A bit of evolution, Torvalds style?

[npetreley]

I mentioned to a friend of mine that Linus Torvalds supposedly said something about evolution and parallelism. My friend's response was very amusing...

If Linus Torvalds believes the Neo-Darwinian Theory of Evolution, I have a challenge for him: insert random "point mutations" [in] future releases of Linux and see what happens.

It could be done at the source level, but it would be more interesting at the binary level (and ignore the possibility of recompiling for now). In each copy of the Linux executable released for the major distributions, randomly change, say, every 10,000th bit (so that the random changes are different for each copy).

According to the theory of evolution, some changes should improve the operating system, and hence it should increase the user demand for it. The users of that version should then proliferate copies of it for their friends, and eventually that version should become dominant.

 

[blader]

It would probably work too, if:

#1. All Linux users would be willing to reinstall Linux just to see evolution work.

#2. There's a great battle for survival where superior Linux distributions kill off the inferior ones. Which probably means that users would have to choose between tens of thousands of distributions, each with their own features, just to play npetreley's little game.

#3. The superior Linux distributions could mate and create offspring with mutations.

#4. The offspring repeats step #2.

Yes, it's funny how ridiculous assumptions gives ridiculous conclusions.

 

[LiveFreeOrDie]

Repeat after me: "Living things are not computer programs."

 

[Morat]

  Amusingly enough, I've used genetic algorithms myself. They work quite well.

 

[Jerry Smith]

I hear there is a genetic algorithm that can improvise a jazz solo now... Anyone have a link?

 

[Morat]

  I've not heard of one, but I have seen them used for everything from designing jet engines to supervising warehouse space for a whiskey manufacturer.

 

 

[npetreley]

You keep talking about computer programs that use so-called genetic algorithms to design OTHER things for which you have standards. What I'd love to see you do is literally introduce bit-flip mutations into your binaries and see how well they run.

 

[Jerry Smith]

Why? Are bit-flip mutations in any way analogous to any sort of mutation that occurs in nature?

 

[Morat]

  Nope. To mimic evolution the way Nick wants, you'd merely have to write programs were the fitness function is "they run" and "they reproduce".

   You get viruses that each your hard drive alive.

   Nick's objection to GA's is akin to claiming you can't discuss ice freezing in the Arctic using tests run on water in your freezer. It's an objection to any sort of experiment or model.

 

 

[npetreley]

Originally posted by Jerry Smith
Why? Are bit-flip mutations in any way analogous to any sort of mutation that occurs in nature?

Is using a genetic algorithm to design something specific analogous to evolution?

 

[npetreley]

Originally posted by Morat
  Nope. To mimic evolution the way Nick wants, you'd merely have to write programs were the fitness function is "they run" and "they reproduce".

   You get viruses that each your hard drive alive.

   Nick's objection to GA's is akin to claiming you can't discuss ice freezing in the Arctic using tests run on water in your freezer. It's an objection to any sort of experiment or model.

 

You don't have to write any new programs to mimic evolution the way it works in the evolutionist's mind. There are plenty of programs that let you specify a beginning picture and a target picture, after which it will calculate and draw as many transitionals as you like. In fact, there's a booth at the local Dave and Buster's that takes two pictures (of two different people) and combines them into one person. Very cool. Very evolutionist.

 

[Jerry Smith]

Roughly, yes. And roughly, bit-flip changes in code are too. Both analogies do break down relatively quickly. Neither can demonstrate very much. That's the cost of over-extending an analogy.

 

[LiveFreeOrDie]

To see what a real (and successful BTW) simulation of biological evolution looks like, see here:

http://www-lecb.ncifcrf.gov/~toms/paper/ev/

 

[blader]

Originally posted by npetreley


Is using a genetic algorithm to design something specific analogous to evolution?

It certainly is.

Evolution isn't random bitflips to get life. Much like genetic algorithms, biological evolution designs for something particular as well: survival in the natural world.

Genetic algorithms use the exact same methods, only it designs for, say, how to pick good stocks. Or any of the other myriad fields that evolution has been implemented in.

I'd like to remind you that if you're using Windows, the Virtual Memory Manager was designed using genetic algorithms. Ironic, isn't it?

 

[npetreley]

Originally posted by blader
It certainly is.

Genetic algorithms use the exact same methods, only it designs for, say, how to pick good stocks. Or any of the other myriad fields that evolution has been implemented in.

I see. So evolution is just a word for intelligent design. I never heard it described that way before.

 

[Morat]

  I'd imagine because it takes a special sort of thinker to draw that conclusion.

  Let no one say you aren't special, Nick. Very special.

  I can only imagine you've never used or written a genetic algorithm.

  Because the end result is completely undesigned, fitting an objective criteria (reproduction in life, performing a certain task in GAs) in a way no human would have designed it, and often far more effective.

   Using GA's, some engineers have created circuits to do certain tasks. The circuits not only work, but have fewer gates than the best designed human solution, and often work in ways no one can figure out.

 

[choccy]

Using GA's, some engineers have created circuits to do certain tasks. The circuits not only work, but have fewer gates than the best designed human solution, and often work in ways no one can figure out.

I'm guessing you're talking about something like what I've quoted below from this articel in New Scientist.

Thompson realised that he could use a standard genetic algorithm to evolve a configuration program for an FPGA and then test each new circuit design immediately on the chip. He set the system a task that appeared impossible for a human designer. Using only 100 logic cells, evolution had to come up with a circuit that could discriminate between two tones, one at 1 kilohertz and the other at 10 kilohertz.

To kick off the experiment, Thompson created a population of 50 configuration programs on a computer, each consisting of a random string of 1s and 0s. The computer downloaded each program in turn to the FPGA to create its circuit and then played it the test tones (see Diagram). The genetic algorithm tested the fitness of each circuit by checking how well it discriminated between the tones. It looked for some characteristic that might prove useful in evolving a solution. At first, this was just an indication that the circuit's output was not completely random. In the first generation, the fittest individual was one with a steady 5-volt output no matter which audio tone it heard.

After testing the initial population, the genetic algorithm killed off the least fit individuals by deleting them and let the most fit produce copies of themselves-offspring. It mated some individuals, swapping sections of their code. Finally, the algorithm introduced a small number of mutations by randomly switching 1s and 0s within individual programs. It then downloaded the new population one at a time onto the FPGA and ran the fitness tests once more. By generation 220, the fittest individual produced outputs almost identical to the inputs-two waveforms corresponding to 1 kilohertz and 10 kilohertz-but not yet the required steady output at 0 volts or 5 volts (see Diagram). By generation 650, the output stayed mostly high for the 1 kilohertz input, although the 10 kilohertz input still produced a waveform. By generation 1400, the output was mostly high for the first signal and mostly low for the second. By generation 2800, the fittest circuit was discriminating accurately between the two inputs, but there were still glitches in its output. These only disappeared completely at generation 4100. After this, there were no further changes.

Now, how is this not analogous to random mutations and natural selection?

Choccy

 

[blader]

Originally posted by npetreley


I see. So evolution is just a word for intelligent design. I never heard it described that way before.

If by "intelligent design", you mean the improvement of a species through random mutations and and natural selection to fit certain survival objectives, (as is the case with both biological evolution and genetic algorithms that we have discussed above) then yes, yes it is.

I'm glad we're getting somewhere.

 

[npetreley]

Originally posted by blader


If by "intelligent design", you mean the improvement of a species through random mutations and and natural selection to fit certain survival objectives, (as is the case with both biological evolution and genetic algorithms that we have discussed above) then yes, yes it is.

I'm glad we're getting somewhere.

No, I guess we're not getting anywhere. I'm talking about Morat's comments. His idea of applying evolution to programming is, "I want to design a jet engine, and I'm going to use genetic algorithms to do so."

But that's not evolution at all. That's intelligent design, using genetic algorithms as a tool to achieve a specific goal. Evolution doesn't have goals.

Abiogenesis would be...

"I'm going to let this genetic algorithm run on a set of unrelated numbers and see if it produces a program (and when you get one to run at all, let alone do anything useful, give me a ringy-dingy)."

And evolution would be...

"I'll introduce random mutations in this binary, and if it improves the way the program behaves, I'll replace the old program with the mutated one."

 

[blader]

Originally posted by npetreley


But that's not evolution at all. That's intelligent design, using genetic algorithms as a tool to achieve a specific goal. Evolution doesn't have goals.

Wrong. Evolution has a goal. The goal is to be the fittest for survival in your environment. To apply evolution in other fields, all one has to do is change this goal to something else, such as, say, "to be the most efficient jet engine." I don't know why you're even arguing against this, because this has been tried, done, and has worked very well, no matter how much you may want to deny it.

Abiogenesis would be...

"I'm going to let this genetic algorithm run on a set of unrelated numbers and see if it produces a program (and when you get one to run at all, let alone do anything useful, give me a ringy-dingy)."

And evolution would be...

"I'll introduce random mutations in this binary, and if it improves the way the program behaves, I'll replace the old program with the mutated one."

If this is what you think evolution is, no wonder you don't believe in it. I'm not surprised. You don't find many Creationists who dare study evolutionary computation.

http://alife.ccp14.ac.uk/ftp-mirror/alife/zooland/pub/research/ci/Alife/karl-sims/creatures-demo.mpg

^This is an evolved virtual creature using fitness constraints of being the fastest at running, jumping, and swimming. First seen at SIGGRAPH '94.

Here is the paper relating to the above:
http://www.genarts.com/karl/papers/siggraph94.pdf

You seem to have a pretty picture of evolutionary algorithms working as so:

Start with random bits.
Mutate and select until the bits matches the latest binaries of Windows XP.

I suggest you read the above paper for an excellent clarification of what is evolutionary algorithms. The virtual creatures that seem so lifelike in the video were evolved with only one survival constraint among these:

Walk the fastest.
Swim the fastest.
Jump the highest.

You can see the results for yourself.