At another site, a person posted that there was a papered titled “Evolving Circuits with Mutation and Selection” at http://www.cogs.susx.ac.uk/users/adrianth/ices96/paper.html and asked for comments. These researchers results were mentioned at this site elsewhere. What follows are portions of my responses from the other site.
DNAunion: The selection was performed by intelligently-created algorithms – not by the circuit or the environment. Each “generation” consisted of 50 different “genotypes” being tested for their respective “phenotypic” expression (I love the way the authors throw around biological terms so loosely), and according to the results of a formula applied to the outputs, the most “fit” was retained. The researchers called upon a separate mathematical formula (which they had a hand in creating and tuning) to determine which “genomes” to retain and which to discard – the circuitry itself had no idea which were more fit – the source of the selecting was external to it.
The researchers’ algorithm basically looked at the average of the output voltages for the 1 and 10KHz inputs, and retained the “genome” that showed the greatest difference between the two outputs.
And that formula had to be “fine tuned” by the researchers in order to be useful:
[quote]“The calibration constants k1 and k2 were empirically determined, such that circuits simply connecting their output directly to the input would receive zero fitness. Otherwise, with k1 = k2 = 1.0, small frequency-sensitive effects in the integration of the square-waves were found to make these useless circuits an inescapable local optimum.”[/quote]
And
[quote]“Earlier experiments, where the evaluation method only paid attention to whether the output voltage was above or below the logic threshold, met with failure.”[/quote]
And
[quote] “It should be recognized that to evolve non-trivial behaviors, the development of an appropriate evaluation
technique can also be a non-trivial task.”[/quote]
The circuitry itself did not do the selecting, and the only way the software (genetic algorithm) was able to select the “most fit” was because it was programmed with an intelligently-designed formula that had a specified, predetermined goal – this is not how biological evolution works. In fact, trying to relate this to undirected biological evolution is vulnerable to the same debunking of Dawkin’s “cumulative selection via a combination lock” analogy in Behe’s “Darwin’s Black Box”.
***************************************
There are 2 primary components of this in silico circuit evolution experiment – the software and the hardware, both of which were designed.
SOFTWARE:
[anti-IDist stated that it made no difference that someone wrote the algorithm because it modeled natural processes]
That someone wrote the algorithm is not the reason this algorithm fails to model natural processes. The problem is that the selection criteria programmed were directed – there was a predetermined, fixed goal and each improvement (step towards that goal) was meaningless without the foreknowledge of that goal – natural biological evolution has no targets: but this experiment did.
[anti-IDist argued that my rejection of the model meant that I would also reject a computer simulation modeling planetary orbits.]
No. Designing an algorithm to follow the laws of physics, such as for your planetary orbit model, is perfectly acceptable – and in fact mandatory. Gravity (to bind the planets to the Sun) is not a goal – it is one of the four natural forces according to Newton. Inertia (horizontal momentum to keep the planets from falling into the Sun) is not a goal – it is Newton’s first law of motion. Elliptical orbits (as opposed to circular orbits) for planets in our solar system are not a goal – but a fact (Kepler stated this and Newton later explained it). Etc. Incorporating all of these into an algorithmic model of planetary orbits is required – to leave one or more of them out, or to “fiddle” around with them, would invalidate the model. A computer model of planets in orbit is perfectly valid if it takes into consideration the position, mass, momentum, etc. of each object and follows the laws of physics (whether Newtonian or Ein!
st!
einian) governing their motions
.
Using intelligence and/or information to create an algorithm does not invalidate it – again, the problem is that the model of circuit evolution involved a predetermined goal, and that goal was used to determine which “individuals” in each generation were more fit. This use of external teleological mechanisms does not model natural selection.
[anti-IDist argued that having a goal just models nature, where there is a goal of life: survival]
How was the “survival” of the chip enhanced by outputting 5 volts for a 1KHz input and outputting 0 volts for a 10KHz input? Not at all. We must exclude the selection performed directly or indirectly by the experimenters themselves – the chips did not “die” if their outputs were inappropriate – they survived just as well as the others. It was the human-created algorithm that “killed” them, not their physical attributes. What natural law states that outputting 5 volts for a 1KHz input is better than outputting 0 volts? None. It was the genetic algorithm that determined such “fitness”, and the current fitness of each selection was determined by referencing a predetermined, fixed, future goal.
[discussion of Dawkin’s METHINKSITISLIKEAWEASEL program snipped – will post elsewhere]
HARDWARE:
The fact that a computer and its logic gates are used in modeling planetary orbits does not mean that the “naturalness” of the model is invalidated – the computer hardware is merely a tool used to perform calculations and is peripheral to the model. But if the computer hardware itself (such as the logic gates) is one of the components of the actual system being modeled, as is the case in the in silico evolution of circuits, then the fact that the logic gates themselves arise only with the aid of intelligent agents does invalidate the “naturalness” of the model.
So the question is, “How did the apparatus on which "random mutation and natural selection" occur come into being?” Answer: only with the assistance of intelligence (that is, unless someone can present peer-reviewed literature of field programmable gate arrays (FPGAs) and software to run them forming spontaneously in nature).
[anti-IDist claims the FPGA is simply a collection of simple devices]
If the individual devices are so simple, then surely you can provide some references for their purely natural formation and assembly into FPGAs. In fact, we don’t even need the actual devices – appropriate analogous natural devices would suffice. Let me provide you with a list.
(1) Natural NAND gate (or analogous device).
All logic gate functions can be derived from one or more NAND gates. For example; a two-input AND logic gate can be constructed by using the same two inputs for two separate NAND gates, then wiring their outputs as the inputs into a third NAND gate; a NOT logic gate can be constructed by splitting a single signal into two identical signals and subjecting them to a single NAND gate; a two-input OR logic gate can be constructed by using two NOT logic gates (constructed as above from NAND gates), one for each of the two inputs, and wiring their outputs as inputs to a third NAND gate; etc. So a natural NAND gate (or analogous device) would allow for all possible logic gate functions.
(2) Reusability: Whatever the natural NAND gate device might be (if one exists), it must be reusable. For example, if you come up with some method of using the outcomes of trees falling (somehow) to build such a logic gate, you must explain how such a mechanism can be used again and again.
(3) Construction of other logic gates from NAND gate.
Whatever the natural NAND gate device might be (if one exists), a plausible method of hooking them up properly to allow for the construction of the various other logic gates (AND, OR, XOR, NOT, etc.) must be provided.
(4) Reconfigurable assembly.
Whatever the natural NAND gate device might be (if one exists), a plausible method for changing the logic gate function (for example, from AND gates to OR gates) in the assembly must be provided. In addition, how one “cell” of these devices can have its output reassigned to different “cells” at different times must be explainable.
So if anyone can meet the above 4 criteria, then they will have demonstrated that the hardware portion of the experiment might be able to arise in nature (unless I have overlooked a property). However, as it stands, the only method of constructing the hardware (the actual hardware, or even an analogous device capable of performing the same logic-gate functions) absolutely requires the intervention of intelligent agents.
What does the experiment actually demonstrate? We all know that cars, computers, radios, etc. have all undergone intelligently-directed evolution – but the evolution required continual redesigns and recreation. However, this in silico evolution of circuits had almost all of its intelligence provided at the beginning – the initial conditions were specified and actualized (the hardware and software were designed and created), then the rest was basically accomplished without further intervention by intelligent agents (those steps that humans did perform during the experiment can eventually be performed completely within the hardware and software itself). What this in silico evolution of circuits demonstrated was that devices created by intelligent agents are capable of undergoing “hands-off” evolution.
************************
[anti-IDist presents his own example, a computer model of bear evolving long hair, and states that my criteria would reject it also because there would be goal involved]
Your example differs from the experiment for several key reasons, making your model irrelevant to the question at hand.
First, the conditions/environment are changing throughout yours, but not in the circuit experiment (which involved the same two frequency levels being input for each of the thousands of rounds).
Second, and more importantly, in your example, incremental function is possible as the environment is ever changing.
Over many years, say, the average temperature drops 2 degrees. Hair that is 1/8 inch or so longer would be an advantage – it would serve a useful function and could be selected for were it to arise. Over another period of years, the temperature rises 2 degrees. Now, hair that is 1/8 inch shorter would be an advantage and selected for were it to arise. Over the next several cycles, the temperature continues to drop about 2 degrees per cycle. In each case, slightly longer hair would be beneficial to the possessor and would be retained by natural selection were it to arise.
Thus, your model moves from one functional outcome to another functional outcome – multiple examples of individual selection events. And we have valid reasons why each selection is retained – and it is because of its current utility: it is not based on some future goal.
On the other hand, in Dawkin’s “METHINKSITISLIKEAWEASEL” and the in silico circuit evolution there is only the 1 final function/goal – hitting the target Shakespearean sequence in one case, and hitting +5 volts and 0 volts in the other. Again, why are the intermediate selections advantageous if they fail to be functional. Why would “ME THNIKS IT BE LIK A WASIL” be selected over “ME OMERTIME LKIS WLSEAS TAKN”? Only because the first string is closer to the single, predetermined goal. Each mutated string is compared to the final goal and a difference between the two is calculated – the mutated string with the smaller difference is then considered to be the “most fit”. But why? There is no reason unless there is a predetermined goal – a future target guiding the process, and this is not how biological evolution operates. Selection works only on current function, not on future function. These example!
s !
of selection demonstrate externa
l teleological mechanisms – exactly what Darwinian evolution prohibits. Why if random processes (instead of a predetermined goal) are guiding the processes do we end up with “METHINKSITISILIKEAWEASEL” instead of “MYDARLINGCLEMENTINE”? Why are we so lucky that the end results just happen to match our pre-specified goals? Because we explicitly directed the processes towards those goals.
The word “goal” should never appear in any discussion relating exclusively to natural selection: evolution has no goals – naturalistic evolution abhors the use of external teleological mechanisms. A goal implies a future target towards which the process strives – it turns “natural” selection into “directed” selection.
Face it. The in silico circuit evolution experiment absolutely required both the hardware and the software, and they both absolutely required intelligence for their design and creation, and the selection does not model nature – it incorporates the external teleological process of comparing each output along the way with a desired, predetermined, fixed, goal and retaining the one that is the closest match.
This archive was generated by hypermail 2b29 : Mon Oct 09 2000 - 13:28:12 EDT