|
Author
|
Topic: Information. ID versus non-ID
|
Roger R
Member
Member # 200
|
posted 26. January 2003 11:55
quote:
Originally posted by Frances:
Cutting and pasting part of a program is hardly equivalent to showing that information was smuggled in, not to mention some of the other unsupported accusations I have heard about Ev.
When I cited Schneider's words about what he was attempting to do, you claimed that he didn't translate that into the program. When I show you where in the program code it is, you dismiss that as well. Like I said, this is one on which we will probably never agree. It goes to the heart of the basic dispute.
But feel free to tell me what exactly would pass your "test" as introducing "intelligent teleology" into the program. Is there any potential evidence? Do you recognize it as even being possible to do so?
quote:
You highlighted the definition of an integer m used to count the current number of integers. Could you explain the relevance of this to the question asked?
I assume that was a typo and you meant the "current number of mistakes". Because the term "mistake" itself imples something about a goal. And he defines exactly what that is, how to determine and quantify it, and selects intelligently on that very basis. And all done in advance.
That isn't natural selection, but intelligent teleology.
IP: Logged
|
|
Frances
Member
Member # 169
|
posted 26. January 2003 14:04
Roger,
When asked about teleology, you mistook Schneider's statement that he was going to see if he could evolve R_freq to R_seq using RM&NS as if Schneider would thus code this goal R_freq=R_seq into his program.
Now you suggest that you 'showed' how his program was coded to achieve this teleology but you have so far been unable to show that the program was coded for R_seq and R_freq to be the same. In fact as I have shown the simulations lead to many different R_freq which statistically are close to R_seq.
If Roger could show that the fitness function for instance was explicitly minimize (R_freq-R_seq) then we have a teleological goal and Schneider's simulations would have little relevance. But that is not what the program does.
Now you suggest that the term 'mistake' is a goal. If you read the program mistake is not a future goal but a present mismatch. Your claims about teleology, Schneider and the program still suggest limited familiarity with what he really did and how he achieved this.
quote:
The recognizerof each creature is translated from the gene form to a perceptron-like weight matrix. This matrix is scanned across the genome. The number of mistakes made is counted. There are two kinds of mistake:
how many times the recognizer misses a real site
how many times a non-site is detected by the recognizer.
Added:
The time to write the program has no relevance on the issue. The goal was to show that R_freq evolves to R_seq but that does not mean that thus the program was coded to ensure this goal. In fact what the program shows is that mutation and selection are able to explain the observation that R_seq and R_freq tend to be very close. Roger seems to confuse the goal "showing that R_freq can evolve to R_seq" with forcing such evolution. As others have shown as well the hypothesis was simply: given RM&NS can we show that R_freq becomes close to R_seq as observed in nature? The program supported the hypothesis, however if the program had failed to show that RM&NS can explain that R_seq and R_freq are close then Schneider's hypothesis would have failed. So in short, Roger should be careful to differentiate betwen 'the goal' and 'goal driven programming'. While the program did support the hypothesis, the goal "R_seq approximates R_freq" was not hard coded in the program. So no teleological requiremens.
If Roger can point out any such teleology then I am looking forward to such but so far no such teleology has been shown beyond the hypothesis formation "Can RM&NS explain that R_seq approximates R_freq".
That's all.
[ 26. January 2003, 15:41: Message edited by: Frances ]
IP: Logged
|
|
Roger R
Member
Member # 200
|
posted 26. January 2003 15:17
quote:
Originally posted by Frances:
When asked about teleology, you mistook Schneider's statement that he was going to see if he could evolve R_freq to R_seq using RM&NS as if Schneider would thus code this goal R_freq=R_seq into his program.
Now you suggest that you 'showed' how his program was coded to achieve this teleology but you have so far been unable to show that the program was coded for R_seq and R_freq to be the same. In fact as I have shown the simulations lead to many different R_freq which statistically are close to R_seq.
Sorry Frances, but that is factually incorrect. YOU glommed onto to the R_freq to R_seq issue, but what I was citing is that Schneider said "I sat down and wrote the ev program in about 2 days (this is only possible if you have a clear picture of where you are going!) and discovered to my delight that in only 500 generations I could get the full evolution, and Rsequence did indeed approach the predicted Rfrequence and oscillate around that level." IOW, he achieved his goal. And I immediately followed that quote in my same post with:
quote:
Sounds like he had a teleological goal in mind. And that goal drives the "mistake" counting in his program. Run the program over again, and it goes toward the same goal. Yes, it varies slightly in how quickly it gets there, due to the randomizing function, but it always goes toward that specific goal, and not other goals that other GA's go towards. Why would that be, if it wasn't goal driven?
I'm not "now" shifting gears. Quite to the contrary, that has been my point all along. I don't have any obligation to show a coding for the equality when that was never my position. It is you who appears obsessed over the R_freq = R_seq equality to the point where you miss the teleology for the trees.
IP: Logged
|
|
|
Printer-friendly view of this topic