See Also: Book Notes, (me), Notes on Consciousness, LeDoux: Deep History of Conscious Brains, Seth: Being You, Dehaene: Consciousness and the Brain, Haidt: Happiness Hypothesis, Barrett: 7.5 Brain Lessons, Barrett: How Emotions Are Made, Sterling: Allostatis, Human: Makes Us Unique, Consciousness: Confessions, Blank Slate, Neuroscience of Human Relationships, Thinking, Fast and Slow

In Consciousness We Trust

The Cognitive Neuroscience of Subjective Experience by Hakwan Lau
BSP Interview about the book
This book is a good survey of various cognitive processes and failures of processes in the service of discussing three theories of consciousness: ‘global', local, and ‘higher-order theories (HOT)’. The structure of the book is to weigh the evidence as supporting either the local or the global theory while building the case for Lau's version of HOT, aka, PRM. It is a good read for consciousness nerds, but, probably start Consciousness reading with one of the books in the See Also list above. As a bonus, the Elephant and the Rider Metaphor makes an appearance.

As usual, these are notes for me to refer to and is in no way, shape or form a summary!!!

See more about theories of Consciousness here: An academic survey on theoretical foundations, common assumptions and the current state of consciousness science
Most Promising Theories: 1) ‘Predictive processing theory’, 2) ‘global neuronal workspace theory (GNWT)’, 3) ‘higher-order theories (HOT)’, 4) ‘local recurrency theory (LRT)’, 5) ‘integrated information theory (IIT)’

p.29 "We also went through four different notions of consciousness: 1) subjective experience, 2) access consciousness, 3) consciousness as the state an individual is in, and 4) consciousness as purposeful control. Subjective experience"
Perceptual Reality Monitoring, aka PRM.

One frustration I have with the book is that he never clearly articulates what his version of PRM really is.

p.23 "... if we define subjective experience as always having to do with sophisticated information processing, allowing rational access and control, we are automatically loading the dice in favor of global theories. If we define subjective experience as having decidedly nothing to do with these sophisticated information processes, we are likewise tilting the table in favor of local theories. This is why we cannot assume one way or the other from the outset. Nor can we end the debate by saying it is just a matter of definitions, followed by a shrug. These"

p.122 "The literature on nonconscious priming is vast... The question here is whether consciousness may be associated with higher cognitive functions. The global view says yes: consciousness is linked to the prefrontal and parietal cortices, where global coordination and exchange of information take place. Higher cognitive functions should be facilitated by these mechanisms. The local theorists say no: consciousness takes place within the sensory circuitries. Higher cognitive functions are downstream to consciousness. A percept may be nonconscious because it isn't accompanied by the right kind of local dynamics. But the signal may still be able to get out of the local sensory areas to impact the higher cognitive functions downstream."

Filling-in & Inflation

Hakwan Lau coined the term inflation for how details cover more area in the periphery. Filling-in is a more general term for how we fill-in missing or under sampled data. Other mechanisms are covered because p.99 "Multiple mechanisms are needed to account for the phenomenology of the unattended periphery"

p.92 "Perhaps the answer to the puzzle of apparent richness is that the details are never represented as such—neither in explicit nor compressed summary forms. Instead, we just interpret the sensory representations as if they are rich in detail, even when they are not (Knotts et al. 2019). This may not be such a misleading kind of misinterpretation in the end, because the details are actually out there in the world, often just one shift of gaze or attentional focus away. So this may be somewhat like the light inside the refrigerator: because whenever we check it, it is on, we may mistakenly think it is always on (Kouider, de Gardelle, and Dupoux 2007). We can call this an inflation account. That is, the richness of a perceptual representation may be overinterpreted or somehow exaggerated at a later stage."

p.97 "Although the mechanisms for inflation aren't fully fleshed out, one commitment of the account is clear: apparent richness is not supported by actual detailed sensory content. This is in contrast with another popular account: "filling-in," which suggests that the details missing from the input stage are visually reconstructed internally. Typically, filling-in accounts involve having the details reconstructed at the early sensory level via top-down influence. If that's right, perhaps a local view can account for the apparent richness of experience too. That is, the higher areas may be involved, but only as distal causes. Ultimately, the constitutive mechanisms may be all within the early visual areas."

p.118 " is also difficult to pinpoint what functions will always be compromised in nonconscious processing. A perceptual process can fail to be conscious in different ways (Block 2016). To get at this issue, we probably need a theoretical model outlining what really is the mechanism of consciousness. We probably need to understand how the different components interact to support what really is the mechanism of consciousness. We probably need to understand how the different components interact to support subjective experience. But even then, when that mechanism breaks down, there may be "back-up" systems. Understanding this is complicated business."

p.133 "... subjective experience seems rich, ... prefrontal mechanisms have limited processing capacity .. "overflown" by the richness of experience. But .. the role of the prefrontal cortex may not be to "duplicate" the sensory information. Rather, it may just monitor and redirect information in the sensory cortices, using something akin to indexing mechanisms. If so, the putative limited capacity may not be an issue."

Lau gives a quick nod to predictive coding, aka, predictive processng and quickly switches to Generative Adversarial Network (GAN)s

p.142 "This GANs architecture may be related to consciousness in several ways. First, it has been suggested on theoretical grounds that a discriminator-like mechanism may reside within the prefrontal cortex (Gershman 2019; Lau 2019). In support of this claim, there has been physiological evidence (Mendoza-Halliday and Martinez-Trujillo 2017) showing that neurons in the dorsolateral prefrontal cortex can distinguish between external perceptual content and endogenous generation of the same content (i.e., maintenance of the same information during working-memory delay). This is interesting because holding an image in working memory tends to induce somewhat similar activity in the early visual areas as normal perception (Harrison and"

p.154 "So this assertoric force seems strangely stubborn, in the way it has this tendency to inform us about what's happening here and now, regardless of what background beliefs we have. With effort we can resist believing in what we currently see, but things would still seem to us a certain way, given our conscious percepts. If we are anything like an optimal Bayesian system, the process of evidence accumulation is really not supposed to work this way. As such, perhaps we shouldn't expect nonconscious perception to ever behave like this either. This assertoric force may be a unique and curious feature of consciousness that calls for an explanation."

p.154 "However, for one to consciously perceive, one may additionally need to have certain higher-order states or representations. That is, the first-order states alone may drive visual behavior and performance. But without the relevant higher-order states, that would only constitute nonconscious perception, as in blindsight. The higher-order states may be reflected by activity in, for example, the prefrontal cortex. The content of these representations may be more conceptual, symbolic, and relatively sentence-like."

p.154 7.4 Higher-Order Thought or Beliefs?
"One way to account for the assertoric force discussed in Section 7.3 is to theorize that conscious perception always involves two sets of representations. We can call the state of early sensory activity the first-order state or first-order representation. This reflects the perceptual content (e.g., what objects are involved and in what spatial location, with specific features like colors, size, and motion direct). These representations are likely within the sensory cortices. We can say they are relatively picture-like, carrying analog content; we will discuss more what "analog" means in Section 9.5.

However, for one to consciously perceive, one may additionally need to have certain higher-order states or representations. That is, the first-order states alone may drive visual behavior and performance. But without the relevant higher-order states, that would only constitute nonconscious perception, as in blindsight. The higher-order states may be reflected by activity in, for example, the prefrontal cortex. The content of these representations may be more conceptual, symbolic, and relatively sentence-like.

What may be the specific content of the higher-order states? Given the discussion in the Section 7.3, one may be tempted to think that these higher-order states could be the corresponding perceptual beliefs. That is, the first-order state contains t he picture-like sensory information, for example, about a cat in front of us. For us to consciously see the cat, we need to have the higher-order belief that there is the cat in front of us. This belief can then guide our rational decision-making."

p.157 "Let's assume that an agent is capable of reasoning with beliefs and goals. If the mechanisms for this kind of general symbolic-level cognition receive input from both the first-order states and the discriminator, one can see how conscious perception can attain its assertoric force. Essentially, conscious perception happens when the first-order state represents the cat, and the higher-order (discriminator) state indicates that the first-order state is a true reflection of the world right now. Together, these two representations constitute something akin to the premises of a syllogistic inference. In the absence of conflicting background beliefs, it is rational to form the belief that there is a cat in front of us right now. Such a belief is in a sense "justified" by the premises; it logically follows (Figure 7.1)."

p.158 "Against the higher-order thought view, there may be some concerns regarding whether the thought-like higher-order representation can capture the richness of perceptual experience. As reviewed in Chapter 4, the issue is controversial. But a standard higher-order thought theorist is committed to a relatively "sparse" view, on which perceptual experience is no richer than what can be captured by conceptual, thought-like representations. The view argued for here makes no such commitment; the rich content of the first-order state also contributes. The higher-order state does not duplicate the first-order content, but merely serves as a gating mechanism to direct the first-order information to the relevant downstream processes.

The reader may wonder: how does the higher-order (discriminator) state refer to the corresponding first-order state, without duplicating its content? In current artificial GANs, we tend to deal with just one first-order state at a time. But in the actual human brain, there may be multiple concurrent perceptual states in different sensory modalities. At a given time, some of these first-order states may lead to subjective experience while others may not. As such, there are likely multiple discriminator outputs, and one needs to keep track of which refers to which first-order states. Of relevance is that indexing or variable binding mechanisms have been proposed for prefrontal functions (Kriete et al. 2013). Essentially, the prefrontal cortex must have some ways of referring to specific first-order activities via some form of "addressing" system.

For a simplistic analogy, we can think of this as a phone numbers system, where each individual referent is given a unique identifier. So, as in modern computational systems, a higher-order mechanism can refer to first-order representations by these addresses, without duplicating or redescribing the full content. In Chapter 9 we will revisit how such mechanisms may actually work in the mammalian sensory cortices."

p.159 for addressing to work it must access both "higher-order and first-order Content. On the view advocated here, the idea is that these contents are subsequently read out by the mechanisms for general symbolic-level cognition and logical reasoning. In this sense, consciousness is the gating mechanism by which perception impacts cognition; it selects what perceptual information should directly influence our rational thinking.

Note that this is not to identify consciousness with access consciousness (as discussed in Section 1.6). Consciousness here refers to subjective experience, as we do throughout most of this book. The point is that subjective experiences are causally connected with access consciousness in the ways described above. Subjective experiences are characterized by their availability for potential conscious access. But I'm not suggesting that the two are one and the same. When the discriminator decides that a first-order representation correctly represents the world right now, global broadcast and access are likely to happen. But these consequences are not constitutively part of the subjective experience, according to this view."

p.160 "According to PRM, there is another possibility why aphantasia or weak imagery experience may happen. We have argued that the neural mechanisms for the discriminator may also contribute to metacognition. So in total, this discriminator-like mechanism has three different output conditions. That is, a first-order state is one of the following: i) externally triggered, ii) internally generated, or iii) just noise. The first condition should lead to normal subjective perceptual experience, and that the third condition should entail the lack of subjective experience. When the discriminator decides that a certain first-order state is internally generated (second condition), a distinct output is needed. Whether this output is more similar to the first or the third condition may vary across people; when we say two outputs are similar, we mean that they are more easily confused to be the same by downstream readout. To the extent that it is more similar to the third condition (noise) rather than the first (externally triggered), PRM predicts that subjective experience may be absent or relatively feeble too—even if the first-order activity is actually robust."

p.87 "Much of the supporting evidence for Load Theory comes from behavioral studies (Lavie 2005). But in some experiments, functional magnetic resonance imaging (fMRI) was employed to assess how perceptual load at a central task may impact the processing of some peripheral, task-irrelevant stimuli (Schwartz et al. 2005). As predicted by the theory, under high load, there was less brain activity associated with the processing of task-irrelevant stimuli, as if such stimuli were filtered out early on. Interestingly, this effect was observed as early as in V1 (the primary visual cortex), where neurons respond to specific orientations of line segments, among other low-level features. This means that when the perceptual task is challenging enough, attentional filtering may happen even earlier than early selectionists suggested. It seems to take place before the processing of "meaning," as object identity is typically thought to be represented in area IT (inferotemporal cortex), well after V1 processing."


p.180 ""conscious." So, like the political theorists, they also often use the term consciousness to refer to our rational (or rationalized) narratives. But psychoanalysts suggest that the "unconscious" mind is just as capable of sophisticated forms of reasoning, as if it also has its own independent narrative system of some sort. Nevertheless,psychoanalysis has not become an empirically successful sci ence. This leaves us doubting if there really are such things as fully fledged "nonconscious" narrative mechanisms. The view I propose is that there isn't. However, on a more practical level, these two questions are also left open: are there "nonconscious" ways to change our "conscious" narratives? And how about the opposite: can our "conscious" narratives impact "nonconscious" processes too? Below I will argue positively for both: our "conscious" narratives and at least some "nonconscious" processes influence each other. This is true especially for affective processes, which often seem to happen outside of our conscious control. This may in a sense suggest that the psychoanalysts aren't entirely wrong. Not only does this matter for our theory of consciousness, but also these questions have clinical and societal implications too."

p.181 "This is another reason why understanding this narrative system is so important. Philosophical theories often link our sense of free will and moral responsibility to consciousness. But it is unclear if subjective experience per se is the relevant notion (Levy 2014). Instead, our overall sense of agency may arise at the narrative level. This may be why it is relatively malleable (Chapter 5 Section 5.3). And yet, it does not mean that our sense of agency is necessarily always constructed after the facts, as a mere illusion. That is because the narrative system can have causal impacts on both our subjective experiences and rational behavior. As such, disturbances in this sense of agency can have devastating effects, as, for example, observed in psychosis."

p.184 "this breaks down into three key points: i) emotions influence narrative processing; ii) narratives influence emotions; and iii) emotions can be contagious. That is to say, for a patient suffering from psychosis in modern societies, it may be difficult to form a very positive self-narrative because one usually goes through significant emotional distress. Some of this negative affect may come from other people, who do not see the situation in a very positive light. But all the same, the resulting emotion may also influence the patient's own feelings toward their condition. Let us unpack these points.

The first point may be straightforward. The narratives we've been discussing are often self-narratives. When we feel afraid, typically the only immediate belief that we can form is just that we are afraid; we may not know why or what really causes our fear. The causal analysis may take place at least in part at the narrative level. In seeking coherence, we tend not to accept easily that we are frightened for no reason. Instead, the mind seeks plausible explanations.

The social psychologist Jonathan Haidt has given the analogy that our "conscious" decision-making is a bit like riding an elephant (2012). As "rational" agents, we'd like to think we are in control of our emotions. But the slow and powerful animal we are riding has its own mind. We can nudge it toward our desired directions here and there. But in the end, we are often just going along for the ride. We make up post hoc rationalizations to justify our "decisions." But maybe emotions are really the driving forces."

Rider & Elephant Metaphor

(See previous paragraph) Total bonus for using Haidt's Elephant And The Rider Metaphor. On the next page, Lau also compares conscousness to Daniel Kahneman's Thinking, Fast and Slow. So, is the autoplilot part fo conaciousnes? This beautiful New England fall day I was driving, listening to music and rehearsing a conversation in my head at the same time. How does that meash with Subjective Experience?

p.203 "Recall that according to PRM, consciousness is in a sense the interface between perception and cognition. It selects the deserving first-order states for direct impact on higher-cognitive processing at the symbolic level. On this level, thoughts and beliefs are expressed in formats somewhat akin to sentences. As such, noisy sensory inputs are better filtered out, as they may cause dramatic and unpredictable errors on this higher level—they can cause multiplicative errors. Real percepts and our own imagination must be also delineated clearly as such early on, as they have vastly different implications for reasoning."

p.209 "in other words the various different brain regions likely work in concert in support of autobiographical, subjective narrative processing. Within the hippocampus, it is known that the storage of long-term memories does not take the form of a frame-by-frame detailed video- like recording. Instead, one enduring idea is that some kind of index system is used for efficient storage (Teyler and Rudy 2007; Tanaka and McHugh 2018). With these indexes one can retrieve the sensory details from the representations throughout the cortex. This suggests that the "addressing" system proposed earlier for the prefrontal mechanisms to refer to early sensory activity may not be unique."

p.210 "Perhaps it makes sense for downstream areas to all refer to these same addresses or indexes when they communicate with each other about the relevant sensory content. For an analogy, this is a bit like the way we pass hyperlinks for Internet webpages in emails, without duplicating the detailed content—except that here the "hyperlinks" themselves are structured enough that we know similar links will take us to subjectively similar contents. In a sense, this means that the different downstream brain regions communicate with an internal phenomenal "language": When the prefrontal cortex signals to the hippocampus that "this" sensory activity reflects the state of the world right now, the hippocampus "knows" that what should be encoded into our narratives is a sensory stimulus that looks like something, and yet unlike something else, for example. This may be how the qualitative nature of subjective experience comes about. The primary function of reality monitoring is for routing sensory information, to direct such information toward appropriate downstream symbolic-level processing. But, in doing so, the system implicitly knows what the stimulus in question is subjectively like. Because this "knowledge" is implicit, it may be difficult for the subject to articulate it. But it is part of the language through which our different brain mechanisms communicate."

Other notes:

p.139 - Axel Cleeremans' self-organizing metarepresentational (definition) of consciousness: (Cleeremans et al. 2020). According to the view, subjective experience arises when the brain learns to represent its own sensory states for the purpose of hierarchical control of perceptual information and action.

The topological space of subjective experience (alt.URL) Catherine Tallon-Baudry:
Emotional feelings share at least two di- mensions, valence (positive/negative) and arousal (relaxing/exciting)
Perceptual dimensions differentiate emotions "Participants in two studies (one where respondents reported on abstract emotion concepts and a second where they reported on specific emotion episodes) rated the extent to which features anchoring 29 perceptual dimensions (e.g., temperature, texture and taste) are associated with 8 emotions (anger, fear, sadness, guilt, contentment, gratitude, pride and excitement). "
Revealing the multidimensional mental representations of natural objects underlying human similarity judgements

p.XX "That is, the information in the early sensory states are being redescribed at a later stage, for self-monitoring purposes."

Nit: p.202, Lau claims a qualitative different twixt & HSV ( - See:
typo p.57 claustrumc

Metaphor for Frontal Pole - Railroad switch yard for trains of thought. It selects which train will be the dominant coalition.

2022-10-19 <> jch