Re: Evidence for Big Leaps?

Windy wrote:

Would you use this formulation in physics? Is there "intelligence"
picking which atom will experience fission next in a lump of
radioactive substance?

Well, the least action principle formulation of dynamical laws
tells you that every particle and every field behaves, precisely
in all respects, as if each element is seeking to minimize some
'cost' or 'penalty' function which depends on the past and future
states of the system. If there is interaction, the 'cost' that
each particle is minimizing includes not only contributions
from its own state but also the contributions from the states
of other particles and fields.

Note that at the fundamental level (Quantum Field Theory/QFT),
the basic laws are irreducibly indeterministic i.e. for a
given initial state of a system, there are multiple future
states in which it can transform, without any interaction
with other systems and with initial state replicated absolutely
identically from one test to another. Hence, one could say
that the laws and the initial/boundary conditions do not fix
what will particle/field do next (in its general pursuit
to minimize the 'cost') i.e. there is a sort of 'free will'
at the most fundamental level. Further, there is a formulation
of fundamental physical equations (Maxwell, Schrodinger, Dirac)
in which these equations are coarse grained statistical
properties of actions of large number of cellular automata,
each automaton pursuing its own little utility function
(e.g. see http://www.cft.edu.pl/~birula/publ/lattice.pdf and
misc. papers by G. N. Ord: http://www.scs.ryerson.ca/~gord/ ).
Now, I don't what is it like to be an atom in pursuit of
the 'minimum cost', but I do know exactly what is it like
to be a particular collections of atoms in such pursuit.
I thus see no reason why there shouldn't be something that
it is like to be an atom (see on philosophical perspective of
panpsychism: http://plato.stanford.edu/entries/panpsychism/ ).

A virus has been assembled. Not all the way from atoms or
> anything like that, because that would be stupid, but
> it would be possible.

Well, that was done with using quite a few of commercial
biotech ingredients (enzymes, proteins) of varying
complexity, produced by the live organisms in the first
place.

How does cellular differentation alter the DNA states? And since the
massive mixing of states with recombination must occur before your
single nucleotide mutation can be expressed, is your model of picking
of any use?

I was talking of persistent DNA state changes, that involve massive
synchronized changes to multiple far away locations. The "state" in
case of differentiation is not the coarse grained state, such as
coding sequence, but the finer grained quantum state (or just a
sufficiently detailed chemical state), but which still has a
permanence across multiple cell generations, just like the
coding sequence.


So? My example shows that including "all possible DNA states" in the
probability calculation leads to an erroneous conclusion.


It is not the inclusion of all states that is the problem
but the simplification of assigning _equal probabilities_
to all final states. That simplification is fine for the
purpose it was used (to explain how one could formulate
criteria for empirically distinguishing RM from IA conjectures,
i.e. to show existence of such criteria), or to get a rough
idea of kind of magnitudes involved, but not for much detail.

If one were to use the exact probability distribution of
the final states (which exists, since, at least in principle,
it follows from the initial state and the dynamical laws),
you could get correct values for the probability of any
subset of final states, such as the 'favorable' subset.


No, it doesn't postulate "a very particular relation". It postulates no
relation. This has been demonstrated. If you want randomity to be
tested separately in all possible experiments on natural selection,
sure, you would find some that deviate from randomity by chance because
of the small sample size. How do you propose the researchers should
estimate the relation in the mouse case with only one known beneficial
mutation?


The exact and the three simplifying models of DNA state (max entropy,
RM and ID conjectures) are all probabilistic models, they are
all expressed in terms of probability distributions of the final
DNA states. Hence there is nothing meaningful in testing for
"randomness", since they all have randomness. It is the finer
properties of the randomness, which by itself is common to all,
that distinguish one conjecture from another.


The opposing (benevolent) ID conjecture is, like the
RM conjecture, an additional constraint on the final
distribution of DNA configurations, which says that
the final distribution of the DNA configurations
_will_ be biased in favor of the configurations which
will turn out to be (statistically) more favorable
later.


And this has been tested in several cases and no bias in favour of
beneficial mutations has been detected. Why continue? Do you have some
evidence that suggests otherwise?

You can't tell whether there is a 'bias toward favorable' unless
you know what the outcome (distribution) would be without the
'bias toward favorable'. The 'bias toward favorable' is not
the same thing as 'favorable', which is what you and others
here seem to be assuming. Consider a gambler who is cheating,
which is form of a 'bias toward favorable'. Does that imply
that he is also making money? Not at all. He still may be
losing money, and that depends on the baseline probability
distribution for 'making money' without his bias. In the
gambling analogy, the RM conjecture is that no one, neither
players nor the casino are cheating, while the ND conjecture
is that at least the players are cheating and possibly
the casino.

Your argument (and of others here) in gambling analogy is
that one does not need to know or consider the precise rules
of the game that would allow you to compute the baseline odds
(the game need not be symmetrical or fair) or the skills of
the players, or how the variety of color coded tokens being
exchanged translate into money (e.g. there could be a game
where some tokens may have negative value), to declare that
no one is cheating simply by observing that you don't see
anyone having a much bigger token pile front of them than
the others, hence you conclude that no one can be cheating,
hence the RM conjecture must be correct. My argument is that
you need to know at least what the baseline odds are, what
are the values of different token colors and how many
tokens did each player start with, before you can deduce
from the rough sizes of token piles alone that no one is
cheating. You might not even know what the cheating would
look like in such a game, hence you may not recognize
it even if you are looking straight at it in the middle
of the act.

So no bias towards favourable mutations there, either.
> Do you have a problem with that conclusion?


There was a bias toward some favorable and some unfavorable
mutations (among all possible final states), the net
outcome of all biases being favorable (the bacteria managed
to adapt to the challenge and survive). Note that the bias
here consisted in the increased weight for the mutated
final DNA states, and decreased weight for the non-mutated
states (probabilities for all possible final states have
a fixed sum, 1). The resulting distribution had a property
of the (statistical) net increase in the probability of
favorable states, which is what ID requires. The states
whose porbability was decreased were the non-mutated states,
which were unfavorable in this envirnoment (a certain
starvation).

The ID does not require that all unfavorable states must
have decreased probability or that all states with increased
probability must be favorable. It only requires that the net
result of all the IA processes/algorithms must be statistically
favorable (compared to the case of no IA activity). In this
case the IA process/algorithm was implemented in the
biochemical substratum in the form a 'stress response'
mechanism which increases the general mutation rate.

The ID conjecture does not require or prohibit any
particular implementation (or even the nature of
substratum) of the IA algorithms. It only requires
that its basic performance standard for such algorithms
is met -- the net statistical gain compared to not
executing the algorithm(s). ID being a statistical
requirement means that any candidate algorithm has to
go through a phase of large statistical uncertainty,
hence there will be candidate algorithms which result
in the net loss. Further, since the rest of the system
is not staying fixed (but each component is running
and developing its own IA algorithms), even the
algorithms which have passed the initial candidate phase
in the original environment, remain candidates for any
new environmental challenges. A few bits on how all
these interacting and overlapping IA processes might
fit together into a larger pattern was sketched in
an earler post:

http://groups.google.com/group/talk.origins/msg/2c5884a907f10c22

As you may have noticed, an aspect of ID is only a perspective,
a particular (algorithmic) way of looking at the biochemical
processes, a heuristic. But it also contains a probabilistic
conjecture with sharp mathematical criterium which, at least
in principle, can discriminate between ID and RM conjectures.
As suggested in previous post, the Cairn's experiment has
already falsified the RM1 conjecture, the original RM at the
time, and now we have RM2 which excludes from its prohibited
mutagenic processes list the particular 'stress response'
found to lead to the net favorable outcome in the Cairn's
example. It also provided an early example of the IA process,
with likely many more to follow (along with further evolution
of RM2 into RM3, RM4,... with ever more crossed entries and
exceptions on its prohibited mutagenic processes list).


An elevated general mutation rate was never a bias toward
> favourable states.

Without the elevated mutation rate they wouldn't have survived.
Hence the net result of all biases for that environment was
favorable for the bacteria.

All that ID conjecture implies in a general case
is that the rate of beneficial mutations would be
greater than whatever the RM conjecture would predict
in any given circumstances.

So what is responsible for neutral mutations?

If you shift a probability distribution curve (such
as Gaussian) centered roughly around some nominal
'neutral' value M to the right, that doesn't mean
that the probability for M or for points left of M
becomes zero. It only may mean that it is smaller
than before the shift. Your argument here seems to
be that since there is a nonzero probability for
value M and for values left of M, the curve could
not have been shifted from the initial state. My
argument is that you cannot make such deduction
from the observed nonzero probability at M and left
of M, alone. You need to know more, such as what
were their probabilities before the shift. Note
that depending on where precisely the nominal
"neutral" value M is relative to the curve maximum,
the shift could also lead to the increase of
probability for M (due to changing enough of
negative values into the neutral).

.

Relevant Pages

  • Re: Computability
    ... I am sorry if my difficulties with exposition led you to believe ... And the conjecture _is_ interesting. ... probability and that that order will be the order of exposition next ... client cognitions in a pattern of correalations like "when client ...
    (sci.math)
  • Re: Am I a crank?
    ... Goldbach Conjecture and was thinking of programming it up to see if I ... could find any patterns. ... probabilistic reasons. ... that the conjecture is true with a probability of about 1-10^. ...
    (sci.math)
  • Re: x^n + y^n (mod p)
    ... the probability that ... conjecture is false is the infinite product ... hence the infinite product is 0. ... number theory is full of innocent-looking problems that turn out ...
    (sci.math)
  • Re: x^n + y^n (mod p)
    ... the probability that ... conjecture is false is the infinite product ... hence the infinite product is 0. ... For each positive integer n, and all sufficiently large primes p ...
    (sci.math)
  • Re: Hardy-weinberg Equilibrium
    ... while panmixis means equal probability of any ... But suppose we assumed a normal distribution? ... Are you claiming that statistical randomness requires a uniform ... For example, mutation ...
    (talk.origins)