poniedziałek, 30 lipca 2012

Reading Thomas Metzinger -- transparency


Reading "Being No One", the biggest problem I've found so far is the
"Transparency" constraint (3.2.7, of 3.2 "Multilevel Constraints: What
Makes a Neural Representation a Phenomenal Representation?", of 3 "The
Representational Deep Structure of Phenomenal Experience"), and the
related "Homogeneity" constraint (3.2.10 "“Ultrasmoothness”: The
Homogeneity of Simple Content"). Perhaps I'll grasp it making this

Warning: the notion of attention I use below is the technical one
defined in the book.

Although the author appears to define "transparency" precisely, I'm
not sure the bundle of accompanying examples fits into a single
concept. It is spanned between two aspects.

The first is phenomenal simplicity: experience is transparent, when we
cannot direct attention to any more of the details of the process over
which the experience supervenes.  I.e. there is a strict unpenetrable
border of attention, the aspects of the preception process which (at
least potentially) are part of the experience, are the "portions" of
this process to which we can direct attention. And there are large
portions of the perception process to which we just cannot direct
attention at all, at least in normal conditions (e.g. not under
psychedelic drugs, some drug-induced artifacts might be transgressions
of this attentional border). This view of transparency makes the
experience "substantial": things (and I'll add thoughts as well)
appear to us as they appear to us to fundamentally be, we have no
experiential clue that the appearance could be, for example, an
abstraction or a statistical inference.

The second aspect is phenomenal givenness (veridicity?). In normal
wakeful awareness (and in some nocturnal dreams, those that are
"realistic" i.e. vivid and non-lucid), we are predisposed to be naive
realists. We experience being immersed in a world as it independently
of our act of perception is. Experience is transparent when it is
experienced as exclusively about the actual world (or the actual
us). Experience is transparent when its content is experienced as its
only cause. Under this aspect, thoughts, imaginations and lucid dreams
are phenomenally opaque (and therefore also plans, mental rehearsal,
etc.) They (usually) are perceived as "representing", as standing
for something (since only actual stuff can be present).

Of course these two aspects are related, for example we can talk about
ineffability and immediacy in both cases. But there might be
experiences that are one but not the other. It's likely that in vivid
dreams (and lucid dreams are vivid since a lot of the brain is waken
up) we experience qualia. Is Thomas Metzinger taking the two aspects
as two sides of a single coin, or just defining "transparency" as
conjunction of "givenness" and "simplicity"?

The author's definition of phenomenal transparency:

"For any phenomenal state, the degree of phenomenal transparency is
inversely proportional to the introspective degree of attentional
availability of earlier processing stages."

Earlier processing stages are "temporally earlier" in the aspect of
phenomenal givenness, and are "constitutionally earlier" in the aspect
of phenomenal simplicity. In both cases the "earlier processing
stages" are the "internal causes" of the experience.

I think from what Thomas Metzinger writes that he might think, if we
could direct attention to, for example, edge detectors in visual
processing, as long as we rest our attention at the edge detectors, we
would have to "will the perception into existence" of objects that
constitute the normal visual experience, for them to appear to us. By
analogy to how we experience imagination, where we have to "will"
imagined objects "into existence". The objects wouldn't just
effortlessly appear besides the edges. In 3.2.10 paragraph
"Homogeneity as an Informational-Computational Strategy" (he writes
such paragraphs for each constraint) he states: "Without homogeneity
we could introspectively penetrate into the processing stages
underlying the activation of sensory content. One obvious consequence
of this would be that the multimodal, high-dimensional surface of our
phenomenal world would start to dissolve. We would then phenomenally
experience the model as an ongoing global simulation permanently
generated from scratch, as it were, and thereby it would inevitably
lose the phenomenal character of being an untranscendable reality."
What is normally perceived is always a persistent, "maximum a
posteriori" object. Fixing too early processing stages, like edge
detectors, might be disruptive to this inferential process.

What are your thoughts? Do you think that phenomenal givenness implies
phenomenal simplicity? What about vice-versa? Do you think that the
objects of imagination are always given to us relationally, we always
grasp the processuality of their coming about? Are there imaginary
qualia? Do you think that attentional access to (constitutionally)
earlier processing stages dissolves the experiential immediacy of
later stages?

Lucid dreams are interesting because we can affect the scene
construction by directing our attention to toplevel fragments of the
top-down information propagation in scene construction, while the
bottom-up information propagation proceeds unaffected. I'd say that we
have "givenness" when we do not clamp any fragment of the toplevel
layers for the top-down information propagation of a given scene
(either because we cannot since the bottom-level layers are clamped by
sensory input, or because we don't realize that we can). (I'm sorry
for the homunkulus-like way of speaking.) I'd say we have "simplicity"
when the bottom-up information propagation starts below the lowest
attentionally available layer, rather than in the middle, for the
scene. I'm finishing 3.2.10, lucid dreams are covered somewhere in the
next chapter.

Chapter 4 "Case Studies I".

Let me pick up more issues in "Being No One", perhaps the glitches are
my own stubbornness.

The first problem is when the author claims (4.2 "Deviant phenomental
models of reality", 4.2.1 "Agnosia" p. 220/221) that a patient who
uses chromatic information for shape formation and motion detection,
but has no experience of color (his visual experience is in shades of
gray), has color cognitively but not attentionally avialable. The fact
that chromatic distinctions feed into shape formation has obviously
nothing to do with recognizing color conceptually, no? Chromatic
vision feeds into formation of concepts here, but not of color

4.2.2 "Neglect". Similarily, hemineglect no doubt is an attentional
deficit (and a deficit of the "model of intentionality relation"), but
likely the deficit is simply because of the lesion of processing
stages leaving nothing to attend to and model. There's less of a
problem because the author doesn't say otherwise. He sort-of analyzes
the minimal conditions that could generate hemineglect.

Anton's syndrome discussion focuses on the self-modeling deficit
without mentioning whether the offline phenomenal experience
(nocturnal dream like, but top-down modulated) is also absent. I guess
it is.

4.2.4 "Hallucinations", Charles-Bonnet syndrome; "percepts are missing
characteristic features and are simply superimposed on, but not
semantically embedded in, the phenomenal model of external reality" is
contradicted by the following example. The patient reports pragmatic
and (slight) phenomenal abnormalities as distinguishing hallucinated
content, semantically it seems to be OK.

Phenomenal transparency/opacity is again distincitly used in two
senses, one is transparency as "phenomenally normal experience", and
the other, inflated meaning is opacity as "believed to not correspond
to reality".

I don't think it's likely that "earlier processing stages" can become
directly available for attention, because of architectural
limitations. It's more likely that they become available indirectly by
"polluting" the bottom-up signal with "vehicle properties" (the
"maximum aposteriori distribution" puts too much weight on the
consistency of the lower layers, so they fixate before propagating
information upwards). It is still a form of attentional
availability... Perhaps this could be not indirect, but an important
(if not the primary) mechanism of attentional availability? For
Metzinger, attentional availability is "subsymbolic
re-representation", i.e. additional neurodynamical structure is formed
that correlates with the original phenomenon and propagates
information about it.

You might be asking how this differs from cognitive availability,
i.e. "symbolic re-representation". The attentional structure is
(neurodynamically) homomorphic with the original phenomenon and so
cannot be reassembled in arbitrary contexts, while the symbolic
structure is only activationally linked with the original phenomenon.

4.2.5 "Dreams", "phenomenal dream content is not attentionally
available" -- obviously, "All there is is salience-driven, low-level
attention." Again we have two concepts conflated, "attentional
salience" and "volitional attention". Phenomena have to be
attentionally available (in the attentional salience aspect) to be
even minimally conscious. As later noted, this is related to the
distinction from chapter 2 (p. 36) between four forms of

1. external attention
2. consciously experienced cognitive reference
3. inward attention / inner perception
4. consciously experienced cognitive self-reference

Introspection 1 is attention (subsymbolic re-representation) toward a
"world" experience. Introspection 3 is attention toward the
self-model, it "is generated by processes of phenomenal
representation, which direct attention toward certain aspects of an
internal system state, the intentional content of which is being
constituted by a part of the world depicted as internal".
Introspection 2 and 4 are the symbolic variants. Metzinger says
"introspection 3 is almost impossible in a dream state, because
high-level attention is absent." Obviously, most dreams feature a
phenomenal self, only it is not a volitional self because of the lack
of deliberation. Metzinger seems to say that in dreams, attention
cannot rest on the self-model, it only integrates features of the
self-model selected by the generated world-model. But it's an
attentional deficit (lack of high-level attention). "You cannot
introspectively attend even to your most simple sensory perceptions in
the dream state, because you are not an attentional subject."

(Anecdotally, I think I've had dreams without a phenomenal self -- dreamt
from the third person perspective, or, remembered as movies; I've certainly had
dreams with a phenomenal self that was not my actual (waking)
self-model, remembered as movies with a distinguished
character. Normally I have dreams remembered in roughly the same
format as normal memories.)


Later the author makes it even more evident that he thinks of the presentational content, i.e. the processes that are only attentively but not cognitively available, as transparent, and of representational content, i.e. cognitively available processes, as opaque. As a tendency perhaps.