Mansell, W. (2022, July 4). An Integrative Control Theory Perspective on Consciousness. Psychological Review. Advance
online publication. DOI:10.1037/rev0000384
Warren had his own fireworks display on our American ‘Independence Day’ holiday. I read this a couple of weeks ago, but have not been able to focus on it again until now as I recover from pneumonia, fortunately caught early and clearing but known to be a nasty thing worthy of respect.
The most important reference of the term ‘integrative’ in the title is the critical role that Warren attributes to processes of integrating novel information. By reorganization, a perceptual input function may be created which integrates perceptions of lower order into a perception of a more abstract order that did not previously exist.
He also provides a masterful survey of other proposals in consciousness studies, and indicates how PCT can integrate aspects of each. “Two of the most widely discussed theories of consciousness are global workspace theory (GWT; Newman & Baars, 1993) and integrated information theory (IIT; Tononi, 2008; Tononi et al., 2016).”
Three forms of consciousness are considered in these studies and here, Primary or phenomenal consciousness, Secondary consciousness in which perceptions which might be inputs providing ‘new information’ are accessed, and Tertiary consciousness involving self-awareness.
He says that qualia, the ‘suchness’ of perception, “emerge” as a function of integrating novel information. This exposes what to me is the real ‘hard problem’ of consciousness, the relation between neurochemical processes and subjective experience. Warren does not discuss this problem.
Primary (phenomenal) consciousness arises with conflict of intrinsic systems and consequent reorganization. With such conflict, attention is drawn to the inputs of those systems. PCT postulates that attention keeps reorganization confined to the systems that are engaged by the conflict.
As an aside in this review (in the article it bears on the integration of IIT with PCT), I want to mention some distinctions as to what is information. Shannon information theory is observer-dependent, a function of the recipient’s uncertainty. IIT proposed that ‘intrinsic information’ is observer-independent because it results from feedback, the system presenting the information to itself. Warren sees ‘control information’ as intermediate between these.
Control information is “the capacity (know how) to control the acquisition, disposition and utilization of matter/energy in purposive (teleonomic) processes” (Corning, 2007, p. 302). It is the information used within a control system while it is in the process of controlling. The use of control information is inherently a process involving feedback, but it is a specific use of that feedback within a control system or hierarchical network of control systems that defines control information.
If the emergence of consciousness were only with formation of perceptual input functions that create new perceptions (novel information), then consciousness would be episodic. Continuity is a property of Secondary consciousness. Warren proposes that continuity is a consequence of controlling a newly identified intrinsic variable which is related to the net rate of reorganization in the hierarchy.
Gary Cziko (1997) and Warren (2015) have proposed an analogy of reorganization to genetic algorithms, which perform best at an optimal mutation rate. On this basis, he proposes that the rate of creating new input functions and new perceptions (“the rate of novel information integration”) is itself an intrinsic variable. He proposes
that living organisms who develop perceptual hierarchies are intrinsically motivated to find new ways to integrate information at a self-governed rate, and they do so “creatively,” testing out different transformations (e.g., different delays, derivatives) to their input signals to continually optimize the specification and control of a perceptual variable. The information integration rate could be controlled either by accessing novel perceptual inputs through actions such as eye movement, object manipulation, locomotion, or social interaction, or otherwise by altering the intrinsic, trial-and-error mutation rate of reorganization that is applied to these inputs. Humans can also use a third method to add variants in input—their imagination….
These means of ‘paying attention’ are limited resources, narrow channels relative to the abundance of potential inputs. By recurring to the same constrained set of perceptual inputs their episodic bases for consciousness become continuous. Perhaps the familiar principle of successive still images perceived as a movie is an apt analogy. Memory and imagination can fill gaps in an incipient, emergent novel perception even as they do in a mature input function.
Warren proposes that self-awareness and Tertiary consciousness emerge from cooperation and competition between intrinsic systems in the somatic branch of the hierarchy and ‘propositional’ systems above where the somatic branch and the behavioral branch merge.
For ‘propositional systems’ he refers to the principle of ‘order reduction’ mentioned on p. 318 in the second part of Powers et al. (1960) as the basis for symbols. A symbol, it is said, is a ‘content free’ perception at a relatively low level (such as a configuration) which is linked in memory to a much more abstract and complex perception so that perception of the former evokes the latter from memory. The distinctions between icons, signs, and symbols call for more elaborate discussion elsewhere. The brief 1960 passage is no more than a promissory note that Kent McClelland began to pay up in the 1990s, the promise that by modeling “sophisticated systems of symbols that “represent” specific perceptual variables” PCT may account for the huge domains of the perceptual universe of a human that no animal can know. ‘Order reduction’
… allows elements to be combined and reorganized sequentially within the language or as if-then plans in more elaborate ways than the perceptions themselves and communicated to others at high fidelity to facilitate shared pursuits and verbal learning. Among the propositional systems in use are semantic knowledge, autobiographical knowledge, sciences, languages, cultures, and religions. These systems include the “beliefs” described by some predictive processing accounts. The relevance here is that sustaining a perceptual input in consciousness may be necessary to link it with a specific symbol. Once learned, this allows the individual to bring and sustain a perceptual signal to consciousness via its symbol (e.g., by writing down one’s thoughts), or indeed to remove it from consciousness by switching to another line of thinking. This enables humans to learn new roles, abilities, skills, and tasks in systematic ways, often by focusing conscious awareness on each separate perceptual variable involved in the skill to allow improved control, and then shifting to another.
“Powers incorporated local memory into each unit of the hierarchy” and indeed neuroscience affirms that memory is stored at each synapse. What this means is fogged by the mystery of how we get subjective experience when neurons fire. Materialist commitments require us to ask what the heck is subjective experience anyway, and claims of Dennott and others that consciousness (the term used as a surrogate for subjective experience) is an emergent ‘epiphenomenon’ is merely the current incarnation of the ‘war of science and religion’ that so plagued the past two centuries, from which Warren wisely steers clear. The question, what is subjective experience, is paradoxical because it is only by subjective experience that we know anything. It is the water in which we fish swim. PCT would give us a perception of perception. There are self-deceits in abstraction.
[T]here are also potential disadvantages of propositional coding if it is not used in synergy with other control systems. First, logic and language can give the illusion of an agreed perception of reality between two or more individuals by “smoothing over” the divide between them; one person’s perception of a “house,” or “justice,” may be different from another person’s, especially if they grow up in different cultures. Second, and relatedly, through attempting to control only symbols, the practical limitations of putting a plan into practice can be missed or poorly conceived—this is the classic divide between theory and implementation. Third, and importantly for the conscious individual, a plan, knowledge, or “facts” that are mutually agreed within a propositional format may conflict with the intrinsic goals of the individual, for example, within societal policies that are discriminative, or at an extreme, within harmful religious practices or suicide pacts.
Abstraction is not the same as generalization. Two individuals may generalize the same environmental influences, as experienced at lower levels of the hierarchy, in different ways, but label them socially as the same abstraction.
I dropped an important thread earlier, episodic lower-level perceptions give rise to continuity. Accepting the muddled use of ‘qualia’ for these perceptions, the ignored mystery of subjective perception and ‘objective’ perceptual signal, Warren says
if the rate of novel information integration were to be an endogenously controlled variable in itself, this might sustain qualia over time. Thus, based on PCT, I propose that living organisms who develop perceptual hierarchies are intrinsically motivated to find new ways to integrate information at a self-governed rate, and they do so “creatively,” testing out different transformations (e.g., different delays, derivatives) to their input signals to continually optimize the specification and control of a perceptual variable. The information integration rate could be controlled either by accessing novel perceptual inputs through actions such as eye movement, object manipulation, locomotion, or social interaction, or otherwise by altering the intrinsic, trial-and-error mutation rate of reorganization that is applied to these inputs. Humans can also use a third method to add variants in input—their imagination, as described later.
Imagination is one of four modes in which an elementary control system within the hierarchy may function. Every mode engages memory, a perceptual signal that is generated locally at synapses.
This reproduced signal carries the same information, or some significant portion of it, that the original feedback signal carried. To all intents, it is a sensory signal, but one arising from a past event rather than a present one (p. 17 of Powers et al. 1960 Part 1).
In imagination, a system must substitute a copy of its reference signal in place of the signal created by its perceptual input function. The 1960 and 1973 block diagrams imply that this is done by neurons breaking or suppressing the synaptic connection of the perceptual input function to the comparator and the synapses branching from the comparator to the reference input functions of lower-order systems while simultaneously establishing another a synaptic connection from the reference input function to the perceptual input function. For that imagined perception to have any detail, a higher-order system must request that the relevant lower-level systems synapsed to its perceptual input function go into imagination mode. For more detail, those systems in turn must obtain imagined input in the same way. (I’m ignoring here that sensory input and imagination may be combined.)
Similarly for automatic mode and observation mode ‘switches’ must disconnect some of the synapses of a normally operating elementary control loop even as they connect alternative synapses. All this targeted and precisely timed making and breaking of synapses (or neurochemical activation and suppression) has to my knowledge not been supported by neurological evidence, but absence of evidence from that field is not unusual.
Remembering the great capacity for ‘redundancy’ in the brain, it possible that systems operate on the same perceptual inputs in parallel but in different modes.
For imagination, I have proposed that the reference input always branches to perceptual input. In the absence of inputs from lower levels, elementary loops are always ‘idling’. In the absence of changes in input from above, a reference input function generates a weak signal which branches to perceptual input and is controlled with no error. It may be that neurons too long inactive start to reorganize; nerve cells in vitro spontaneously branch, make synaptic connections, and start to influence one another’s rates of firing.
The automatic mode does not require that reference input be cut off, it only requires that the perceptual signal not cause error at the higher level. Routine walking is an example given of automatic mode, but a stumble switches us to control mode.
The passive observation mode requires perceptual input not to cause error. The block diagrams assert that this is done by cutting error output off from lower-level reference inputs. But we already have claimed that perceptual input is stored on the output side of the comparator. I believe the routine storing of perceptual input as memory is all that is required. This is done in part by copying perceptual input to the reference input function.
The block diagrams place this memory at the reference input function. I would say rather that it is at two locations. The more important is the output function of the higher-level system requesting that perception, where the error signal branches with different weights to the several lower-level systems providing signals to its perceptual input function, how much of this signal, how much of that. The other locations are at the reference input functions of those lower-level systems.
In both places the aspect of memory which makes a given memory distinctive is structural: where does it stand in the hierarchy? To which lower-level system does the error output function connect? To which higher-level systems does the reference input function connect? The architecture of these connections as well as the degree of amplification of each signal together constitute a memory. (An axon or dendrite amplifies its signal by branching and synapsing the branch(es) to the same destination.)
The higher-order system that requests observation does so by sending reference signals that cause no error in the lower-level systems that provide its perceptual input. How it might do that would go even farther out of scope for this review, but that is all that is needed for observation mode.
To the ‘creative activities’ by which we regulate the intrinsic ‘information integration rate’ we should add ‘mulling it over’, ‘daydream’, ‘taking a break’, switching tasks so that when one is ‘stuck’ another unrelated one can progress, and sleep.
The correlation of this intrinsic control to fidgeting and mental ‘distraction’ in meditation is quite interesting. The advice for meditation is to perceive a distracting perception in observation mode and waiting for it to subside instead of galloping off on imaginative elaborations of it. When it does subside, we continue to observe the perception commanded by the ‘meditate’ control system, which might be a blank wall as in Zen or in Warren’s experiment until we are observing another ‘distraction’ perception. It is the strengthening of skill in observation mode that is important. Only by observation are memories made accurate rather than distorted by prejudice.
Discussion of the contributions of the paleomammalian brain and the PCT model of emotion are not included in Warren’s survey here, but that is a major source of such prejudicial influence on memory and therefore of subsequent perceptions. (I refer here to the functions of memory discussed above.) The paleomammalian brain and the ‘reptilian’ functions closer to the brainstem demand slower rates of ‘information integration’ (less uncertainty) than cortical systems do. Conflict here is an important direction for research.
Temperamental differences arise from differences in resolution of this conflict. The larger amygdala found in conservatives may be a genetic endowment and it may be due to childhood development in an authoritative and punitive environment where the need for certainty is elevated as means of avoiding punishment. Or both. The open/closed mind parameter investigated by Rokeach may correlate with skill in observation mode. The rigid/flexible parameter may correlate with the balance of paleomammalian vs. cortical control of the intrinsic information-integration variable. A high demand for certainty correlates with reluctance to go back to a prior step taken in problem-solving, taking it as an established certainty rather than considering alternatives. Related to this is ‘premature closure’. Creativity requires “willing suspension of belief” (and disbelief) as much or even more than does audience participation.
I am sorry that I was not able to take up the opportunity that was given me to read this during the process of its writing. There may be suggestions here for the future, but they are surely too late for the transition from online publication to print. The references to person-specific computational models should include work by Tom Bourbon, if that is possible. There are a few bumps in the road that may already have been attended to. On p. 7:
The next step in bridging IIT and PCT is to consider the integration of input signals to form an input function[, a process] that occurs at each successive level of the perceptual hierarchy.
(This to avoid the ambiguity of “to form an input system that occurs at each level” vs. the intended reading as “the integration of input signals … that occurs at each level”.)
on p. 13:
Examples of disturbances that lower the potential resource for novel information integration include environments -with- [that are] highly familiar-,- or very simple-, environments-.
The picture that Warren presents of PCT is a very appealing one.
This is conscious human life—the adaptive switching between plans and multiple instincts, using conversation, language, and the realm of cultural artifacts to establish and maintain control within the bounds of biological limits.
Warren has taken up a very challenging topic, executed a masterful PCT account of it in relation to the leading alternative proposals, and by this publication is leading people who are skilled in observation mode to look into PCT more closely, with willing suspension of prior convictions.
For citations here see the reference list in Warren’s article.
