|
Information as lingua franca
Hi Mikhail, You suggested: One could attempt
to measure mutual information I(X;Y) between *gene sequences* X and *functional
behaviours* Y - without specifying or even assuming a precise
relationship between f and Y
["f" and "between" switched. RA].
It could be said that
genes (or the proteins they produce) encode *information* RATHER THAN specific
regularities about the environment!
Would you be willing to elaborate on this? For example, if you apply this to software, would this imply that you can measure the mutual information between the code and the function it computes? Can you work out a simple example for me? One of my favorites is the idiom that exchanges the values in two variables x and y.
temp = x;
x = y;
y = temp;
How would you express the mutual information in the code (above) vs. the function (the values in x and y were exchanged)?
-- Russ On Nov 24, 2007 11:00 PM, < Mikhail.Prokopenko@...> wrote:
Hi Russ and all,
I agree, of course, that genotype-phenotype
relationship is not a simple X = f(Y) relationship :) And emergence plays
a role in the process: I completely agree that "there is probably significantly
more to be said than that genes (or the proteins they produce) encode specific
regularities about the environment."
Russ argued that
---------
Nature typically works that way; it builds new stuff
on top of existing stuff. So if one has a genetic mechanism that results
in some functionality, that functionality may be incorporated into some other
functionality if the new functionality turns out to be useful.
This view
would propose to look at biological mechanisms as increasingly complex levels of
abstraction. Some of those levels of abstraction produce functionality in the
world. Others produce internal functionality. The overall effect result is that
beings with certain functionalities survive better in certain
environments.
---------
This may be a good representation. If we are
to learn from Nature (and I don't mean bio-mimetics) we have to start
somewhere. Information Theory may be one good tool that transcends simple X
= f(Y) relationships (read: simple correlations), and captures the [probably
nested] relationships leading to functionality. For example, one could attempt
to measure mutual information I(X;Y) between *gene sequences* X and *functional
behaviours* Y - without specifying or even assuming a precise
relationship f between and Y. It could be said that
genes (or the proteins they produce) encode *information* RATHER THAN specific
regularities about the environment!
If we assume that a) environment has a "statistical
structure", b) agents' genome has a "statistical structure", c) agent's
dynamics/functionality in the environment have "statistical structure", then
What's the information dynamics among these (and other conceivable) abstraction
levels, is a valid question, I think.
In short, the view you
advocated (increasingly complex levels of abstraction), as far as I
understand, is not at odds with IDSO.
I'm curious if "neutral emergence" view (Andrew Weeks,
Susan Stepney, Fiona Polack)
is related to this thread? Susan, would you like
to comment on this?
Cheers,
Mikhail
P.S. I'm still using IDSO but happy to switch to
"information-mediated" rather than "information-driven" at any time when we
reach a consensus :)
From: Abbott, Russ
Sent: Saturday, 24
November 2007 6:25 AM
To: Prokopenko, Mikhail (ICT Centre, North
Ryde)
Cc: IDSO-CSIRO
Subject: Re: [IDSO] strong
IDSO
Thanks, I'll have a look at the paper. Regarding your
question:
If, say, y
is a gene, and h(y) is the protein that corresponds to some regularity in nature
... then wouldn't one be
able to say that gene y itself also corresponds to this
regularity?
It seems to me that neither the gene nor the protein
typically corresponds to some regularity in nature. In my original message I
should have said "is advantageous given some regularity in nature" instead of
"corresponds to some regularity in nature." Pardon the following
fairly long discussion. The example I used was skin color. I freely
admit to not being an expert in the field. The Wikipedia article (and
I don't claim that Wikipedia is always a reliable source) paints a fairly
complex picture of the relationship between genes, skin color, latitude, etc.
One of the basic mechanisms (it claims) is that like chimpanzees, humans had
light skin under our fur. When we lost our fur, dark skin became essential in
those parts of the world where the sun was brightest as a way to protect against
skin cancer. The genetic configuration for dark skin (lots of melanin) is
apparently fairly specific. Any mutation results in significantly lighter skin,
which tends to be disadvantageous for survival in the hot sun of Africa.
According to this theory, the environment selected for dark skin, not light
skin. When that environmental pressure wasn't present, mutation wasn't so
disadvantageous, and lighter skin survived. In addition, lighter skin is
advantageous for the production of vitamin D. So the cartoon version of
this is that there is a gene for skin color. One specific DNA sequence
produces dark skin; all other versions produce lighter skin. Dark skin is
advantageous where the sun is hot and strong (because of the skin cancer
connection). Light skin is advantageous where the sun is weak (because of the
vitamin D connection). Both of these connections are correlated with the
strength of the sun. But neither of them is directly connected. That is, dark
skin with weak sun is OK if you have another source of vitamin D (as the article
says the Inuit do). Light skin with strong sun is OK if you have lots of fur (as
our primate ancestors did). So I think it's a stretch to say that either
a gene or the protein it produces records information about the environment.
It's much more complex than that. And even if it were more directly
connected (without the fur or the other sources of vitamin D), the gene controls
the production of melanin. Assume that's all it does. (It's probably even
more complex than that.) The amount of melanin has the effect of making the skin
"lighter" or "darker." But why does that matter? It matters because that affects
how the skin will respond to sunlight. But isn't that backwards? Light colors
reflect sunlight, and dark colors accept it. In this case, though melanin acts
as a sun (UV) block rather than as a reflective color. So the more melanin, the
less sun. But:
- Do we care how the specific configuration of nucleic acids produces
lots of melanin and other configurations produce less? Or is that a level of
abstraction that we can ignore?
- Do we care about the mechanism whereby melanin acts as a sun block?
Or is that another level of abstraction that we can ignore?
- And finally, do we care about what happens when UV rays are blocked or
not blocked , i.e., the biological mechanisms that produce skin cancer and
vitamin D? Or is that still another level of abstraction we can
ignore?
As you know, I'm all in favor of understanding nature in
terms of emergent phenomena. So some of these mechanisms may be subsumed
under the mantle of emergence, and we may be able to talk about them as levels
of abstraction. But even so, it seems to me that there is probably
significantly more to be said than that genes (or the proteins they produce)
encode specific regularities about the environment. Having gone this
far, I would say that it probably makes more sense to understand genes and the
proteins they generate as part of a larger collection of mechanisms that working
together produce useful (or not-so-useful) results. Nature typically works
that way; it builds new stuff on top of existing stuff. So if one has a
genetic mechanism that results in some functionality, that functionality may be
incorporated into some other functionality if the new functionality turns out to
be useful. This view would propose to look at biological mechanisms as
increasingly complex levels of abstraction. Some of those levels of abstraction
produce functionality in the world. Others produce internal functionality. The
overall effect result is that beings with certain functionalities survive better
in certain environments. For example, being able to extract oxygen from
water is good if you live in the water, and being able to use oxygen in the
atmosphere is good if you live in an atmosphere with oxygen. It's not so much
that genes encode information about the environment; it's that genes provide
certain functionalities, which are useful in certain environments.
Again, apologies for the length of this post. I got carried
away. -- Russ
On Nov 23, 2007 12:21 AM, < Mikhail.Prokopenko@...>
wrote:
Hi Russ and
all,
The paper I mentioned
earlier
M. Piraveenan,
D. Polani, M. Prokopenko. Emergence of Genetic Coding: an
Information-theoretic Model, in F. Almeida e Costa, L. M. Rocha, E. Costa, I.
Harvey, A. Coutinho (eds). Advances in Artificial Life: 9th European
Conference on Artificial Life (ECAL-2007), Lisbon, Portugal, September 10-14,
Lecture Notes in Artificial Intelligence, vol. 4648, pp. 42-52, Springer,
Berlin, 2007.
cab be d/loaded
from
I should clarify that this paper is
not intended as a statement on IDSO at all, but attempts to crudely model
possible emergence of proto-symbols (rudimentary genes) that
capture dynamics of an object (a rudimentary cell) within its
environment.
Russ, you
make an interesting distinction that "the genes themselves generally
don't encode those regularities", but rather "just generate proteins that
result in light skin, which correspond to certain facts about nature where
those genes evolved". If, say, y is a gene, and h(y) is the protein that
corresponds to some regularity in nature - let's ignore the
trivialisation of the functional form h(.) for the moment - then wouldn't one
be able to say that gene y itself also corresponds to this regularity?
...and maybe even encodes it in some way?
Thanks,
Mikhail
From: Abbott, Russ
Sent: Thursday,
22 November 2007 3:41 PM
To: Boschetti, Fabio (CMAR,
Floreat)
Cc: Prokopenko, Mikhail (ICT Centre, North Ryde); Lafusa,
Antonio; IDSO-CSIRO
Subject: Re: starting IDSO discussions: weak and
strong IDSO
Hi,
I can't resist jumping in here.
Mikhail's
original statement was, I gather, in support, for example, of Adami's claim
that "the evolutionary process extracts
valuable information and stores it in the genes." That seems quite
true. Genes do seem to do that sort of thing. That claim, though doesn't say
that everything that happens in nature (or in man-made systems like sodaplay)
involves the explicit encoding and passing of information.
Furthermore, it seems that even strong IDSO is an example of weak IDSO
in the sense that genes do not store information about the world. Yes, one can
say that certain genes evolved to correspond to certain regularities in
nature. But the genes themselves generally don't encode those regularities.
Genes for, say, light skin that evolved in higher latitudes don't say anything
about the length of the day or the strength of the sun. They just generate
proteins that result in light skin, which correspond to certain facts about
nature where those genes evolved.
So I'd like to request that we
attempt to clarify a bit what the issue is that's being discussed.
--
Russ
|
|
 RE: Information as lingua franca
Hi,
Sorry I hasn't replied sooner (too busy lately). I'll
try to answer some of the questions, but first, let me mention again that it's
important to identify the channel(s) over which information (flow) is to be
maximised. For example, Adami argues that one channel may be between "genes" X
and "environment" Z - hence I(X:Z) as a measure. In another example
(Piraveenan et al. paper) we considered what happens to possible "genes" X
when a coupled "phenotype" Y is affected by noise and shifts into
Y' - hence, I(Y;Y') as a measure driving/mediating emergence of
X.
In the following I assume that we are simply interested
in how to compute I(X;Y) in the software example.
Any two code fragments can be same - it doesn't
really matter. Although if all fragments are the same, the entropy H(X) would be
zero. The better n fragments represent the total collection, the better is the
analysis, but it's not necessary to represent the totality. Adami, for example,
used the sequence length (e.g., if we deal with assembler code,
that would be the max length of a program) to approximate max
entropy, H_max(X). The same would apply to "results". If the "results"
depend on the context, we need another variable (Z), and that's what
"environment" in Adami's paper is, I think. It's hard to enumerate
environments/contexts - Adami's paper shows one way to work around
this.
Moving to the questions on how to compute entropies.
These are mostly technical questions, and can be answered in many ways. One
way is to use frequencies (how often fragments occur, etc.). Another way
(parametric) is to assume a specific distribution, say Gaussian, compute
entropies analytically, and estimate parameters from data. One may also use
kernels. The same techniques apply to "behaviours/results" if the space is
defined. We used a dynamical system to represent the "phenotype" level. H(Y) is
computed independently of X, but to compute joint entropy, you'd need to
consider pairs: which fragment results in which behaviour, and how often each
pair occurs among all pairs, etc. In short, it comes down to enumerating
possibilities (if one doesn't use parametric approaches) and building
histograms :)
I'm not sure if we want to get into a technical
discussion on how to compute entropies in this list - there are more interesting
questions, I think, like the one Antonio asked about flows between levels, or
the questions on explicit communication vs stigmergy, or the co-evolution of
perception-action loops, related to your recent post on
squirrels-and-rattlesnakes.
Thanks,
Mikhail
So let's say we had n code fragments. Each one has a corresponding
result when executed ("expressed"). Is it possible for two of the code
fragments to be the same, or must they all be different from each other. Same
question about the results. I'm assuming that the n code fragments must
represent to total collection of code fragments. Is that true? That is, we
can't be interested in a subset of the code fragments. Or can we? Same
question about the results. Although since the results depend on context,
it's harder to say what the total collection of results are. That is, a
code fragment may have different results when executed (expressed) in different
contexts. However the preceding questions are answered, how would you go
about computing H(X), H(Y), and H(X, Y)? Do you need to know how often
each of the code fragments appears in the overall population to compute H(X)? Do
you need to know how often each of the results occurs in the lifetime of the
population elements to compute H(Y). To put this question another way, do
you need to know how often each of the code fragments is expressed (executed)
and thereby produces its result to compute H(Y)? How much of the above do you
need to know to compute H(X, Y) and how is it computed? As you can see,
I'm not at all clear about the overall idea and am hoping for a simple concrete
example to help me understand. I know I should probably read the papers, but you
know how those things are. If I can get a shortcut intuitive explanation by
reading my email, that's so much easier! Thanks. -- Russ
On Nov 26, 2007 12:29 AM, < Mikhail.Prokopenko@...>
wrote:
OK. To compute
I(X;Y), we'd need more examples of X - other code fragments, and corresponding
examples of Y - functional outcomes. Then, having the samples (x1, x2, ...,
xn) and (y1, y2, ..., yn), we can compute entropies H(X), H(Y) and H(X,Y).
Having one
example of X - the three lines of
code x1 - does not allow
to compute the entropy H(X). Adami's paper, on the other hand, does consider
the software world with code fragments, etc. For example, if the line "x = y;"
in your example is mutated away in a fragment x2, the
functionality y2 would change, affecting entropies and mutual information. The
way these quantities are affected depends on the entire population (the
samples).
So the main
point is that we consider statistical structure of the code-fragment space,
the outcome space, and so on.
Mikhail
From: Abbott, Russ
Sent: Monday, 26
November 2007 7:14 PM
To: Prokopenko, Mikhail (ICT Centre, North
Ryde)
Cc: IDSO-CSIRO
Subject: Re: Information as
lingua franca
I was hoping my variable swapping 3-liner would be a simple
concrete example for which you could illustrate how mutual information
worked. I was thinking of the program as the gene sequence (or the
proteins they generate) and the function of reversing the variables as the
function that the proteins perform. So the three lines of code is the X
(the gene
sequence) and the reversal of the values in the variables
is the Y (the functional
behaviour ). How would you work out I(X:Y) in this
case? -- Russ
On Nov 25, 2007 11:47 PM, < Mikhail.Prokopenko@...> wrote:
Hi
Russ,
The sentences
in question have an "X" missing (sorry), and should read:
"one
could attempt to measure mutual information I(X;Y) between *gene sequences*
X and *functional behaviours* Y - without specifying or even
assuming a precise relationship f between X ["X" was missing. MP]
and Y. It could be said that genes (or the
proteins they produce) encode *information* RATHER THAN specific
regularities about the
environment!"
I simply
mean a black box f between input X and output Y. If we apply it to software,
then different X's may, for instance, be different initial conditions to a
problem-solver f, producing different solutions Y. Then, even if the
relationship Y = f(X) is not precise - e.g., there is some random noise in
f, we still can compute mutual information I(X;Y). That is, mutual
information between initial conditions and solutions - not between initial
conditions and the problem-solver.
A more
involved software analogy is a process by which different queries/keys
K to a database/memory X produce, as a result of some mapping f, different
outcomes Y. In this case, one may want to analyse conditional mutual
information I(Y;K|X) - how much a key is necessary to identify the output of
the mapping, given the memory. In this crude analogy, K mimics "proteins"
accessing the "DNA" X, and producing the "functionalities" Y.
I don't
think I understand what are X, Y and f in your example, i.e.
what are the respective analogies of genes, functionalities they
produce, and the process that links them.
Not sure
if the following is related to your question on the function that exchanges
the values of two variables X and Y [please ignore it if it's not
related]:
There are some ways to compute information
flow (not mutual information) between, say, temporal strings X =
0,1,0,1,0,1,.... and Y = 1,0,1,0,1,0 - dependent on whether there is a
function f that connects them. For example, (causal) information flow would
be 1 bit if the function f swaps x_i and y_i at every step i, yet the
flow would be 0 if X and Y are independent of each other and each
simply inverts previous value. Note that the results
differ -
but observationally, these two cases are equivalent - and neither mutual
information nor transfer entropy would detect the difference. This is an
example of interventionist (causal) approach:
Ay, N., and Polani, D., (2007). Information
Flows in Causal Networks. Advances in Complex Systems. Accepted; also Santa
Fe Institute Working Paper 06-05-014
Mikhail
From: Abbott, Russ
Sent: Monday,
26 November 2007 5:09 AM
To: Prokopenko, Mikhail (ICT Centre, North
Ryde)
Cc: IDSO-CSIRO
Subject: Information as
lingua franca
Hi Mikhail, You suggested:
One could attempt to measure mutual information I(X;Y) between *gene
sequences* X and *functional behaviours* Y - without specifying or even
assuming a precise relationship between f and Y ["f" and "between" switched.
RA]. It could be said that genes (or the proteins they produce) encode
*information* RATHER THAN specific regularities about the environment!
Would
you be willing to elaborate on this? For example, if you apply this to
software, would this imply that you can measure the mutual information
between the code and the function it computes? Can you work out a simple
example for me? One of my favorites is the idiom that exchanges the
values in two variables x and y.
temp = x;
x = y;
y = temp;
How would you express
the mutual information in the code (above) vs. the function (the values in x
and y were exchanged)? -- Russ
On Nov 24, 2007 11:00 PM, < Mikhail.Prokopenko@...> wrote:
Hi Russ and
all,
I agree, of course, that genotype-phenotype relationship is not a
simple X = f(Y) relationship :) And emergence plays a role in the
process: I completely agree that "there is probably significantly more to
be said than that genes (or the proteins they produce) encode specific
regularities about the environment."
Russ argued
that
---------
Nature typically works that way; it builds new stuff on top of
existing stuff. So if one has a genetic mechanism that results in
some functionality, that functionality may be incorporated into some other
functionality if the new functionality turns out to be useful.
This view would propose to look at biological mechanisms as
increasingly complex levels of abstraction. Some of those levels of
abstraction produce functionality in the world. Others produce internal
functionality. The overall effect result is that beings with certain
functionalities survive better in certain
environments.
---------
This may be a good representation. If we are to learn
from Nature (and I don't mean bio-mimetics) we have to start somewhere.
Information Theory may be one good tool that transcends simple X =
f(Y) relationships (read: simple correlations), and captures the [probably
nested] relationships leading to functionality. For example, one could
attempt to measure mutual information I(X;Y) between *gene sequences* X
and *functional behaviours* Y - without specifying or even
assuming a precise relationship f between and Y.
It could be said that genes (or the proteins they produce) encode
*information* RATHER THAN specific regularities about the
environment!
If we assume that a) environment has a "statistical structure", b)
agents' genome has a "statistical structure", c) agent's
dynamics/functionality in the environment have "statistical structure",
then What's the information dynamics among these (and other conceivable)
abstraction levels, is a valid question, I
think.
In short, the view you advocated
(increasingly complex levels of abstraction), as far as I
understand, is not at odds with IDSO.
I'm
curious if "neutral emergence" view (Andrew Weeks, Susan Stepney, Fiona
Polack)
is
related to this thread? Susan, would you like to comment on
this?
Cheers,
Mikhail
P.S. I'm still using IDSO but happy to switch to
"information-mediated" rather than "information-driven" at any time when
we reach a consensus :)
From: Abbott, Russ
Sent:
Saturday, 24 November 2007 6:25 AM
To: Prokopenko, Mikhail (ICT
Centre, North Ryde)
Cc: IDSO-CSIRO
Subject: Re: [IDSO]
strong IDSO
Thanks, I'll have a look at the paper. Regarding
your question:
If,
say, y is a gene, and h(y) is the protein that corresponds to some
regularity in nature ... then wouldn't one be able to say that gene y itself also
corresponds to this
regularity?
It seems to me that neither
the gene nor the protein typically corresponds to some regularity in
nature. In my original message I should have said "is advantageous given
some regularity in nature" instead of "corresponds to some regularity in
nature." Pardon the following fairly long
discussion. The example I used was skin color. I freely admit
to not being an expert in the field. The Wikipedia article (and I don't claim that Wikipedia is
always a reliable source) paints a fairly complex picture of the
relationship between genes, skin color, latitude, etc. One of the basic
mechanisms (it claims) is that like chimpanzees, humans had light skin
under our fur. When we lost our fur, dark skin became essential in those
parts of the world where the sun was brightest as a way to protect against
skin cancer. The genetic configuration for dark skin (lots of melanin) is
apparently fairly specific. Any mutation results in significantly lighter
skin, which tends to be disadvantageous for survival in the hot sun of
Africa. According to this theory, the environment selected for dark skin,
not light skin. When that environmental pressure wasn't present, mutation
wasn't so disadvantageous, and lighter skin survived. In addition,
lighter skin is advantageous for the production of vitamin D. So
the cartoon version of this is that there is a gene for skin color.
One specific DNA sequence produces dark skin; all other versions produce
lighter skin. Dark skin is advantageous where the sun is hot and strong
(because of the skin cancer connection). Light skin is advantageous where
the sun is weak (because of the vitamin D connection). Both of these
connections are correlated with the strength of the sun. But neither of
them is directly connected. That is, dark skin with weak sun is OK if you
have another source of vitamin D (as the article says the Inuit do). Light
skin with strong sun is OK if you have lots of fur (as our primate
ancestors did). So I think it's a stretch to say that either a
gene or the protein it produces records information about the environment.
It's much more complex than that. And even if it were more
directly connected (without the fur or the other sources of vitamin D),
the gene controls the production of melanin. Assume that's all it
does. (It's probably even more complex than that.) The amount of
melanin has the effect of making the skin "lighter" or "darker." But why
does that matter? It matters because that affects how the skin will
respond to sunlight. But isn't that backwards? Light colors reflect
sunlight, and dark colors accept it. In this case, though melanin acts as
a sun (UV) block rather than as a reflective color. So the more melanin,
the less sun. But:
- Do we care how the specific configuration of nucleic acids
produces lots of melanin and other configurations produce less? Or is
that a level of abstraction that we can ignore?
- Do we care about the mechanism whereby melanin acts as a sun
block? Or is that another level of abstraction that we can ignore?
- And finally, do we care about what happens when UV rays are
blocked or not blocked , i.e., the biological mechanisms that
produce skin cancer and vitamin D? Or is that still another level
of abstraction we can ignore?
As you know, I'm all in
favor of understanding nature in terms of emergent phenomena. So
some of these mechanisms may be subsumed under the mantle of emergence,
and we may be able to talk about them as levels of abstraction. But
even so, it seems to me that there is probably significantly more to be
said than that genes (or the proteins they produce) encode specific
regularities about the environment. Having gone this far, I would
say that it probably makes more sense to understand genes and the proteins
they generate as part of a larger collection of mechanisms that working
together produce useful (or not-so-useful) results. Nature typically
works that way; it builds new stuff on top of existing stuff. So if
one has a genetic mechanism that results in some functionality, that
functionality may be incorporated into some other functionality if the new
functionality turns out to be useful. This view would propose to
look at biological mechanisms as increasingly complex levels of
abstraction. Some of those levels of abstraction produce functionality in
the world. Others produce internal functionality. The overall effect
result is that beings with certain functionalities survive better in
certain environments. For example, being able to extract oxygen
from water is good if you live in the water, and being able to use oxygen
in the atmosphere is good if you live in an atmosphere with oxygen. It's
not so much that genes encode information about the environment; it's that
genes provide certain functionalities, which are useful in certain
environments. Again, apologies for the length of this post. I got
carried away. -- Russ
On Nov 23, 2007 12:21 AM, < Mikhail.Prokopenko@...> wrote:
Hi Russ and
all,
The paper I mentioned
earlier
M.
Piraveenan, D. Polani, M. Prokopenko. Emergence of Genetic Coding: an
Information-theoretic Model, in F. Almeida e Costa, L. M. Rocha, E.
Costa, I. Harvey, A. Coutinho (eds). Advances in Artificial Life: 9th
European Conference on Artificial Life (ECAL-2007), Lisbon, Portugal,
September 10-14, Lecture Notes in Artificial Intelligence, vol. 4648,
pp. 42-52, Springer, Berlin, 2007.
cab be d/loaded
from
I should clarify that this
paper is not intended as a statement on IDSO at all, but attempts to
crudely model possible emergence of proto-symbols (rudimentary genes)
that capture dynamics of an object (a rudimentary cell)
within its environment.
Russ, you
make an interesting distinction that "the genes themselves
generally don't encode those regularities", but rather "just generate
proteins that result in light skin, which correspond to certain facts
about nature where those genes evolved". If, say, y is a gene, and h(y)
is the protein that corresponds to some regularity in nature -
let's ignore the trivialisation of the functional form h(.) for the
moment - then wouldn't one be able to say that gene y itself also
corresponds to this regularity? ...and maybe even encodes it in
some way?
Thanks,
Mikhail
From: Abbott, Russ
Sent:
Thursday, 22 November 2007 3:41 PM
To: Boschetti, Fabio (CMAR,
Floreat)
Cc: Prokopenko, Mikhail (ICT Centre, North Ryde);
Lafusa, Antonio; IDSO-CSIRO
Subject: Re: starting IDSO
discussions: weak and strong IDSO
Hi,
I can't resist jumping in here.
Mikhail's
original statement was, I gather, in support, for example, of Adami's
claim that "the evolutionary process
extracts valuable information and stores it in the genes." That
seems quite true. Genes do seem to do that sort of thing. That claim,
though doesn't say that everything that happens in nature (or in
man-made systems like sodaplay) involves the explicit encoding and
passing of information.
Furthermore, it seems that even strong
IDSO is an example of weak IDSO in the sense that genes do not store
information about the world. Yes, one can say that certain genes evolved
to correspond to certain regularities in nature. But the genes
themselves generally don't encode those regularities. Genes for, say,
light skin that evolved in higher latitudes don't say anything about the
length of the day or the strength of the sun. They just generate
proteins that result in light skin, which correspond to certain facts
about nature where those genes evolved.
So I'd like to request
that we attempt to clarify a bit what the issue is that's being
discussed.
--
Russ
-- -- Russ
Abbott _____________________________________________ Professor, Computer
Science California State University, Los Angeles o Check out my blog at
http://russabbott.blogspot.com/
-- -- Russ
Abbott _____________________________________________ Professor, Computer
Science California State University, Los Angeles o Check out my blog at http://russabbott.blogspot.com/
|
