Quoted text from (Grossberg 2021)

Table of Contents

• Grossberg OR[anticipated, predicted, unified] the [experimental result, model]s
• What is consciousness?
• How does one evolve a computational brain?
• Grossberg's comments for some well-known consciousness theories

Grossberg OR[anticipated, predicted, unified] the [experimental result, model]s

• Grossberg 2021 p045c2h0.85 ART's predicted links between attention and synchronous oscillations have been experimentally supported by experiments reported by (edited list) :
  • (Timothy Bushman, Earl Miller 2007)
  • (Andreas Engel, Pascal Fries, Wolf Singer 2001)
  • (Georgia Gregoriou, Stephen Gotts, Huilhui Zhou, Robert Desimone 2009)
  • (Georgia Gregoriou, Robert Desimone, Anfrew Rossi, Leslie Ungerleider 2014)
  • (Daniel Pollen 1999)
• Grossberg 2021 p046c1h0.05 Of particular interest are data supporting ART's predicted link between synchronous oscillations and consciousness that neuroscientists such as ;
  • (Victor Lamme 2006)
  • (Rudolfo Llinas, Ribrary, Contreras, Pedroarena 1998)
  • (Wolf Singer 1998)
  • and their colleagues
  have reported.
• Grossberg 2021 p063c1h0.45 The fact that LTM (Long Term Memory) traces in the brain exhibit gated steepest descent properties started to be confirmed experimentally by neurophysiologists like Wolf Singer and William "Chip" Levy 20 years later :
  • (Levy, Brassel, Moore 1983)
  • (Levy, Desmond 1985)
  • (Rauschecker, Singer 1979)
  • (Singer 1983)
• Grossberg 2021 p229c1h0.50 (Scott Brincat, Earl Miller 2015) reported neurophysiological data that support the ART distinction between category learning within the attentional system, that includes prefrontal cortex (PFC), and the orienting system, that includes the hippocampus (HPC). ... These results contribute to a long history of neurobiological experiments that have implicated the hippocampus in mismatch processing, notably the processing of novel events, including experiments and theoretical concepts about habituation of novelty in the hippocampus as learning proceeds. ... The ART hypothesis testing and category learning cycle hereby clarifies and unifies an important series of psychophysiological experiments that go back at least 50 years.
• Grossberg 2021 p229c2h0.60 SMART computer simulations demonstrate that a good enough match of a top-down expectation with a bottom-up feature pattern generates an attentive resonance during which the spikes of active cells synchronize in the gamma frequency range of 20-70 Hz (Figure 5.40). Many labs have reported a link between attention and gamma oscillations in the brain, including two articles published in 2001, one from the laboratory of Robert Desimone when he was at the the National Institute of Mental Health in Bethseda (Fries, Reynolds, Rorie, Desimone 2001), and the other from the laboratory of Wolf Singer in Frankfurt (Engel, Fries, Singer 2001). You'll note that Pascal Fries participated in both studies, and is an acknowledged leader in neurobiological studies of gamma oscillations; eg (Fries 2009). .."
• Grossberg 2021 p067c1h0.60 Teuvo Kohonen at Helsinki University also made a transition from linear algebra concepts, such as the Moore-Penrose pseudo-inverse (eg Albert 1972), to more biologically motivated models that include nonlinear interactions. In the latter category, Kohonen used a simplified version of a Self-Organizing Map, or SOM, model in various applications (eg Kohonen 1984). ... The SOM had earlier been introduced and developed in the early 1970s by myself working in Cambridge and Boston, and Christof von der Malsburg, who was then at the Max-Planc Institute for Biophysical Chemistry in Gottingen(Grossberg 1972c, 1976a, 1976b; Malsburg 1973). I had even earlier introduced and mathematically analyszed the learning laws and competitive networks that go into the SOM model, starting in the 1960s (eg Grossberg 1968c, 1969c, 1971b, 1973, 1974).
• Grossberg 2021 p204c2h0.30 Models that combine instar learning and competition are often called competitive learning or self-organizing map models. Such models were introduced and computationally characteried by Christoph von der Malsburg and myself between 1972-1978 (Grossberg 1972c, 1976a, 1978a; von der Malsburg 1973; Willshaw, Malsburg 1976). They were subsequently applied and further developed by many authors, notably Teuvo Kohonen, whose 1984 book on the subject, and its subsequent editions, helped to popularize them (Kohonen 1984).

What is consciousness?

• Grossberg 2021 page xii I will explain that "all conscious states are resonant states". The importance of this assertion motivated the title of this book. I will describe the resonances that seem to underlie our conscious experiences of seeing, hearing, feeling, and knowing. These explanations will include where in the brain these resonances take place, how they occur there, and why evolution may have been driven to discover conscious mental states in the first place. I will also clarify how these resonances interact when we simultaneously see, hear, feel, and know something, all at once, about a person or na event in our world. This description is part of a burgeoning classification of resonances. I will also explain why not all resonant states are conscious, and why not all brain dynamics are resonant. These results contribute to solving what has been called the Hard Problem of Consciousness.
• Grossberg 2021 p040c2h0.80 sub-section: Why did consciousness evolve? In particular, the book will show how and why multiple processing stages are needed before the brain can construct a complete and stable enough representation of the information in the world with which to control effective behaviours. This happens because the sensory signals that our brains process are often noisy and ambiguous, as are the representations that our brains form in response to them at the earliest processing stages. Complementary computing and hierarchical resolution of uncertainty overcome these problems until perceptual representations that are sufficiently complete, context-sensitive, and stable can be formed. The brain regions where these representations are completed are different for seeing, hearing, feeling, and knowing. But then how do our brains select these representations? My proposed answer is: A resonant state is generated that selectively "lights up" these representations and thereby renders them conscious. These conscious representations can then be used to trigger effective behaviours.
  
  Consciousness hereby enables our brains to prevent the noisy and ambiguous information that is computed at earlier processing stages from triggering actions that could lead to disastrous consequences. Conscious states thus provide an extra degree of freedom whereby the brain ensures that its interactions with the environment, whether internal or external, are effective as possible, given the information at hand. These conclusions will be supported by theoretical analysis throughout the book that enable explanations of many psychological and neurobiological data about normal and clinical behaviours that have no other mechanistic explanations at the present time.

How does one evolve a computational brain?

• Grossberg 2021 p081-83 sub-section: How does one evolve a computational brain?
  • Behavioral Success drives::>> Behavioral Data -> Art of Modelling -> Design Principles -> Mathematical Models -> Method of Minimal Anatomies (Occam's Razor) -> Behavioral Predictions -> explain Neural Data -> novel Brain Predictions <<::links behavior-mind-brain
  • "burn the candle from both ends" by pressing both top-down from Behavioral Data and bottom-up from Brain Data to clarify what the model can and cannot explain at its current stage of derivation. No model can explain everything.
  • >characterize the boundary between the known and unknown: the shape of this boundary helps to direct attention to new design principles
  (keys : behavior-mind-brain link)
• Grossberg 2021 p081-83 The cycle of model evolution :
  • The next step is to show how these new design principles can be incorporated into the evolved model in a self-consistent way, without undermining its previous mechanisms, thereby leading to a progressively more realistic model, one that can explain and predict ever more behavioral and neural data.
  • In this way, the model undergoes a type of evolutionary development, as it becomes able to cope behaviorally with environmental constraints of ever increasing subtlety and complexity.
  • The Method of Minimal Anatomies may hereby be viewed as way to functionally understand how increasingly demanding combinations of environmental pressures were incorporated into brains during the evolutionary process.
  (keys : brain evolution, cycle of model evolution)
• Grossberg 2021 p081c2h0.66 The cycle of model evolution has been carried out many times since 1957, leading today to increasing numbers of models that individually can explain and predict psychological, neurophysiological, anatomical, biophysical, and even biochemical data. In this specific sense, the classical mind-body problem is being incrementally solved
  (keys : mind-body -classical- problem)
• Grossberg 2021 p081c2h0.66 sub-section: How does one evolve a computational brain?
  The above discussion illustrates that no single step of theoretical derivation can derive a whole brain. One needs a method for deriving a brain in stages, or cycles, much as evolution has incrementally discovered ever more complex brains over many thousands of years. The following theoretical method has been successfully applied many times since I first used it in 1957. It embodies a kind of conceptual evolutionary process for deriving a brain.
  
  Because "brain evolution needs to achieve behavioural success", we need to start with data that embodiey indices of behavioral success. That is why, as illustrated in Figure 2.37 Modelling method and cycle, one starts with Behavioral Data from scores or hundreds of psychological experiments. These data are analyszed as the result of an individual adapting autonomously in real time to a changing world. This is the Arty of Modeling. It requires that one be able to infer from static data curves the dynamical processes that control individual behaviors occuring in real time. One of the hardest things that I teach to my students to do is "how to think in real time" to be able to carry out this speculative leap.
  
  Properly carried out, this analysis leads to the discovery of new Design Principles that are embodied by these behavioral processes. The Design Principles highlight the functional meaning of the data, and clarify how individual behaviors occurring in real time give rise to these static data curves.
  
  These principles are then converted into the simplest Mathematical Model using a method of minimal anatomies, which is a form of Occam's Razor, or principle of parsimony. Such a mathematical model embodies the psychological principles using the simplest possible differential equations. By "simplest" I mean that, if any part of the derived model is removed, then a significant fraction of the targeted data could no longer be explained. One then analyzes the model mathematically and simulates it on the computer, showing along the way how variations on the minimal anatomy can realize the design principles in different individuals or species.
  
  This analysis has always provided functional explanations and Behavioral Predictions for much larger behavioral data bases than those used to discover the Design Principles. The most remarkable fact is, however, that the behaviorally derived model always looks like part of a brain, thereby explaining a body of challenging Neural Data and making novel Brain Predictions.
  
  The derivation hereby links mind to brain via psychological organizational principles and their mechanistic realization as a mathematically defined neural network. This startling fact is what I first experienced as a college Freshman taking Introductory Psychology, and it changed my life forever.
  
  I conclude from having had this experience scores of times since 1957 that brains look the way they do because they embody a natural computational realization for controlling autonomous adaptation in real-time to a changing world. Moreover, the Behavior -> Principles -> Model -> Neural derivation predicts new functional roles for both known and unknown brain mechanisms by linking the brain data to how it helps to ensure behavioral success. As I noted above, the power of this method is illustrated by the fact that scores of these predictions about brain and behavior have been supported by experimental data 5-30 years after they were first published.
  
  Having made the link from behavior to brain, one can then "burn the candle from both ends" by pressing both top-down from Behavioral Data and bottom-up from Brain Data to clarify what the model can and cannot explain at its current stage of derivation. No model can explain everything. At each stage of development, the model can cope with certain environmental challenges but not others. An important part of the mathematical and computational analysis is to characterize the boundary between the known and unknown; that is which challenges the model can cope with and which it cannot. The shape of this boundary between the known and unknown helps to direct the theorist's attention to new design principles that have been omitted from previous analysis.
  
  The next step is to show how these new design principles can be incorporated into the evolved model in a self-consistent way, without undermining its previous mechanisms, thereby leading to a progressively more realistic model, one that can explain and predict ever more behavioral and neural data. In this way, the model undergoes a type of evolutionary development, as it becomes able to cope behaviorally with environmental constraints of ever increasing subtlety and complexity. The Method of Minimal Anatomies may hereby be viewed as way to functionally understand how increasingly demanding combinations of environmental pressures were incorporated into brains during the evolutionary process.
  
  If such an Embedding Principle cannot be carried out - that is, if the model cannot be unlumped or refined in a self-consistent way - then the previous model was, put simply, wrong, and one needs to figure out which parts must be discarded. Such a model is, as it were, an evolutionary dead end. Fortunately, this has not happened to me since I began my work in 1957 because the theoretical method is so conservative. No theoretical addition is made unless it is supported by multiple experiments that cannot be explained in its absence. Where multiple mechanistic instantiations of some Design Principles were possible, they were all developed in models to better underestand their explanatory implications. Not all of these instantiations could survive the pressure of the evolutionary method, but some always could. As a happy result, all earlier models have been capable of incremental refinement and expansion.
  
  The cycle of model evolution has been carried out many times since 1957, leading today to increasing numbers of models that individually can explain and predict psychological, neurophysiological, anatomical, biophysical, and even biochemical data. In this specific sense, the classical mind-body problem is being incrementally solved.
  
  Howell: bold added for emphasis.
  (keys : Principles-Principia, behavior-mind-brain link, brain evolution, cycle of model evolution)
  see also quotes: Charles William Lucas "Universal Force" and others (not retyped yet).

Grossberg's comments for some well-known consciousness theories

• Grossberg 2021 p047c2h0.20 the neural global workspace model that was published in 2014 by Stanislas Dehaene (Dehaene 2014) builds upon the global workspace of Bernard Baars (Baars 2005) and claims that "consciousness is global information broadcasting within the cortex [to achieve] massive sharing of pertinent information throughout the brain" (p13). Dehaene also makes a number of other useful observations, including that "the time that our conscious vision spends entertaining an interpretation is directly related to its likelihood, given the sensory information received" (p97) and that "the conditioning paradigm suggests that consciousness has a specific evolutionary role: learning over time, rather than simply living in the instant. The system formed by the prefrontal cortex and its interconnected areas, including the hippocampus, may serve the essential role of bridging temporal gaps" (p103). Such claims are consistent with the analysis in this book, but they do not describe the underlying organizational principles, neural mechanisms, or brain representations that embody subjective conscious aspects of experience.
• Grossberg 2021 p048c1h0.25 In 2012 and 2015, Tononi further develops postulates for his Integrated Information Theory (IIT) for physical systems that include consciousness (Tononi 2012, 2015). These postulates are intrinsic existence, compositionality, information, integration, and exclusion, These postulates summarize some basic facts about consciousness, but do not explain them. p048c1h0.85 Both Dehaene and Tononi used the word "information" as a critical component of their hypothesis. But what is "information"? The scientific concept of "information" in the mathematical sense of Information Theory was defined and characterized in 1948 by the great American mathematician, electrical engineer, and cryptographer Claude Shannon (Shannon 1948). ... In order to even discuss what "information" is requires that a set of states exist whose information can be computed, and that fixed probabilities exist for transitions between these states. In contrast, the brain is a self-organizing system that continuously creates new states through development and learning, and whose probability structure is continually changing along with them. Without a theory that explains how their transition probabilities may change through time in response to changing environmental statistics, the classical concept of information is useless. How such states arise is a key explanatory target of ART, and is one reason why ART can offer a classification of the resonances that are proposed to embody specific conscious experiences.
• Grossberg 2021 p048c2h0.25 The influential philosopher of mind, Daniel Dennett, wrote a highly cited book in 1991 with the arresting title Conciousness Explained (Dennett 1991). In this book, Dennett argued against the a Cartesian Theater model; that is, a place in the brain where "it all comes together" and generates subjective judgements. Instead, Dennett advocated a Multiple Drafts model where discriminations are distributed in space and time across the brain, a concept that, without elaboration, is too vague to have explanatory power.
• Grossberg 2021 p048c2h0.90 Perhaps the theory of the highly influential Portugese-American neuroscientist and author, Antonio Damasio, comes closest to theoretically linking brain to mind in his beautifully written 1999 book with the title The Feeling of What Happens: Body and Emotion in the making of consciousness (Damasio 1999). ... Damasio used the cliical data that he elegantly summarized in his book to guide him to what is, in effect, a heuristic derivation of the Cognitive-Emotional-Motor, or CogEM, model that I had first published in 1991. ... Unlike CogEM and its refinements, Damasio's theory provides no mechanistic account, and could therefore provide no data simulations or predictions based on the model's emergent properties. Nor could he situate his heuristic concepts within a larger theory of how brain resonances and conciousness may be linked.
• Grossberg 2021 p204c2h0.30 In their 1990 article, Crick and Koch described two forms of consciousness : "a very fast form, linked to iconic memory...; and a slower one [wherein] an attentional mechanism transiently binds together all those neurons whose activity relates to the relevant features of a single visual object]". This conclusion was consistent with available results about ART at that time, but did not offer a linking hypothesis between brain dynamics and the perceptual, cognitive, and cognitive-emotional representations whose resonances support different conscious qualia. A great deal of additional experimental evidence for neural correlates of consciousness has been reported since 1990, but has typically led to theoretical conclusions that fail to make the crucial linking hypothesis between specific conscious pyschological qualia (eg Baars 2005; Dehaene 2014; Dennett 1991; Edelman, Tononi 2000; Koch etal 2016; Tononi 2004, 2015). ... Such claims are consistent with the analysis in this book, but they do not describe the underlying organizational principles, neural mechanisms, or brain representations that embody subjective conscious aspects of experience.