Transcript Optimism vs. skepticism in cognitive neuroscience

Optimism vs. skepticism
in cognitive neuroscience
(Bechtel + Gallant’s lab
vs.
Uttal + Hardcastle)
Optimism
I. Bechtel (‘11)
• “Heuristic identity theory” (fMRI) over 30 areas
in for visual processing (occipital lobe, parietal
and temporal cortex) (‘08)
• fMRI – notions “localization” “brain areas” reconceptualized
• Bechtel’s alternative: mechanism + dynamical
system = dynamical mechanisms
• In 2008 – B: reductionism + emergence
• Autonomy of system (from Bernard’s notion of
“internal environment”, Cannon’s “homeostasis”,
Varela’s “autopoiesis”)
• “In fact, living systems has typically highly
integrated despite the differentiation of
operations between different organs and cell
types. The mind/brain seems to be no
different on this score – it consists of
component processing areas that perform
different computations which has
nonetheless highly integrated with each
other. Such a mechanism does not typically
include encapsulated modules, and one is not
likely to find them in the mind/brain.” (Bechtel
2009)
• Mental mechanisms with specific functions
could be localized, but he emphasizes the
integration
• Sporns and Zwi’s (2004) “dual role of cortical
connectivity”:
(1) The functional specificity of certain cortical
areas that manipulates specific information
(2) The integration of this kind of information in
a coherent behavior and cognitive states
(“integration into coherent global states
through oscillations in thalamic neurons play
in producing global states”)
II. Gallant’s lab
(1) Gallant 2008 + 2009
• 120 pictures in V1, V2, V3
• Training set: fMRI signals to a larger (1,750) library of
natural images and measured the fMRI responses
produced by each of a number of voxels
• A representation of each picture - formed - spatial
frequency and orientation information summarized as
a “predicted activity pattern” for each of voxels
associated with picture = Quantitative representation
of fMRI responses to each image. (Uttal, p. 112)
Based on similarities, the voxel pattern of each image
(from those 120) was compared with information
from library, the best fitting image being selected.
(2) “Reverse retinopy” Thiron et al (2006) and
Miyawaki et al. (2008) (in Uttal, p. 114)
• Shape of simple geometrical forms is preserved
enough in early retinotopic regions of visual
system
= Preservation of topology of original stimulus
pattern and spatial resolvability of spatial pattern
• Examine spatial pattern of activated voxels, and
infer shape of stimulus. (…) Spatial information is
retinotopically persevered in early portion of the
visual system. (Uttal, p. 113)
(3) Gallant et al. second work – not “reverse retinopy”
method but “complex natural scenes from
nonisomorphic fMRI images”
• Uttal: “a process of recognition, selection, or
identification of an image from a known library of
alternatives rather than a reconstruction in either
the psychological or the neuroscientific sense.”
(Uttal 2011, p. 114)
• Uttal: Reconstructed images - “not pictures directly reconstructed from fMRI data but pictures
produced by combining parts of pictures that were
selected from the library of images, that is, the
Bayesian priors on which the system was originally
trained.” (p. 114)
• They selected pictures from their library based on
the pattern of activations
• This is not reconstruction per se; it is once again
selecting from a predetermined “deck of cards”.
Gallant 2011
• fMRI, computer progamm = Quantitative modeling
of brain activity measurements
• Reconstructed natural movies
• In the past, main problem: blood oxygen leveldependent (BOLD) signals measured by fMRI are
much slower than neuronal activity in relationship
with dynamic stimuli
• “Motion-energy encoding” model has to fit two
components: visual motion information and slow
hemodynamics mechanisms
• V1, V2, V3 (occipitaltemporal visual cortex areas)
= How spatial and temporal information are
represented in thousands of voxels of visual cortex
• Watched several hours of movies – brain scanned
fMRI --- measurements in computer
• Comparing fMRI data and details of each movie,
computer program constructs “dictionaries” for
shape, edge and motion. Each voxel has such a
dictionary.
• Second set of movies- new fMRI data --reconstruction
“The goal of movie reconstruction is to use the
evoked activity to recreate the movie you observed.
To do this, we create encoding models that describe
how movies are transformed into brain activity, and
then we use those models to decode brain activity
and reconstruct the stimulus.” (Gallantlab.org)
[Meaning + binding problem (oscillations)?]
Skepticism
I. Hardcastle (2007)
• Cognition: More and more interests in
development neurobiology (more and more
molecular) or gene-environment interactions
• Perception: from where its stability?
“… perceptual systems represent the external
environment to us. Through complex
computations we do not yet understand, our
sensory systems derive stable images from the
ever-fluctuating raw signals of our transducers.” (p.
296)
How and where are our sensory signals encoded and
stored? How do we separate “figure” from
“ground”? How are incoming signals “mixed” with
our memories, attention, and our understanding of
the world so that we get full-blown representational experiences? How do we combine information from different sensory modalities? How do
other brain systems transform and use this information? How do they modulate the represen- tations
to meet our behavioral goals and biological needs?
How do we use representations to regulate action,
planning, and other outputs? (p.297)
• The neuroscientists can record no more than
simultaneously 150 neurons, they can summed LFP
from no more than several thousands.
“But brain areas have hundreds of thousands of
neurons, several orders of magnitude more than
they can access at any given time. And these
neurons are of different types, with different
response properties and different interconnections
with other cells, including other similar neurons,
neurons with significantly different response
properties, and cells of other types completely.” (p.
304)
Problems with
• Recording activity of one cell inserting an electrode
(feedback projections in the brain)
• Lesions and fMRI investigation (fMRI has quite well
spatio-temporal limits)
“Neuroscience is a victim of imprecise
instrumentation. If scientists extrapolate from what
they might learn with more sensitive measures, it
can easily be seen that there will come a time when
this whole approach just will not work anymore. Put
in the harshest terms, brain imaging seems to
support reductionism because the imaging
technology is not very good yet.” (p. 306)
II. Uttal’s skepticism (2011)
“In effect, we are doing what we can do when we
cannot do what we should do.” (2011)
• Role of fMRI in mind-brain problem – status of
cognitive neuroscience
• “explosive growth of [brain imaging] - No
comprehensive and synoptic evaluation of huge
number of studies” (p. 1)
• Against localization: any simple thought involves
whole brain
Ontological Postulate
(1) All mental processes are the outcome of neural
activity.
(2) All mental processes are the outcome of the
microscopic interactions and actions of the great
neuronal networks of the brain.
• fMRI grasp macro-level not micro-level
• Against localization: bidirectionality (+ complexity
of brain) and degeneracy (Edelman) of neuronal
areas
Epistemological postulates
(1) Brain activity associated with mental activity is
broadly distributed on and in the brain →
Phrenological localization - replaced with broadly
distributed neural systems for mental activity
(2) Great complexity and number of neural networks
→ Not possible to find correlations neural-mental
• Brain imaging techniques do best - where the brain
activity is observed when a stimulus is presented
• “neural network approach is computationally
intractable” → Mind-body problem cannot be
solved