[extropy-chat] DARPA Ends Brain Reverse Engineering Project

[Anders Sandberg]

Robert Bradbury wrote:
> I would be interested in whether anyone really *knows* what is going on in
> this area (really)?  Unlike, say the visual processing system, or the
> auditory processing system (which Ray touched on in TSIN), is the
> cognitive
> processing system currently understood at all?  I don't follow this field,
> so I have no idea whether neuroarchitectural understanding (perhaps via
> fMRI) has reached the point where we have tied our external understanding
> of
> what is going on to precise biological architectures (which IMO is what
> they
> were going after).

My answer would be that nobody really knows what is going on. We have a
lot of piecemal information that makes sense or have empirical support,
for example the emerging picture of the prefrontal lobes as having
subsystems for working memory, linking to motivational states, estimation
of value etc and them in turn tied to a motor planning/motor program
storage in the cortex-basal ganglia loop system. We even have some bridges
between the macrolevel and microlevel, such as working memory models and
associative learning. The big problem is that there are a lot of holes in
this understanding, and I personally think we are missing one or two key
principles (such as how to make the neural assemblies form useful
hierarchical or modular concepts). That might change very soon or not for
a long time, it is impossible to predict beforehand.

I know of at least one of the previously funded projects (the Gluck one).
They do have a cool model but I doubt it could be easily or quickly
extended to reach the high aims of the project. The other projects
involved all seemed to use less connectionist models, which probably
improves their ability to mimick psychology but makes mapping onto the
brain very hard.

> So was the problem that we simply do not have the tools yet (e.g. fMRI
> resolution, or fine scale mapping of dendrite trees at billion neuron
> level?) to accomplish stated goals of the project?

None of the projects involved looked like they would be doing anything
uploading-like. The problem is that this is all understanding-based
models: to work we have to understand the high-level principles of the
brain and thinking, and that is hard. Uploading gets around it by just
brute-forcing the system. And as you said, most models used were models of
how the brain ought to think according to the models already assumed by
researchers.



-- 
Anders Sandberg,
Oxford Uehiro Centre for Practical Ethics
Philosophy Faculty of Oxford University