Connectionists: Brain-like computing fanfare and big data fanfare

[james bower]

Mark,

Interesting points - a couple of comments:

On Jan 28, 2014, at 10:00 PM, Mark H. Bickhard <mhb0 at lehigh.edu> wrote:

>  but one way to think of the difference I have in mind is to consider theories as resources and frameworks for the construction of models.

this is precisely the direction that, in neuroscience, I think is a large part of the problem (and you seem to agree).  People come up with theories about how a particular part of the brain works (often based on what is almost antidotal descriptions of its architecture) and then build models as a kind of existence proof for the idea.  These theories and models then take on a life of their own.  One example in the cerebellum, is the learning theory invented by David Marr and Jim Albus (independently) in the late 60s.  While Marr later disparaged his own model and precisely its antidotal nature in the introduction to his epistemological book “Vision”, the Marr/Ablus theory and the pseudo-realistic, i.e., Ptolemaic  (see previous definition) models that were built based on the theory have generated literally thousands of experimental papers - whose methods were, in effect, manipulated to support the presumed function.  I would be happy to send anyone interested references.  To this day the majority of cerebellar cortical models that have been published implement one or another modern version of the original theory.  The models all violate the “conservation laws’ of Purkinje cells that I mentioned previously. Furthermore, the Marr / Albus theory can not be right, if those “conservation laws” are correct.

As I keep returning to in the case of Newton and the inverse square law.  He did not have a theory of gravitation, that he decide to build a model to demonstrate how it works.  he posed a physical question to a model of a physical system and something unexpected popped out.  What popped out became the inspiration for Newtons theory of mechanics. 

So, the point again is, that while today physicists can invent theories, and implement them as models - they are not working in the theoretical or modeling vacuum that exists for neuroscience.  

The field is absolutely stuck in a framework based on false presuppositions - many of them.  But the solution is not the a priori generation of a new framework - it is to go back to the system itself, using tools that allow the system itself to generate the framework.  Once we have that - speculation and theory can flow much more freely.  

We don’t really even know what kind of machine this is - and we don’t know very much about what it does either I don’t believe.

Jim


> 
> For example, does the brain function in terms of the processing of semantic information, or via endogenously oscillatory processes (at multiple spatial and temporal scales, within complex topologies) that engage in modulations amongst themselves?  The former assumption is almost universal today, but there are serious problems with it.  I advocate the latter (paper available, for those who might be interested: http://www.lehigh.edu/~mhb0/ModelCNSFunctioning19Jan14.pdf).
> 
> This raises some related issues.  E.g., is there a possibility that the field is stuck in a framework with a false presupposition?  If so, we might be in a position similar to that of attempting to determine (model?) the mass of fire (phlogiston).  That example is from a while ago (and the assumption that fire is a substance goes back millennia, at least in the Western tradition), but more recently we find people devoting careers to figuring the nature and properties of "associations" or of how two-layer Perceptrons could engage in full human level pattern recognition (and, later, overlooking for a decade or so that the in-principle proof of impossibility did not apply to multiple layer connectionist models).  False presuppositions are rather common in history.  I argue that the framework of semantic information processing is such a false presupposition.
> 
> Even without the problems of false presuppositions, there are also the problems of missing conceptual frameworks.  You could not "reverse engineer" a table without notions of atoms, molecules, van der Waals forces, and so on — it is not a problem solvable via reverse engineering alone without regard for the conceptual resources available.  Presumably, a similar point holds for understanding how the brain works, only more so, and it seems likely that we are in such a position of conceptual impoverishment, if not (also) in a position of working within a false framework of presuppositions.
> 
> I have argued that we are in fact working with false presuppositions and with an impoverishment of conceptual resources, and have made some attempts to contribute to resolving those problems.  The history of science provides strong "inductive" support that all science has been and still is caught in such problems and that major advances require changes at those levels.  Including science(s) of the brain.  But my arguments have tended to be more "in-principle" (and, thus, have driven me into theory, and even [horrors!] philosophy).
> 
> This speaks to the points made in the discussion about theory as a differentiated subdomain within physics, and that it perhaps should also be recognized as such in studying the brain (and in psychology and cognitive science, etc. more generally).  I fully agree.  Recall, however, that the first Nobel for purely theoretical work in physics was for Dirac in 1933 (they would not and did not award the Nobel to Einstein for his purely theoretical work).  So, physics is a positive model in this regard, but it too suffered from a serious hesitancy to make and honor the distinction.
> 
> Mark
> 
> Mark H. Bickhard
> Lehigh University
> 17 Memorial Drive East
> Bethlehem, PA 18015
> mark at bickhard.name
> http://bickhard.ws/
> 
> On Jan 28, 2014, at 7:52 PM, Ping Li wrote:
> 
> 
> 
> Hi John, 
> In psychology, it's often the opposite -- when the theory (or model, in this case) and experiment don't agree, it's the theory that's to blame. Hence we have to "simulate the data", "replicate the empirical findings", and "match with empirical evidence" (phrases used in almost all cognitive modeling papers -- just finished another one myself)... That's why Brad pointed out it's so hard to publish modeling papers without corresponding experiments or other empirical data.
> 
> Best,
> Ping
> 
> 
>  That's not the whole story.  For modern physics, a common happening is that when theory and experiment disagree, it is the experiment that is wrong, at least if the theory is well established.  (Faster-than-light neutrinos are only one example.)
> 
> John Collins
> 
> 
> 
> 
> 
> 

 

 

Dr. James M. Bower Ph.D.

Professor of Computational Neurobiology

Barshop Institute for Longevity and Aging Studies.

15355 Lambda Drive

University of Texas Health Science Center 

San Antonio, Texas  78245

 

Phone:  210 382 0553

Email: bower at uthscsa.edu

Web: http://www.bower-lab.org

twitter: superid101

linkedin: Jim Bower

 

CONFIDENTIAL NOTICE:

The contents of this email and any attachments to it may be privileged or contain privileged and confidential information. This information is only for the viewing or use of the intended recipient. If you have received this e-mail in error or are not the intended recipient, you are hereby notified that any disclosure, copying, distribution or use of, or the taking of any action in reliance upon, any of the information contained in this e-mail, or

any of the attachments to this e-mail, is strictly prohibited and that this e-mail and all of the attachments to this e-mail, if any, must be

immediately returned to the sender or destroyed and, in either case, this e-mail and all attachments to this e-mail must be immediately deleted from your computer without making any copies hereof and any and all hard copies made must be destroyed. If you have received this e-mail in error, please notify the sender by e-mail immediately.

 


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20140129/cfede036/attachment.html>