[Bmi] some clarification

Paul Werbos pwerbos at nsf.gov
Mon Oct 24 09:24:22 EDT 2011

Good mor ing!

I apologize if my message to John was not so coherent. It was not intended
as something for the group.

So maybe I owe you a bit of clarification.

There are many worthwhile goals out there related to the brain, all the way from
routine clinical practice to basic science.

In basic science, I would want to focus energy especially on two "grand challenges":

(1) The ability to understand and replicate the ability of the basic mammal brain ("the mouse")
to do what NFS Engineering has called "cognitive optimization and prediction."
Here I am referring to what you would see if you searched on "COPN" at www.nsf.gov.
This is a relatively crisp goal, which nonetheless requires some greater cooperation between 
a number of disciplines and approaches.

(2) "Human potential."

It is easier in many ways to cooperate towards the first, focused goal.  
My paper in Neural Networks 2009 basically gives my view of how we could get there,
and it also stresses the two "subchallenges" -- the cognitive optimization part and the cognitive prediction part.
I view the image processing work recently discussed by Leonid, by LeCun, by Schmidhuber
and even by Grossberg as basically part of the "cognitive prediction" stream of this...

The second one basically tries to go beyond the individual mouse brain ... to the human mind in 
its fullest expression, trying to use what we learn from the mouse as a foundation for better addressing
larger issues which humans have grappled with for a long time. This is NOT such a crisp challenge,
but I view it as extremely important, and I often feel bad about how much we have neglected 
many aspects of it. I view issues related to language as one part of the second stream.

I am not saying which of the two is "the right one." They are both right, maybe for different people in different mixes
at different times. And so too are a few other important basic challenges which may USE neural networks but are not 
ABOUT neural networks... such as energy challenges, where several new technologies need to work together foir
best results.

Best of luck,


More information about the BMI mailing list