PSYCH 4320 / COGST 4310 / BioNB 4330

Consciousness and Free Will

Theme II

  Week 5: being a vertebrate II

Week 5: what is it like to be a bat vertebrate?

the neurobiological level: the big picture

The three principal domains of "world" (target selection), "body" (action selection), and "motivation" (needs) that must interact to optimize decision processes in real time, as implemented in the roof of the midbrain. The dorsolateral to ventromedial path from the surface of the colliculus to the midbrain aqueduct corresponds to a posterior to frontal to medial path in the cortex. In the reverse direction, and in functional terms, it reads "motivation", "action" and "world". S, I and D: superficial, intermediate and deep layers of the superior colliculus. PAG: the periaqueductal gray matter surrounding the midbrain cerebral aqueduct. Bidirectional arrow aligned with the collicular lamina denotes compensatory coordinate transformations.

the neurobiological level: the big picture

Sagittal diagram of cortical convergence (in part via the basal ganglia) onto key structures in the "synencephalic bottleneck" (marked by thick arrows in the main figure and by a black bar in the inset).
Abbreviations: C, nucleus cuneiformis; H, hypothalamus (preoptic area included); M, mammillary bodies; MP, "mesopontine state control nuclei" (locus coeruleus, pedunculopontine and laterodorsal tegmental nuclei, and dorsal raphé); MR, midbrain reticular formation; N, substanta nigra; P, periaqueductal gray matter; Pt, pretectum; R, red nucleus; SC, superior colliculus; V, ventral tegmental area; Z, zona incerta.

from the functional to the phenomenological level: the big picture

Highly schematic depiction of the nested relation between ego-center, neural body and neural world constituting the analog neural simulation ("reality space").

Black depicts the physical universe, one part of which is the physical body (black oval), both of which are necessarily outside of consciousness. One part of the physical body is the physical brain (circle; shaded and unshaded). The heavy black line separating the reality space from other functional domains within the brain indicates the exclusion of those domains from consciousness (unshaded). Arrows mark interfaces across which neural information may pass without entering consciousness.

the phenomenological level

Left: where the self appears to reside.

Right: what the world looks like from [my] left eye.

the phenomenological level

Human vision is phenomenally cyclopean because once the data streams from the two eyes get fused into a single representation they are no longer individually accessible.

To find out what the world looks like from my left eye, I must close my right eye; when both eyes are open, I cannot help seeing a single, integrated panorama (if I had impaired stereopsis due to amblyopia or some other condition, or if I were a rabbit, or a Pierson's puppeteer, things would look differently for me).

The visible world in the cyclopean panorama appears as if it is seen from a vantage point situated inside the skull, behind the bridge of the nose.

Of course, if the "I" (the phenomenal Self) were really where it seems to be, I would see nothing but bits of brain and bone. Instead, it looks like the entire front of my head is missing.

the phenomenological level

My phenomenal world, which includes an image of my body, must therefore be a neural fiction perpetrated by the senses. It is presumably there for my own good, which is probably one reason why this illusion cannot be dispelled at will.

Part of the illusion would persist even if I were to shut the world out altogether, via total sensory deprivation. The part that persists is literally central to the illusion: it is the part at which all the sensory inputs seem to converge, and which is present at all times, because it is fed, in addition to the external senses, by a continuous internally generated somatic input. This is the phenomenal Self.

Merker (2012): back to the neurobiological level

Issues such as how and where our phenomenal experience of a fully articulated three-dimensional world populated by interpreted perceptual objects might be implemented, and whether any conscious contents are in fact implemented at the cortical level itself, were left open.

A brain charged with guiding its body through a complex and lively world from a position of solitary confinement inside its opaque skull faces a set of functional challenges beset with inverse and ill-posed problems at every turn. Uncertainty and ambiguity therefore encumber all cortical labors, making probability distributions the natural medium of its disambiguating inferential operations.

Merker (2012): inverse problems abound

Inverse problems abound in all stimulus dimensions: Is this a highly reflective surface under low illumination or a poorly reflective one brightly lit? Is this acute angle actually an obtuse one viewed sideways?

A full list of inherent sensory ambiguities has never been compiled, but in the modality of vision alone it includes (but is not limited to) the relationship between our conscious percepts and stimulus dimensions such as size, distance, orientation, contour, texture, shading, illumination, reflectance, transmittance, motion, binocular disparity, solid shape, object groupings, and scene segmentation.

Merker (2012): probabilistic honesty: upside and downside

Major efforts in the study of perception and computer vision have been devoted to unravelling the means by which our perceptual systems nevertheless accomplish their tasks in the face of inherently noisy and ambiguous input. Typically they show the utility and even necessity of applying prior constraints, regularization, and inference operations to the resolution of sensory ambiguity. And behind these devices there appears to lie one further, fundamental, key to the successful operation of the cortical sensory hierarchies: their adoption of a policy of HONESTY. Ambiguity and uncertainty is the reality they face, and they appear to represent it as such, by means of probability density distributions over alternatives mapped onto populations of neurons [...].


Merker (2012): what the cortex does

[...] to maintain ready combinability of the contents of its sensory operations, the cortex may exercise a “wisdom of suspended commitments” implemented in probabilistic terms throughout its sensory hierarchies. It would be, as it were, the neural version of the programmer’s or modeler’s “principle of least commitment” (Marr 1976).

Merker (2012): when and where to commit?

Under ordinary, non-laboratory-contrived, circumstances we are surrounded by definite objects rather than ambiguously or probabilistically defined ones. Objects are distinct and well defined on all the seven basic dimensions of vision (Adelson & Bergen 1991), and together they surround us in the form of the well articulated and interpreted three-dimensional extended panorama we call the world. This commitment of sensory consciousness to exclude ambiguity and tentativeness from our conscious percepts appears to belong among its more basic characteristics.

The categorical refusal of sensory consciousness to admit ambiguity suggests the operation of an iron-clad constraint of some kind, or even a capacity limit. The question is what might account for it.

Merker (2012): the problem of "meaning holism"

[...] It is advantageous for the cortex to postpone the precipitation of estimates as long as possible, because doing so allows additional cues and constraints to be recruited to the disambiguating process (meanwhile decision pressure mounts by the fractions of seconds, as we shall see). Ideally, no candidate interpretation at any sensory level should be regarded as definitive until “all the facts are in” (i.e. the analysis is complete; see Edelman 2002, p. 128, ‘meaning holism’). Only then is it safe from being overturned by additional contextual constraints or cues, and ready to find its place in a comprehensive scene interpretation.

But how is any disambiguation at all possible under such circumstances, circumstances which would seem to make the resolution of any given ambiguity dependent on the resolution of all ambiguities?

Merker (2012): constraints on resolving the problem of holism

There are two principal requirements for mastering this logistical dilemma.

  1. Information in all parts of the cortex and across sensory modalities must be capable of interacting rather swiftly and directly [...]. Cortical connectivity helps fill this requirement through the bidirectional organization of cortical sensory hierarchies. It makes the frontolimbic-hippocampal domain of the cortical inter-areal connective graph a “superhub” of a small-world connectivity.
  2. Higher levels [must] take precedence over lower ones in the resolution of ambiguities. [...] This means that somehow a global scene interpretation is required for any disambiguation to be final at any level, i.e. definitive disambiguation can take place only in the form of a GLOBAL BEST ESTIMATE of the current scene as a whole.

We all know what the required global best estimate of the current sensory scene as a whole is like because – this is my claim – it is nothing other than the actual contents of our sensory consciousness, i.e. the world in which we find ourselves every second of our waking hours.

Merker (2012): how many neurons?

The question, then, is how compactly the momentary contents of sensory consciousnes might be represented in a neural medium. [...] How many neurons are required to represent the momentary state of our perception of the world in as far as its purely visual sensory aspects are concerned?

Somewhere between 105 and 106 neurons might suffice to render a realistic image of our visual surroundings. Sensory experience as a whole is multimodal, so the visual estimate must be expanded to accomodate the remaining senses. They are, however, unlikely to add comparable numbers to the total estimate, because they share the spatial frameworkof visual perception, being “contained within it” as it were. This leaves A FEW MILLION NEURONS as a rough and ready estimate for the total representational requirements of full moment-to-moment sensory awareness, once its preliminaries have been accomplished elsewhere.

Merker (2012): how often?

As soon as the gaze moves, the sensory “facts” delivered by sensory arrays change, irrevocably, because the sensory arrays are now differently disposed relative to their sources in the physical world. [...] That leaves only the space between gaze movements – typically a few hundred milliseconds – for an estimate to be extracted from the state of cortical sensory systems, only to become obsolete upon the occurrence of the gaze movement it supports.

The “being in time” of conscious contents, in other words, is rich but ephemeral, barely surviving the present moment, and being consigned to oblivion with each intrusive sensory change or movement of the gaze. The impression of robust continuity of a rich sensory world results from the coherent memory thread trailing focal attention, supplemented and supported by the fact that outside the focus – with its attraction to loci of change – things tend to remain pretty much the same.

After these preliminaries regarding design features it is time to consider the manner in which this set of desiderata might be realized in neural terms.

Merker (2012): where?

The “other thalamus” is demarcated by a heavy black line, within which dotted outlines mark – in rostrocaudal sequence – the mediodorsal, laterodorsal, lateral posterior and pulvinar nuclei, respectively. The territory surrounding the mediodorsal nucleus is the extended intralaminar complex. A sample of cortical areas are marked with symbols which also mark their highly schematic representation in the pulvinar and superior colliculus. The latter structure has been artificially inserted into the diagram in roughly the position and orientation (“upside down”) in which it would be encountered in horizontal sections of the human brain. The cone-like area at the center of the thalamus is the zona incerta, actually located beneath the dorsal thalamus. The thin oblique line through one of the “map stacks” of the pulvinar marks an axis of iso-representation through the aligned maps, each of which is a two-dimensional discoid cut by the plane of the section. Excitatory connections end in a “Y”, inhibitory connections in a “T”.

Merker (2012): the pulvinar

Thus, neurons occupying a pulvinar iso-representation line will individually reflect the functional specialty of their parent cortical map, but taken together will represent the full cortical range of visual functional specialization. This full range will be represented for each retinal location, in the aggregate covering all of retinal space.

Such an arrangement has the potential of forming a single, unitary visual image space whose moment-to-moment contents incorporate the full scope of multi-areal cortical visual activity that feeds it, and to do so in the form of a coherent and comprehensive spatial rendition of the current visual scene.

The dorsal pulvinar is interconnected with parietal and frontal oculomotor regions, but has no caudally directed premotor output of its own. It avails little that the extended intralaminar complex with its prominent oculomotor-related output to the basal ganglia (and hence to the superior colliculus) lies “right next door” if the pulvinar does not project to it.

Merker (2012): the pulvinar AND the superior colliculus

[...] Cortical activity related to gaze, orienting and focal attention would, in parallel with the proposed pulvinar constraint satisfaction operation, undergo radical reduction to final estimate form for those purposes in the superior colliculus.

The fact that both pulvinar and colliculus receive their cortical afference from the axons of layer 5 pyramidal cells, and often by collaterals of the very same axons, would ensure tight temporal coordination of their activities. The relevant pulvinar territory is, moreover, continually informed of collicular outcomes through the direct projection of the intermediate layers of the colliculus to the dorsal pulvinar. This gives the dorsal pulvinar running access to all dimensions of information bearing on a global best estimate of the momentary sensory situation.

The suggestion is, in other words, that the pulvinar does not need a caudally directed output of its own, since the colliculus operates in tandem with it at all times, and on the basis of the very same high-level sources of gaze-relevant cortical information.

the neurobiological level: the big picture

Merker (2012):

some questions to ponder

"The first-person exercise we have just conducted yields a minimal definition of the self as the perceptual egocenter of sensory consciousness and, BY EXTENSION, OF ALL AWARENESS."

What about bees? Nematodes?

Is completely self-less awareness possible? And if yes, what neural — or, better, COMPUTATIONAL — processes are necessary and sufficient for it?

Some tentative answers coming after the fall break.