Connectionists: Brain-like computing fanfare and big data fanfare

[Richard Loosemore]

I have already written at least one paper (with Trevor Harley) 
complaining about much the same cluster of issues as have been raised in 
this discussion.

I will mention just two points, extracted from those papers:

1) Re. Ptolemy's Epicycles, and the bankruptcy of that approach to 
psychology/neuroscience.

It is possible to see this entire issue as stemming from the 
"complex-system-ness" of the system we are interested in.  To the extent 
that cognition is a complex system (i.e. there is a disconnect between 
overall behavior and underlying mechanism), we would [insert here a long 
argument] expect it to be extraordinarily difficult to come up with high 
level theories of brain function that have a tight connection to the 
underlying neural machinery.  That is especially relevant to those who 
think they can understand the brain by just simulating it, or collecting 
vast amounts of signal data:  you're wasting your time, because all that 
effort will boot you nothing.

There IS a way around that issue, but it involves a realignment of how 
we do both cog psych and neuroscience (and AI for that matter).  We need 
more systematic exploration of very large numbers of different types of 
cognitive mechanism models.  Treat the discovery of theories not as the 
work of bright individuals (one theory per individual per lifetime) but 
as a process that is quasi-automated, and which yields a thousand 
theories a day.

2)  Re Brain Imaging.

In the Loosemore and Harley paper I pointed out the massive impact that 
a slightly off-beat theory can have.  If the functional units of the 
brain are actually "virtual" entities that are allowed to move around on 
a physical network of column-like units, some of the neuroscience data 
can be explained in a very elegant way (and BTW the same data is 
inconsistent with all other theories or frameworks).  
But....but....but:  that same off-beat theory, if it were correct, would 
render into nonsense almost all of the defauot assumptions being made by 
those collecting the Big Neuroscience Data right now.  It would do that 
because believe it or not all that BND is pretty much wedded to the idea 
that the functional units are not virtual;  that the physical units 
actually do 'mean' something most of the time.

My conclusion is with James, and with Juyang Weng, who started the 
discussion.  The Big Data/Brain-Simulation-Or-Bust approach is a 
gigantic boondoggle.

Richard Loosemore
Mathematical and Physical Science
Wells College
Aurora NY 13026
USA




On 1/25/14, 12:05 PM, james bower wrote:
> Hi Jose,
>
> Ah, neuroimaging - don't get me started.  Not all, but a great deal of 
> neuroimaging has become a modern form of phrenology IMHO, distorting 
> not only neuroscience, but it turns out, increasingly business too. 
>  To wit:
>
[.....]



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20140125/51161787/attachment.html>