as a “single unifying explanatory framework” (p. 61) in cognitive science, are questionable. I suggest that this conspicuous absence might be partially remedied, at least for the case of remembering, by looking at recent Bayesian accounts of memory retrieval developed after Anderson’s Adaptive Control of Thought-Rational (ACTR) model (Anderson and Milson, 1989; Anderson, 1990; Anderson and Schooler, 1991, 2000). Specifically, I suggest that the ACT-R model can be read as describing how memory retrieval attempts to minimize prediction error when finding the optimal memory given the costs of its retrieval and the organism’s current needs. Originally, the ACT-R model stated that remembering is a cognitive operation whose costs are offset by the gains attained when retrieval is successful. As such, our adaptive memory system would search for a particular memory as long as the probability of recovering it given our current needs is greater than the costs of its retrieval. The ACT-R model captures this insight in Bayesian terms thus: let H i be the hypothesis that a particular memory Predictive memory and the surprising gap