Re: AI approach couples biological computation with value


"feedbackdroid" <feedbackdroid@xxxxxxxxx> wrote in message
news:1169578593.120851.220930@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Alpha wrote:
"feedbackdroid" <feedbackdroid@xxxxxxxxx> wrote in message
news:1169403824.892994.298560@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Alpha wrote:


"Nature has equipped biological computations with a measure of
their
value."



Unlike computers,"... biological computations are not lifeless
streams
of
symbols, totally devoid of meaning [*]. Instead, biological
computation
carry something extra - an extra measure of their overall worth.
Instead
of just computation, there is computation plus something else, and
that
something else is a measure of the value of the computation to the
overall success of the organism, its overall fitness. Biological
computtions know how to care"

...................


BTW, he strongly implies and explicity states that today'sompupters do
not
have these features and do not have meaning/valuations tied to
computations
now (especially not in terms of energy efficiency), bolstering the
view
that
computers manipulate meaningless symbols.




Cool, the term "ompupters" is kind of warm and fuzzy and cute. I like
it. OTOH, "... Biological computtions know how to care ..." is a little
TOO warm+fuzzy+cute for my tastes :).

Thanks for the link to the book. Might have some interesting
perspective.

It does; the more I rad the more I appreciate his take on what brain is
doing. He is big on brain doing modeling as well, for predictive *and
learning* reasons! E.g., our imaginantion enables learning while "not on
the
actual job". Hawkins is big on feedback as prediction_enabler.

He discusses RL at length too - something Curt should be interested in.
Critics and all that.

Clearly, the problem with our computer AI's is they're
simply computations, unconnected with the real-world. OTOH, ogranisms

Cool back to ya: the term "ogranisms" is a reference to granny's orgasims
right! ;^))



Sorry, there's only room for one wiseguy on this forum. Me, not you.

BTW, I got over to Barnsesnoble to check out Montague's book. So far,
I'm very equivocal about it. First, no pictures :), and lots of
rambling. Very loose analogies. More importantly [from memory], about
page 65, his idea of the "mutual modeling principle", and the 2nd rule
of efficient computation = model everything, I feel are well off the
mark.

Hmmm; I don;t think so, the more a sender-receiver (of signals/messages)
know about ach other, the more they can craft such a message with less and
less ambiguity in the message; i.e., it can be crafted to fit the input
needs of the receiver (no translations necessary etc.) leading to more
efficient use of bandwidth and computing resources.


He gives an anecdote about a husband and wife driving to the same
airport from different directions, and how each must have a model of
the other's actions in their minds in order to eventually link up
successfully, given the limited amount of intercommunications available
to them [must be before the days of cell phones]. Then he tries to
extend this analogy to brain operation.

He states that brain areas which form reciprocal connections with other
areas "must" each both model themselves and also model the other areas
to which they're connectioned. This must be done to "reduce
intercommunications" in the brain, apparently in order to improve
energy efficiency, but this is plain impossible. He seems to think that
producing local models takes less energy than sending signals to other
areas.

In fact, a brain area will do well to create even "one" local model,
let alone multiple local models simultaneously.

Why do you say this?

And this, especially
given that, on average, each cortical area is connected to close to
HALF of the other cortical areas, as Edelman is very fond of
illustrating in his various books. Even if a brain area does create
some kind of local internal model, it seems ridiculous to think that it
could also produce DOZENS of additional models of what other areas are
doing simultaneously. Naw.

Ok; perhaps the "must" is too overbearing. Possible may be more like it.
For complex interactions between complex local neuronal groups I can see
modelling overcoming the communications bottleneck in terms of efficiency.
I don;t think each neuron models the activities of each and every other
neuron that it may communicate with.





--
Posted via a free Usenet account from http://www.teranews.com

.