Re: Of Mice and Straw Men


seanpit@xxxxxxxxxxxxxxxxxxxxxxxxxxx wrote:
> Von R. Smith wrote:
>
> > So, basically, you admit that much of your argument depends on your
> > intuitions about the fitness landscape being correct. What is your
> > basis for assuming that your intuitions are correct?
>
> Higher level functions are measurable separated, widely separated, in
> sequence space from all other functions - horizontal and vertical.
> This is not based on intuition. It is measurable and has been
> experimentally demonstrated.


When? By whom? How was it measured and experimentally demonstrated?


>
> > They weren't in
> > the case of Zachriel's word-generator experiments, and that was a
> > fairly tractable case, small in scale, with a straightforward fitness
> > function and the landscape exhaustively laid out for you (the total
> > list of "functioning sequences" being unambiguously given by a Scrabble
> > dictionary).
>
> I never said that the low levels of complexity used in Zach's
> experiments could not be evolved.


No, but you made specific predictions about how long it would take, and
those predictions were wrong. More on that below.


> I only said that each additional
> character increase in size and/or specificity would result in an
> exponential decrease in the ratio of meaningful/beneficial sequences
> vs. non-beneficial sequences.


This was not all that Sean said, even though it may now be all that he
is willing to stand by. Sean made specific predictions about how many
mutational steps short strings would take to appear. He predicted it
would take on average 250,000 steps to evolve 7-character words. But it
didn't; the number of mutations Zachriel obtained in his trials ranged
from the hundreds to the low thousands.

Sean further predicted that, from there, the number of mutational steps
would increase *exponentially* as the length of the target string
increased, meaning that the number of mutations should roughly square
each time the length doubled, or cube each time it tripled. That
proved to be wildly off, even for very low values of L.

Confronted with Zachriel's early word generator results, Sean revised
his predictions to the following:


quote:

> L = 4: 20 mutations on average
> L = 8: 20,000 mutations
> L = 16: > 5e8 mutations
> L = 32: ~ 2.5e17 mutations
> L = 64: ~ 6e34 mutations
> L = 128: ~ 4e69 mutations
> L = 256: ~ 1e139 mutations


end quote

Note that Sean has simply taken the largest known value at the time (8
or so) and extrapolated exponentially from there. Zachriel then ran
his word generator longer, and Sean's predictions foundered
immediately. It didn't take 500 million mutations on average, to
generate a 16-character string. It only took about 20 million, and at
least one 19-character word took only a bit longer. It should be noted
that Zachriel's Scrabble dictionary only had a few dozen words over 16
characters long, and none over 23. A *very* small target to hit, and
yet the program did so far faster than Sean had guessed -I mean,
calculated.

Confronted once more with a glaring discrepancy between the evidence
and his predictions, Sean then revised those predictions *again*, to
the following:


L of 16 = 20,000,000 mutations L^6.07
L of 32 = 20,000,000,000 m L^6.85
L of 64 = 20,000,000,000,000 m L^7.35
L of 128 = 20,000,000,000,000,000 m L^7.75
L of 256 = 20,000,000,000,000,000,000 m L^8.02


Note that this revision abandons the principle that number of mutations
will vary as an exponential function of L; it is more like a
logarithmic function now.

At any rate, strings of L = 32 were well beyond the reach of Zach's
first program because there weren't any single words in his scrabble
dictionary that long. So he constructed a Phrasenator that could
generate longer strings and select for sequences of words found in
Hamlet as well as individual words found in a dictionary. He ran a few
hundred million trials, and guess what? Sean's re-revised predictions
were STILL wrong about how many mutations it would take to evolve
strings of 32 and 64 characters. In one run, Zachriel got:


L = 32 ~ 7.4e7 mutations
L = 65 ~ 2.2e8 mutations


In a second trial, he got:

L = 32 ~ 1.8e7 mutations
L = 55 ~ 1.3 e8 mutations

Sean's predicted numbers?

1st try:

L = 32 ~ 2.5e17
L = 64 ~ 6e34

2nd try:

L = 32 ~ 2e10
L = 64 ~ 2e13



> Clearly lower-level functions are highly
> interconnected. Way before Zach came on the scene I demonstrated this
> interconnection by showing the general attachment of meaningful
> 3-letter sequences in the English language system with my use of the
> sequence: cat to hat to bat to bit to bid to did to dig to dog. These
> links are the result of single character changes even though the ratio
> at the level of 3-character sequences in the English language system is
> about 1 in 18.
>
> So, you see, Zach did not disprove any part of my position. Like many,
> he just created a strawman to represent my position that doesn't really
> exist. He also erroneously extrapolated the high-degree of
> interconnections that exist at lower levels to higher levels of
> functional complexity. He doesn't realize that at higher levels these
> interconnections start to quickly break down. I specifically
> challenged him to evolve beyond the level of 1,000 characters. Zach
> said it would be easy to evolve at such levels of meaningful complexity


As I recall, Zach never said that it would be "easy" to do this, if by
easy one means being able to do in real time with the practical
resources available to him, which only allowed him to run his
experiments for a few hundred million generations per trial. Nobody
ever denied that it would take considerably longer to evolve a 1,000
character string than it would a 3- or 7- character string. The issue
was just how long "considerably longer" was likely to be. Nothing in
the available evidence suggests that Sean's intuitions on this subject,
let alone his calculations, were even close.


> and presented various long "O Sean Pitman" poems as examples.
> The
> problem is that he performed his sequence mutations with the help of
> ID.


That is not a "problem" for the point Zach was making, which is that a
mutationally gradual and potentially selectable pathway exists from a
single letter to a long, meaningful sequence. It might still take a
long time to evolve without the benefit of Zach's mind, possibly even
too long given the right conditions, but not for any reasons that
involve "neutral gaps" of meaningless sequences, because there are
none. That Zach used his mind is irrelevant. "Neutral gaps" is a
claim about the fitness landscape in a given phase space, and if
correct would stop *any* search method constrained to taking gradual
steps through said phase space, not just mindless search methods. For
whatever reason, Sean either fails to understand or simply refuses to
acknowledge this point.




> His computer programs never got close to such levels and they
> never could this side of trillions upon trillions of years of average
> time.


Ah, and here is the crux of the matter. Yes, it is true that
Zachriel's programs only got up to strings of characters in the 50s and
60s in a few hundred million trials. But what justifies the move from
that fact to the conclusion that he could not have done have found a
string of 1,000 words this side of trillions upon trillions of years?
Surely not Sean's calculations and extraplations, which were repeatedly
shown to be highly inaccurate; given a known value for the number of
mutations required to find a valid string of L = 8, he couldn't figure
out how long it would take to evolve one with L = 16. Given a known
value for L = 16, he couldn't predict within two orders of magnitude of
the observed value for L = 32, nor within four orders of magnitude of
the observed value for L = 64. And the numbers showed no tendency
whatsoever to converge upon an exponential function; they diverged
further and further from such a function as L got larger. For a while
they looked a bit like a logarithmic function, but even this
deteriorated measurably within observed values. At most, as Zach
pointed out, we are dealing with something like a quadratic function.

And yet, through all this, Sean claims the evidence supports his claim,
when in fact, the most one can say for his claim is the negative virtue
that no one has been able to rub his face in an observation directly
disproving it. He can present no *positive* basis for his "trillions
upon trillions of years", and the supposed logical, mathematical, and
empirical basis for his estimates were shown to be rubbish. There is a
name for an estimate that has no valid logical, mathematical, or
empirical basis: a wild guess.

All Sean is actually left with, for all his confident pronouncements
and pseudo-mathematical rhetoric, is the hope that somehow, somewhere,
something goes wrong with evolution. The best thing one can say about
his wild guess of "trillions and trillions of years" is that no one has
disproven it, in the same way no one has disproven the claim that the
President of the United States in 2052 will be a blind Eskimo. That is
not a lot upon which to hang a supposedly devastating critique of an
important, and well-supported scientific theory.



>
> In short, Zach's programs don't even come close to challenging my
> position or predictions.


As I have pointed out above, Zach's programs falsified every specific
quantitative prediction Sean made about strings up to 64 characters.
They falsified his claim that the number of mutations needed to find
"meaningful" sequences would tend to vary as an exponential function of
L, a fact which Sean tacitly conceded by finally presenting a set of
predictions for L > 32 that abandoned this principle.

The programs didn't run nearly long enough to test anything over one
billion mutations. They *did* run long enough to show that Sean has no
idea how to estimate the length of time it would take to evolve a
string of a given length, even when spotted observed values for strings
up to half that length. Sean is claiming victory based solely on his
ability to set his goalposts beyond immediately practical testability,
and asking his audience to trust that his wild guesses, shown to be
wrong whenever tested, eventually reverse the observed trend and
eventually become accurate.

.

Relevant Pages

  • Re: Of Mice and Straw Men
    ... Those predictions were based on nothing more ... Sean *did* make all sorts of specific ... Zachriel revolved mainly around the number of mutational steps, ... here is an example of Sean's idea of "real life types of mutations": ...
    (talk.origins)
  • Re: Of Mice and Straw Men
    ... Sean made specific predictions about how many ... > different weights to different types of mutations. ... That's how selection works, Sean. ...
    (talk.origins)
  • Re: Of Mice and Straw Men
    ... Sean made specific predictions about how many ... different weights to different types of mutations. ... at such low levels. ...
    (talk.origins)
  • Re: Observations vs. Theories
    ... sean, with his 3rd rate madrasa education provided by the 7th day ... reason regarding religion vs science. ... and newton's predictions enabled OTHER planets to be discovered since ... and why didn't CREATIONISM make such a prediction? ...
    (talk.origins)
  • Re: Predictions made by string theory
    ... String theory makes no empirically-testable predictions. ... Brian Greene predicted they would find strings at CERN during the ... search for the Higgs Boson, and he said they were much larger than ...
    (sci.physics)