[dorkbotdc-blabber] Bayesian theory of the brain in New Scientist

[Philip Kohn]

You can always make any problem into energy minimization.
If you define energy as a measure of how good the poem is (quite 
dependent on who does rating, of course),
you can make a zillion dimensional space of all possible poems, where 
Shakespeare's sonnets would
be deep holes (slight changes to a poem might still be pretty good, so 
it might not be that steep a pit).
Does this do you any good?   Most likely no.

It is always hard to figure out what is really going on from the lay 
press.  I can't find any recent Friston
publications on it (all his recent Bayesian work has been on how to 
analyze neuroimaging data).
He may have made some claims like that, but I hope not.  Anyone who 
studies the brain knows that
there are most likely many "rules" involved.  That is one difference 
between physics and biology,
evolution is a hacker, it just makes things work without much concern 
for how complex they are.

And even if there was a Schroedinger's equation for the brain, it may 
still be useless in practice just
as quantum mechanics is not useful for understanding how a bicycle works 
(which is actually surprisingly
complicated using classical mechanics).





Lorne Covington wrote:
>
> This looks like a repackaging of known concepts that really only applies
> to perception.  It seems to be about modeling how the mind creates a
> more correct internal representation of the external world at a low
> level, not about how the mind does things with that information, much
> less create novel thoughts.  Bayesian methods are certainly cool, used
> them back in the '80s to deduce likely position and heading from
> multiple nav sources with differing errors.
>
> Let's say you see something in your peripheral vision.  Maybe it's
> surprising, and you turn to look at it to get more information.  Maybe
> it's not, compared to other mental and perceptual processes, and you
> ignore it, probably without even knowing it.  So I don't see how
> "reducing free energy" explains the mental processes going on that
> decide whether or not to seek more information about that object versus
> to continue eating, finishing that soldering job, or finding that
> perfect phrase for your poem.
>
> Certainly the fact that the cortex has feedback paths is no great news.
> Others have gone farther to say those feedback connections also provide
> the way for higher brain layers to model or synthesize sensory
> experience, such as when you play or sing a tune in your mind, or
> picture something in your mind's eye that you have never seen (a zorse,
> say).  Don't see how Bayesian methods can do that.
>
> Jeff Hawkins "On Intelligence" gives a pretty good overview of this type
> of thing, but I think he tends to claim too much as his own insight.
> Plus he's trying to patent it all.  He does have a cortex modeling kit
> you can download for free for non-profit work, but be warned you'll sign
> away a lot of future rights to get it (numenta.com).
>
> So to answer the question posed by the blog, "is there a unified theory
> of the brain?", the answer appears to me to be "no".  Of specific
> perceptual processes, perhaps.
>
> Thanks for the link, I will go read the source papers!
>
> Cheers!
>
> - Lorne
>
> P.S. - Other books I've found fun: "Descartes' Error" by Antonio
> Damasio, and "Sweet Dreams" by Daniel Dennett.
>
>
> Alberto Gaitán wrote:
>> On our topic of modeling creativity, here's an interesting take on 
>> modeling the human brain as a Bayesian engine:
>>
>> http://reverendbayes.wordpress.com/2008/05/29/bayesian-theory-in-new-scientist/ 
>>
>>
>
>
>
> ........................................................................
> .......dorkbot dc: people doing strange things with electricity.........
> ................... http://dorkbot.org/dorkbotdc .......................
> ........................................................................
>